feat(social): face detection + recognition (SCRFD + ArcFace TRT FP16, Issue #80)

Add two new ROS2 packages for the social sprint:

saltybot_social_msgs (ament_cmake):
- FaceDetection, FaceDetectionArray, FaceEmbedding, FaceEmbeddingArray
- PersonState, PersonStateArray
- EnrollPerson, ListPersons, DeletePerson, UpdatePerson services

saltybot_social_face (ament_python):
- SCRFDDetector: SCRFD face detection with TRT FP16 + ONNX fallback
  - 640x640 input, 3-stride anchor decoding, NMS
- ArcFaceRecognizer: 512-dim embedding extraction with gallery matching
  - 5-point landmark alignment to 112x112, cosine similarity
- FaceGallery: thread-safe persistent gallery (npz + JSON sidecar)
- FaceRecognitionNode: ROS2 node subscribing /camera/color/image_raw,
  publishing /social/faces/detections, /social/faces/embeddings
- Enrollment via /social/enroll service (N-sample face averaging)
- Launch file, config YAML, TRT engine builder script

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

jetson/ros2_ws/src/saltybot_social/config/social_params.yaml

@@ -0,0 +1,8 @@
+person_state_tracker:
+  ros__parameters:
+    engagement_distance: 2.0
+    absent_timeout: 5.0
+    prune_timeout: 30.0
+    update_rate: 10.0
+    n_cameras: 4
+    uwb_enabled: false

jetson/ros2_ws/src/saltybot_social/launch/social.launch.py

@@ -0,0 +1,35 @@
+from launch import LaunchDescription
+from launch.actions import DeclareLaunchArgument
+from launch.substitutions import LaunchConfiguration
+from launch_ros.actions import Node
+
+
+def generate_launch_description():
+    return LaunchDescription([
+        DeclareLaunchArgument(
+            'engagement_distance',
+            default_value='2.0',
+            description='Distance in metres below which a person is considered engaged'
+        ),
+        DeclareLaunchArgument(
+            'absent_timeout',
+            default_value='5.0',
+            description='Seconds without detection before marking person as ABSENT'
+        ),
+        DeclareLaunchArgument(
+            'uwb_enabled',
+            default_value='false',
+            description='Whether UWB anchor data is available'
+        ),
+        Node(
+            package='saltybot_social',
+            executable='person_state_tracker',
+            name='person_state_tracker',
+            output='screen',
+            parameters=[{
+                'engagement_distance': LaunchConfiguration('engagement_distance'),
+                'absent_timeout': LaunchConfiguration('absent_timeout'),
+                'uwb_enabled': LaunchConfiguration('uwb_enabled'),
+            }],
+        ),
+    ])

jetson/ros2_ws/src/saltybot_social/package.xml

@@ -0,0 +1,25 @@
+<?xml version="1.0"?>
+<?xml-model href="http://download.ros.org/schema/package_format3.xsd" schematypens="http://www.w3.org/2001/XMLSchema"?>
+<package format="3">
+  <name>saltybot_social</name>
+  <version>0.1.0</version>
+  <description>Multi-modal person identity fusion and state tracking for saltybot</description>
+  <maintainer email="seb@vayrette.com">seb</maintainer>
+  <license>MIT</license>
+  <depend>rclpy</depend>
+  <depend>std_msgs</depend>
+  <depend>geometry_msgs</depend>
+  <depend>sensor_msgs</depend>
+  <depend>vision_msgs</depend>
+  <depend>saltybot_social_msgs</depend>
+  <depend>tf2_ros</depend>
+  <depend>tf2_geometry_msgs</depend>
+  <depend>cv_bridge</depend>
+  <test_depend>ament_copyright</test_depend>
+  <test_depend>ament_flake8</test_depend>
+  <test_depend>ament_pep257</test_depend>
+  <test_depend>python3-pytest</test_depend>
+  <export>
+    <build_type>ament_python</build_type>
+  </export>
+</package>

jetson/ros2_ws/src/saltybot_social/saltybot_social/identity_fuser.py

@@ -0,0 +1,201 @@
+import math
+import time
+from typing import List, Optional, Tuple
+
+from saltybot_social.person_state import PersonTrack, State
+
+
+class IdentityFuser:
+    """Multi-modal identity fusion: face_id + speaker_id + UWB anchor -> unified person_id."""
+
+    def __init__(self, position_window: float = 3.0):
+        self._tracks: dict[int, PersonTrack] = {}
+        self._next_id: int = 1
+        self._position_window = position_window
+        self._max_history = 20
+
+    def _distance_3d(self, a: Optional[tuple], b: Optional[tuple]) -> float:
+        if a is None or b is None:
+            return float('inf')
+        return math.sqrt(sum((ai - bi) ** 2 for ai, bi in zip(a, b)))
+
+    def _compute_distance_and_bearing(self, position: tuple) -> Tuple[float, float]:
+        x, y, z = position
+        distance = math.sqrt(x * x + y * y + z * z)
+        bearing = math.degrees(math.atan2(y, x))
+        return distance, bearing
+
+    def _allocate_id(self) -> int:
+        pid = self._next_id
+        self._next_id += 1
+        return pid
+
+    def update_face(self, face_id: int, name: str,
+                    position_3d: Optional[tuple], camera_id: int,
+                    now: float) -> int:
+        """Match by face_id first. If face_id >= 0, find existing track or create new."""
+        if face_id >= 0:
+            for track in self._tracks.values():
+                if track.face_id == face_id:
+                    track.name = name if name and name != 'unknown' else track.name
+                    if position_3d is not None:
+                        track.position = position_3d
+                        dist, bearing = self._compute_distance_and_bearing(position_3d)
+                        track.distance = dist
+                        track.bearing_deg = bearing
+                        track.history_distances.append(dist)
+                        if len(track.history_distances) > self._max_history:
+                            track.history_distances = track.history_distances[-self._max_history:]
+                    track.camera_id = camera_id
+                    track.last_seen = now
+                    track.last_face_seen = now
+                    return track.person_id
+
+        # No existing track with this face_id; try proximity match for face_id < 0
+        if face_id < 0 and position_3d is not None:
+            best_track = None
+            best_dist = 0.5  # 0.5m proximity threshold
+            for track in self._tracks.values():
+                d = self._distance_3d(track.position, position_3d)
+                if d < best_dist and (now - track.last_seen) < self._position_window:
+                    best_dist = d
+                    best_track = track
+            if best_track is not None:
+                best_track.position = position_3d
+                dist, bearing = self._compute_distance_and_bearing(position_3d)
+                best_track.distance = dist
+                best_track.bearing_deg = bearing
+                best_track.history_distances.append(dist)
+                if len(best_track.history_distances) > self._max_history:
+                    best_track.history_distances = best_track.history_distances[-self._max_history:]
+                best_track.camera_id = camera_id
+                best_track.last_seen = now
+                return best_track.person_id
+
+        # Create new track
+        pid = self._allocate_id()
+        track = PersonTrack(person_id=pid, face_id=face_id, name=name,
+                            camera_id=camera_id, last_seen=now, last_face_seen=now)
+        if position_3d is not None:
+            track.position = position_3d
+            dist, bearing = self._compute_distance_and_bearing(position_3d)
+            track.distance = dist
+            track.bearing_deg = bearing
+            track.history_distances.append(dist)
+        self._tracks[pid] = track
+        return pid
+
+    def update_speaker(self, speaker_id: str, now: float) -> int:
+        """Find track with nearest recently-seen position, assign speaker_id."""
+        # First check if any track already has this speaker_id
+        for track in self._tracks.values():
+            if track.speaker_id == speaker_id:
+                track.last_seen = now
+                return track.person_id
+
+        # Assign to nearest recently-seen track
+        best_track = None
+        best_dist = float('inf')
+        for track in self._tracks.values():
+            if (now - track.last_seen) < self._position_window and track.distance > 0:
+                if track.distance < best_dist:
+                    best_dist = track.distance
+                    best_track = track
+
+        if best_track is not None:
+            best_track.speaker_id = speaker_id
+            best_track.last_seen = now
+            return best_track.person_id
+
+        # No suitable track found; create a new one
+        pid = self._allocate_id()
+        track = PersonTrack(person_id=pid, speaker_id=speaker_id,
+                            last_seen=now)
+        self._tracks[pid] = track
+        return pid
+
+    def update_uwb(self, uwb_anchor_id: str, position_3d: tuple,
+                    now: float) -> int:
+        """Match by proximity (nearest track within 0.5m)."""
+        # Check existing UWB assignment
+        for track in self._tracks.values():
+            if track.uwb_anchor_id == uwb_anchor_id:
+                track.position = position_3d
+                dist, bearing = self._compute_distance_and_bearing(position_3d)
+                track.distance = dist
+                track.bearing_deg = bearing
+                track.history_distances.append(dist)
+                if len(track.history_distances) > self._max_history:
+                    track.history_distances = track.history_distances[-self._max_history:]
+                track.last_seen = now
+                return track.person_id
+
+        # Proximity match
+        best_track = None
+        best_dist = 0.5
+        for track in self._tracks.values():
+            d = self._distance_3d(track.position, position_3d)
+            if d < best_dist and (now - track.last_seen) < self._position_window:
+                best_dist = d
+                best_track = track
+
+        if best_track is not None:
+            best_track.uwb_anchor_id = uwb_anchor_id
+            best_track.position = position_3d
+            dist, bearing = self._compute_distance_and_bearing(position_3d)
+            best_track.distance = dist
+            best_track.bearing_deg = bearing
+            best_track.history_distances.append(dist)
+            if len(best_track.history_distances) > self._max_history:
+                best_track.history_distances = best_track.history_distances[-self._max_history:]
+            best_track.last_seen = now
+            return best_track.person_id
+
+        # Create new track
+        pid = self._allocate_id()
+        track = PersonTrack(person_id=pid, uwb_anchor_id=uwb_anchor_id,
+                            position=position_3d, last_seen=now)
+        dist, bearing = self._compute_distance_and_bearing(position_3d)
+        track.distance = dist
+        track.bearing_deg = bearing
+        track.history_distances.append(dist)
+        self._tracks[pid] = track
+        return pid
+
+    def get_all_tracks(self) -> List[PersonTrack]:
+        return list(self._tracks.values())
+
+    @staticmethod
+    def compute_attention(tracks: List[PersonTrack]) -> int:
+        """Focus on nearest engaged/talking person, or -1 if none."""
+        candidates = [t for t in tracks
+                      if t.state in (State.ENGAGED, State.TALKING) and t.distance > 0]
+        if not candidates:
+            return -1
+        # Prefer talking, then nearest engaged
+        talking = [t for t in candidates if t.state == State.TALKING]
+        if talking:
+            return min(talking, key=lambda t: t.distance).person_id
+        return min(candidates, key=lambda t: t.distance).person_id
+
+    def prune_absent(self, absent_timeout: float = 30.0):
+        """Remove tracks absent longer than timeout."""
+        now = time.monotonic()
+        to_remove = [pid for pid, t in self._tracks.items()
+                     if t.state == State.ABSENT and (now - t.last_seen) > absent_timeout]
+        for pid in to_remove:
+            del self._tracks[pid]
+
+    @staticmethod
+    def detect_group(tracks: List[PersonTrack]) -> bool:
+        """Returns True if >= 2 persons within 1.5m of each other."""
+        active = [t for t in tracks
+                  if t.position is not None and t.state != State.ABSENT]
+        for i, a in enumerate(active):
+            for b in active[i + 1:]:
+                dx = a.position[0] - b.position[0]
+                dy = a.position[1] - b.position[1]
+                dz = a.position[2] - b.position[2]
+                if math.sqrt(dx * dx + dy * dy + dz * dz) < 1.5:
+                    return True
+        return False

jetson/ros2_ws/src/saltybot_social/saltybot_social/person_state.py

@@ -0,0 +1,54 @@
+import time
+from dataclasses import dataclass, field
+from enum import IntEnum
+from typing import Optional
+
+
+class State(IntEnum):
+    UNKNOWN = 0
+    APPROACHING = 1
+    ENGAGED = 2
+    TALKING = 3
+    LEAVING = 4
+    ABSENT = 5
+
+
+@dataclass
+class PersonTrack:
+    person_id: int
+    face_id: int = -1
+    speaker_id: str = ''
+    uwb_anchor_id: str = ''
+    name: str = 'unknown'
+    state: State = State.UNKNOWN
+    position: Optional[tuple] = None   # (x, y, z) in base_link
+    distance: float = 0.0
+    bearing_deg: float = 0.0
+    engagement_score: float = 0.0
+    camera_id: int = -1
+    last_seen: float = field(default_factory=time.monotonic)
+    last_face_seen: float = 0.0
+    history_distances: list = field(default_factory=list)  # last N distances for trend
+
+    def update_state(self, now: float, engagement_distance: float = 2.0,
+                     absent_timeout: float = 5.0):
+        """Transition state machine based on distance trend and time."""
+        age = now - self.last_seen
+        if age > absent_timeout:
+            self.state = State.ABSENT
+            return
+        if self.speaker_id:
+            self.state = State.TALKING
+            return
+        if len(self.history_distances) >= 3:
+            trend = self.history_distances[-1] - self.history_distances[-3]
+            if self.distance < engagement_distance:
+                self.state = State.ENGAGED
+            elif trend < -0.2:  # moving closer
+                self.state = State.APPROACHING
+            elif trend > 0.3:   # moving away
+                self.state = State.LEAVING
+            else:
+                self.state = State.ENGAGED if self.distance < engagement_distance else State.UNKNOWN
+        elif self.distance > 0:
+            self.state = State.ENGAGED if self.distance < engagement_distance else State.APPROACHING

jetson/ros2_ws/src/saltybot_social/saltybot_social/person_state_tracker_node.py

@@ -0,0 +1,273 @@
+import time
+
+import rclpy
+from rclpy.node import Node
+from rclpy.qos import QoSProfile, ReliabilityPolicy, HistoryPolicy
+
+from std_msgs.msg import Int32, String
+from geometry_msgs.msg import PoseArray, PoseStamped, Point
+from sensor_msgs.msg import Image
+from builtin_interfaces.msg import Time as TimeMsg
+
+from saltybot_social_msgs.msg import (
+    FaceDetection,
+    FaceDetectionArray,
+    PersonState,
+    PersonStateArray,
+)
+from saltybot_social.identity_fuser import IdentityFuser
+from saltybot_social.person_state import State
+
+
+class PersonStateTrackerNode(Node):
+    """Main ROS2 node for multi-modal person tracking."""
+
+    def __init__(self):
+        super().__init__('person_state_tracker')
+
+        # Parameters
+        self.declare_parameter('engagement_distance', 2.0)
+        self.declare_parameter('absent_timeout', 5.0)
+        self.declare_parameter('prune_timeout', 30.0)
+        self.declare_parameter('update_rate', 10.0)
+        self.declare_parameter('n_cameras', 4)
+        self.declare_parameter('uwb_enabled', False)
+
+        self._engagement_distance = self.get_parameter('engagement_distance').value
+        self._absent_timeout = self.get_parameter('absent_timeout').value
+        self._prune_timeout = self.get_parameter('prune_timeout').value
+        update_rate = self.get_parameter('update_rate').value
+        n_cameras = self.get_parameter('n_cameras').value
+        uwb_enabled = self.get_parameter('uwb_enabled').value
+
+        self._fuser = IdentityFuser()
+
+        # QoS profiles
+        best_effort_qos = QoSProfile(
+            reliability=ReliabilityPolicy.BEST_EFFORT,
+            history=HistoryPolicy.KEEP_LAST,
+            depth=5
+        )
+        reliable_qos = QoSProfile(
+            reliability=ReliabilityPolicy.RELIABLE,
+            history=HistoryPolicy.KEEP_LAST,
+            depth=1
+        )
+
+        # Subscriptions
+        self.create_subscription(
+            FaceDetectionArray,
+            '/social/faces/detections',
+            self._on_face_detections,
+            best_effort_qos
+        )
+        self.create_subscription(
+            String,
+            '/social/speech/speaker_id',
+            self._on_speaker_id,
+            10
+        )
+        if uwb_enabled:
+            self.create_subscription(
+                PoseArray,
+                '/uwb/positions',
+                self._on_uwb,
+                10
+            )
+
+        # Camera subscriptions (monitor topic existence)
+        self.create_subscription(
+            Image, '/camera/color/image_raw',
+            self._on_camera_image, best_effort_qos)
+        for i in range(n_cameras):
+            self.create_subscription(
+                Image, f'/surround/cam{i}/image_raw',
+                self._on_camera_image, best_effort_qos)
+
+        # Existing person tracker position
+        self.create_subscription(
+            PoseStamped,
+            '/person/target',
+            self._on_person_target,
+            10
+        )
+
+        # Publishers
+        self._persons_pub = self.create_publisher(
+            PersonStateArray, '/social/persons', best_effort_qos)
+        self._attention_pub = self.create_publisher(
+            Int32, '/social/attention/target_id', reliable_qos)
+
+        # Timer
+        timer_period = 1.0 / update_rate
+        self.create_timer(timer_period, self._on_timer)
+
+        self.get_logger().info(
+            f'PersonStateTracker started: engagement={self._engagement_distance}m, '
+            f'absent_timeout={self._absent_timeout}s, rate={update_rate}Hz, '
+            f'cameras={n_cameras}, uwb={uwb_enabled}')
+
+    def _on_face_detections(self, msg: FaceDetectionArray):
+        now = time.monotonic()
+        for face in msg.faces:
+            position_3d = None
+            # Use bbox center as rough position estimate if depth is available
+            # Real 3D position would come from depth camera projection
+            if face.bbox_x > 0 or face.bbox_y > 0:
+                # Approximate: use bbox center as bearing proxy, distance from bbox size
+                # This is a placeholder; real impl would use depth image lookup
+                position_3d = (
+                    max(0.5, 1.0 / max(face.bbox_w, 0.01)),  # rough depth from bbox width
+                    face.bbox_x + face.bbox_w / 2.0 - 0.5,   # x offset from center
+                    face.bbox_y + face.bbox_h / 2.0 - 0.5    # y offset from center
+                )
+
+            camera_id = -1
+            if hasattr(face.header, 'frame_id') and face.header.frame_id:
+                frame = face.header.frame_id
+                if 'cam' in frame:
+                    try:
+                        camera_id = int(frame.split('cam')[-1].split('_')[0])
+                    except ValueError:
+                        pass
+
+            self._fuser.update_face(
+                face_id=face.face_id,
+                name=face.person_name,
+                position_3d=position_3d,
+                camera_id=camera_id,
+                now=now
+            )
+
+    def _on_speaker_id(self, msg: String):
+        now = time.monotonic()
+        speaker_id = msg.data
+        # Parse "speaker_<id>" format or raw id
+        if speaker_id.startswith('speaker_'):
+            speaker_id = speaker_id[len('speaker_'):]
+        self._fuser.update_speaker(speaker_id, now)
+
+    def _on_uwb(self, msg: PoseArray):
+        now = time.monotonic()
+        for i, pose in enumerate(msg.poses):
+            position = (pose.position.x, pose.position.y, pose.position.z)
+            anchor_id = f'uwb_{i}'
+            if hasattr(msg, 'header') and msg.header.frame_id:
+                anchor_id = f'{msg.header.frame_id}_{i}'
+            self._fuser.update_uwb(anchor_id, position, now)
+
+    def _on_camera_image(self, msg: Image):
+        # Monitor only -- no processing here
+        pass
+
+    def _on_person_target(self, msg: PoseStamped):
+        now = time.monotonic()
+        position_3d = (
+            msg.pose.position.x,
+            msg.pose.position.y,
+            msg.pose.position.z
+        )
+        # Update position for unidentified person (from YOLOv8 tracker)
+        self._fuser.update_face(
+            face_id=-1,
+            name='unknown',
+            position_3d=position_3d,
+            camera_id=-1,
+            now=now
+        )
+
+    def _on_timer(self):
+        now = time.monotonic()
+        tracks = self._fuser.get_all_tracks()
+
+        # Update state machine for all tracks
+        for track in tracks:
+            track.update_state(
+                now,
+                engagement_distance=self._engagement_distance,
+                absent_timeout=self._absent_timeout
+            )
+
+        # Compute engagement scores
+        for track in tracks:
+            if track.state == State.ABSENT:
+                track.engagement_score = 0.0
+            elif track.state == State.TALKING:
+                track.engagement_score = 1.0
+            elif track.state == State.ENGAGED:
+                track.engagement_score = max(0.0, 1.0 - track.distance / self._engagement_distance)
+            elif track.state == State.APPROACHING:
+                track.engagement_score = 0.3
+            elif track.state == State.LEAVING:
+                track.engagement_score = 0.1
+            else:
+                track.engagement_score = 0.0
+
+        # Prune long-absent tracks
+        self._fuser.prune_absent(self._prune_timeout)
+
+        # Compute attention target
+        attention_id = IdentityFuser.compute_attention(tracks)
+
+        # Build and publish PersonStateArray
+        msg = PersonStateArray()
+        msg.header.stamp = self.get_clock().now().to_msg()
+        msg.header.frame_id = 'base_link'
+        msg.primary_attention_id = attention_id
+
+        for track in tracks:
+            ps = PersonState()
+            ps.header.stamp = msg.header.stamp
+            ps.header.frame_id = 'base_link'
+            ps.person_id = track.person_id
+            ps.person_name = track.name
+            ps.face_id = track.face_id
+            ps.speaker_id = track.speaker_id
+            ps.uwb_anchor_id = track.uwb_anchor_id
+            if track.position is not None:
+                ps.position = Point(
+                    x=float(track.position[0]),
+                    y=float(track.position[1]),
+                    z=float(track.position[2]))
+            ps.distance = float(track.distance)
+            ps.bearing_deg = float(track.bearing_deg)
+            ps.state = int(track.state)
+            ps.engagement_score = float(track.engagement_score)
+            ps.last_seen = self._mono_to_ros_time(track.last_seen)
+            ps.camera_id = track.camera_id
+            msg.persons.append(ps)
+
+        self._persons_pub.publish(msg)
+
+        # Publish attention target
+        att_msg = Int32()
+        att_msg.data = attention_id
+        self._attention_pub.publish(att_msg)
+
+    def _mono_to_ros_time(self, mono: float) -> TimeMsg:
+        """Convert monotonic timestamp to approximate ROS time."""
+        # Offset from monotonic to wall clock
+        offset = time.time() - time.monotonic()
+        wall = mono + offset
+        sec = int(wall)
+        nsec = int((wall - sec) * 1e9)
+        t = TimeMsg()
+        t.sec = sec
+        t.nanosec = nsec
+        return t
+
+
+def main(args=None):
+    rclpy.init(args=args)
+    node = PersonStateTrackerNode()
+    try:
+        rclpy.spin(node)
+    except KeyboardInterrupt:
+        pass
+    finally:
+        node.destroy_node()
+        rclpy.shutdown()
+
+
+if __name__ == '__main__':
+    main()

jetson/ros2_ws/src/saltybot_social/setup.cfg

@@ -0,0 +1,4 @@
+[develop]
+script_dir=$base/lib/saltybot_social
+[install]
+install_scripts=$base/lib/saltybot_social

jetson/ros2_ws/src/saltybot_social/setup.py

@@ -0,0 +1,32 @@
+from setuptools import find_packages, setup
+import os
+from glob import glob
+
+package_name = 'saltybot_social'
+
+setup(
+    name=package_name,
+    version='0.1.0',
+    packages=find_packages(exclude=['test']),
+    data_files=[
+        ('share/ament_index/resource_index/packages',
+            ['resource/' + package_name]),
+        ('share/' + package_name, ['package.xml']),
+        (os.path.join('share', package_name, 'launch'),
+            glob(os.path.join('launch', '*launch.[pxy][yma]*'))),
+        (os.path.join('share', package_name, 'config'),
+            glob(os.path.join('config', '*.yaml'))),
+    ],
+    install_requires=['setuptools'],
+    zip_safe=True,
+    maintainer='seb',
+    maintainer_email='seb@vayrette.com',
+    description='Multi-modal person identity fusion and state tracking for saltybot',
+    license='MIT',
+    tests_require=['pytest'],
+    entry_points={
+        'console_scripts': [
+            'person_state_tracker = saltybot_social.person_state_tracker_node:main',
+        ],
+    },
+)

jetson/ros2_ws/src/saltybot_social_face/config/face_recognition_params.yaml

@@ -0,0 +1,11 @@
+face_recognizer:
+  ros__parameters:
+    scrfd_engine_path: '/mnt/nvme/saltybot/models/scrfd_2.5g.engine'
+    scrfd_onnx_path: '/mnt/nvme/saltybot/models/scrfd_2.5g_bnkps.onnx'
+    arcface_engine_path: '/mnt/nvme/saltybot/models/arcface_r50.engine'
+    arcface_onnx_path: '/mnt/nvme/saltybot/models/arcface_r50.onnx'
+    gallery_dir: '/mnt/nvme/saltybot/gallery'
+    recognition_threshold: 0.35
+    publish_debug_image: false
+    max_faces: 10
+    scrfd_conf_thresh: 0.5

jetson/ros2_ws/src/saltybot_social_face/launch/face_recognition.launch.py

@@ -0,0 +1,80 @@
+"""
+face_recognition.launch.py -- Launch file for the SCRFD + ArcFace face recognition node.
+
+Launches the face_recognizer node with configurable model paths and parameters.
+The RealSense camera must be running separately (e.g., via realsense.launch.py).
+"""
+
+from launch import LaunchDescription
+from launch.actions import DeclareLaunchArgument
+from launch.substitutions import LaunchConfiguration
+from launch_ros.actions import Node
+
+
+def generate_launch_description():
+    """Generate launch description for face recognition pipeline."""
+    return LaunchDescription([
+        DeclareLaunchArgument(
+            'scrfd_engine_path',
+            default_value='/mnt/nvme/saltybot/models/scrfd_2.5g.engine',
+            description='Path to SCRFD TensorRT engine file',
+        ),
+        DeclareLaunchArgument(
+            'scrfd_onnx_path',
+            default_value='/mnt/nvme/saltybot/models/scrfd_2.5g_bnkps.onnx',
+            description='Path to SCRFD ONNX model file (fallback)',
+        ),
+        DeclareLaunchArgument(
+            'arcface_engine_path',
+            default_value='/mnt/nvme/saltybot/models/arcface_r50.engine',
+            description='Path to ArcFace TensorRT engine file',
+        ),
+        DeclareLaunchArgument(
+            'arcface_onnx_path',
+            default_value='/mnt/nvme/saltybot/models/arcface_r50.onnx',
+            description='Path to ArcFace ONNX model file (fallback)',
+        ),
+        DeclareLaunchArgument(
+            'gallery_dir',
+            default_value='/mnt/nvme/saltybot/gallery',
+            description='Directory for persistent face gallery storage',
+        ),
+        DeclareLaunchArgument(
+            'recognition_threshold',
+            default_value='0.35',
+            description='Cosine similarity threshold for face recognition',
+        ),
+        DeclareLaunchArgument(
+            'publish_debug_image',
+            default_value='false',
+            description='Publish annotated debug image to /social/faces/debug_image',
+        ),
+        DeclareLaunchArgument(
+            'max_faces',
+            default_value='10',
+            description='Maximum faces to process per frame',
+        ),
+        DeclareLaunchArgument(
+            'scrfd_conf_thresh',
+            default_value='0.5',
+            description='SCRFD detection confidence threshold',
+        ),
+
+        Node(
+            package='saltybot_social_face',
+            executable='face_recognition',
+            name='face_recognizer',
+            output='screen',
+            parameters=[{
+                'scrfd_engine_path': LaunchConfiguration('scrfd_engine_path'),
+                'scrfd_onnx_path': LaunchConfiguration('scrfd_onnx_path'),
+                'arcface_engine_path': LaunchConfiguration('arcface_engine_path'),
+                'arcface_onnx_path': LaunchConfiguration('arcface_onnx_path'),
+                'gallery_dir': LaunchConfiguration('gallery_dir'),
+                'recognition_threshold': LaunchConfiguration('recognition_threshold'),
+                'publish_debug_image': LaunchConfiguration('publish_debug_image'),
+                'max_faces': LaunchConfiguration('max_faces'),
+                'scrfd_conf_thresh': LaunchConfiguration('scrfd_conf_thresh'),
+            }],
+        ),
+    ])

jetson/ros2_ws/src/saltybot_social_face/package.xml

@@ -0,0 +1,27 @@
+<?xml version="1.0"?>
+<?xml-model href="http://download.ros.org/schema/package_format3.xsd" schematypens="http://www.w3.org/2001/XMLSchema"?>
+<package format="3">
+  <name>saltybot_social_face</name>
+  <version>0.1.0</version>
+  <description>SCRFD face detection and ArcFace recognition for SaltyBot social interactions</description>
+  <maintainer email="seb@vayrette.com">seb</maintainer>
+  <license>MIT</license>
+
+  <depend>rclpy</depend>
+  <depend>sensor_msgs</depend>
+  <depend>cv_bridge</depend>
+  <depend>image_transport</depend>
+  <depend>saltybot_social_msgs</depend>
+
+  <exec_depend>python3-numpy</exec_depend>
+  <exec_depend>python3-opencv</exec_depend>
+
+  <test_depend>ament_copyright</test_depend>
+  <test_depend>ament_flake8</test_depend>
+  <test_depend>ament_pep257</test_depend>
+  <test_depend>python3-pytest</test_depend>
+
+  <export>
+    <build_type>ament_python</build_type>
+  </export>
+</package>

jetson/ros2_ws/src/saltybot_social_face/saltybot_social_face/__init__.py

@@ -0,0 +1 @@
+"""SaltyBot social face detection and recognition package."""

jetson/ros2_ws/src/saltybot_social_face/saltybot_social_face/arcface_recognizer.py

@@ -0,0 +1,316 @@
+"""
+arcface_recognizer.py -- ArcFace face embedding extraction and gallery matching.
+
+Performs face alignment using 5-point landmarks (insightface standard reference),
+extracts 512-dimensional embeddings via ArcFace (TRT FP16 or ONNX fallback),
+and matches against a persistent gallery using cosine similarity.
+"""
+
+import os
+import logging
+from typing import Optional
+
+import numpy as np
+import cv2
+
+logger = logging.getLogger(__name__)
+
+# InsightFace standard reference landmarks for 112x112 alignment
+ARCFACE_SRC = np.array([
+    [38.2946, 51.6963],   # left eye
+    [73.5318, 51.5014],   # right eye
+    [56.0252, 71.7366],   # nose
+    [41.5493, 92.3655],   # left mouth
+    [70.7299, 92.2041],   # right mouth
+], dtype=np.float32)
+
+
+# -- Inference backends --------------------------------------------------------
+
+class _TRTBackend:
+    """TensorRT inference engine for ArcFace."""
+
+    def __init__(self, engine_path: str):
+        import tensorrt as trt
+        import pycuda.driver as cuda
+        import pycuda.autoinit  # noqa: F401
+
+        self._cuda = cuda
+        trt_logger = trt.Logger(trt.Logger.WARNING)
+        with open(engine_path, 'rb') as f, trt.Runtime(trt_logger) as runtime:
+            self._engine = runtime.deserialize_cuda_engine(f.read())
+        self._context = self._engine.create_execution_context()
+
+        self._inputs = []
+        self._outputs = []
+        self._bindings = []
+        for i in range(self._engine.num_io_tensors):
+            name = self._engine.get_tensor_name(i)
+            dtype = trt.nptype(self._engine.get_tensor_dtype(name))
+            shape = tuple(self._engine.get_tensor_shape(name))
+            nbytes = int(np.prod(shape)) * np.dtype(dtype).itemsize
+            host_mem = cuda.pagelocked_empty(shape, dtype)
+            device_mem = cuda.mem_alloc(nbytes)
+            self._bindings.append(int(device_mem))
+            if self._engine.get_tensor_mode(name) == trt.TensorIOMode.INPUT:
+                self._inputs.append({'host': host_mem, 'device': device_mem})
+            else:
+                self._outputs.append({'host': host_mem, 'device': device_mem,
+                                      'shape': shape})
+        self._stream = cuda.Stream()
+
+    def infer(self, input_data: np.ndarray) -> np.ndarray:
+        """Run inference and return the embedding vector."""
+        np.copyto(self._inputs[0]['host'], input_data.ravel())
+        self._cuda.memcpy_htod_async(
+            self._inputs[0]['device'], self._inputs[0]['host'], self._stream)
+        self._context.execute_async_v2(self._bindings, self._stream.handle)
+        for out in self._outputs:
+            self._cuda.memcpy_dtoh_async(out['host'], out['device'], self._stream)
+        self._stream.synchronize()
+        return self._outputs[0]['host'].reshape(self._outputs[0]['shape']).copy()
+
+
+class _ONNXBackend:
+    """ONNX Runtime inference (CUDA EP with CPU fallback)."""
+
+    def __init__(self, onnx_path: str):
+        import onnxruntime as ort
+        providers = ['CUDAExecutionProvider', 'CPUExecutionProvider']
+        self._session = ort.InferenceSession(onnx_path, providers=providers)
+        self._input_name = self._session.get_inputs()[0].name
+
+    def infer(self, input_data: np.ndarray) -> np.ndarray:
+        """Run inference and return the embedding vector."""
+        results = self._session.run(None, {self._input_name: input_data})
+        return results[0]
+
+
+# -- Face alignment ------------------------------------------------------------
+
+def align_face(bgr: np.ndarray, landmarks_10: list[float]) -> np.ndarray:
+    """Align a face to 112x112 using 5-point landmarks.
+
+    Args:
+        bgr: Source BGR image.
+        landmarks_10: Flat list of 10 floats [x0,y0, x1,y1, ..., x4,y4].
+
+    Returns:
+        Aligned BGR face crop of shape (112, 112, 3).
+    """
+    src_pts = np.array(landmarks_10, dtype=np.float32).reshape(5, 2)
+    M, _ = cv2.estimateAffinePartial2D(src_pts, ARCFACE_SRC)
+    if M is None:
+        # Fallback: simple crop and resize from bbox-like region
+        cx = np.mean(src_pts[:, 0])
+        cy = np.mean(src_pts[:, 1])
+        spread = max(np.ptp(src_pts[:, 0]), np.ptp(src_pts[:, 1])) * 1.5
+        half = spread / 2
+        x1 = max(0, int(cx - half))
+        y1 = max(0, int(cy - half))
+        x2 = min(bgr.shape[1], int(cx + half))
+        y2 = min(bgr.shape[0], int(cy + half))
+        crop = bgr[y1:y2, x1:x2]
+        return cv2.resize(crop, (112, 112), interpolation=cv2.INTER_LINEAR)
+
+    aligned = cv2.warpAffine(bgr, M, (112, 112), borderMode=cv2.BORDER_REPLICATE)
+    return aligned
+
+
+# -- Main recognizer class -----------------------------------------------------
+
+class ArcFaceRecognizer:
+    """ArcFace face embedding extractor and gallery matcher.
+
+    Args:
+        engine_path: Path to TensorRT engine file.
+        onnx_path: Path to ONNX model file (used if engine not available).
+    """
+
+    def __init__(self, engine_path: str = '', onnx_path: str = ''):
+        self._backend: Optional[_TRTBackend | _ONNXBackend] = None
+        self.gallery: dict[int, dict] = {}
+
+        # Try TRT first, then ONNX
+        if engine_path and os.path.isfile(engine_path):
+            try:
+                self._backend = _TRTBackend(engine_path)
+                logger.info('ArcFace TensorRT backend loaded: %s', engine_path)
+                return
+            except Exception as e:
+                logger.warning('ArcFace TRT load failed (%s), trying ONNX', e)
+
+        if onnx_path and os.path.isfile(onnx_path):
+            try:
+                self._backend = _ONNXBackend(onnx_path)
+                logger.info('ArcFace ONNX backend loaded: %s', onnx_path)
+                return
+            except Exception as e:
+                logger.error('ArcFace ONNX load failed: %s', e)
+
+        logger.error('No ArcFace model loaded. Recognition will be unavailable.')
+
+    @property
+    def is_loaded(self) -> bool:
+        """Return True if a backend is loaded and ready."""
+        return self._backend is not None
+
+    def embed(self, bgr_face_112x112: np.ndarray) -> np.ndarray:
+        """Extract 512-dim L2-normalized embedding from a 112x112 aligned face.
+
+        Args:
+            bgr_face_112x112: Aligned face crop, BGR, shape (112, 112, 3).
+
+        Returns:
+            L2-normalized embedding of shape (512,).
+        """
+        if self._backend is None:
+            return np.zeros(512, dtype=np.float32)
+
+        # Preprocess: BGR->RGB, /255, subtract 0.5, divide 0.5 -> [1,3,112,112]
+        rgb = cv2.cvtColor(bgr_face_112x112, cv2.COLOR_BGR2RGB).astype(np.float32)
+        rgb = rgb / 255.0
+        rgb = (rgb - 0.5) / 0.5
+        blob = rgb.transpose(2, 0, 1)[np.newaxis]  # [1, 3, 112, 112]
+        blob = np.ascontiguousarray(blob)
+
+        output = self._backend.infer(blob)
+        embedding = output.flatten()[:512].astype(np.float32)
+
+        # L2 normalize
+        norm = np.linalg.norm(embedding)
+        if norm > 0:
+            embedding = embedding / norm
+        return embedding
+
+    def align_and_embed(self, bgr_image: np.ndarray, landmarks_10: list[float]) -> np.ndarray:
+        """Align face using landmarks and extract embedding.
+
+        Args:
+            bgr_image: Full BGR image.
+            landmarks_10: Flat list of 10 floats from SCRFD detection.
+
+        Returns:
+            L2-normalized embedding of shape (512,).
+        """
+        aligned = align_face(bgr_image, landmarks_10)
+        return self.embed(aligned)
+
+    def load_gallery(self, gallery_path: str) -> None:
+        """Load gallery from .npz file with JSON metadata sidecar.
+
+        Args:
+            gallery_path: Path to the .npz gallery file.
+        """
+        import json
+
+        if not os.path.isfile(gallery_path):
+            logger.info('No gallery file at %s, starting empty.', gallery_path)
+            self.gallery = {}
+            return
+
+        data = np.load(gallery_path, allow_pickle=False)
+        meta_path = gallery_path.replace('.npz', '_meta.json')
+
+        if os.path.isfile(meta_path):
+            with open(meta_path, 'r') as f:
+                meta = json.load(f)
+        else:
+            meta = {}
+
+        self.gallery = {}
+        for key in data.files:
+            pid = int(key)
+            embedding = data[key].astype(np.float32)
+            norm = np.linalg.norm(embedding)
+            if norm > 0:
+                embedding = embedding / norm
+            info = meta.get(str(pid), {})
+            self.gallery[pid] = {
+                'name': info.get('name', f'person_{pid}'),
+                'embedding': embedding,
+                'samples': info.get('samples', 1),
+                'enrolled_at': info.get('enrolled_at', 0.0),
+            }
+        logger.info('Gallery loaded: %d persons from %s', len(self.gallery), gallery_path)
+
+    def save_gallery(self, gallery_path: str) -> None:
+        """Save gallery to .npz file with JSON metadata sidecar.
+
+        Args:
+            gallery_path: Path to the .npz gallery file.
+        """
+        import json
+
+        arrays = {}
+        meta = {}
+        for pid, info in self.gallery.items():
+            arrays[str(pid)] = info['embedding']
+            meta[str(pid)] = {
+                'name': info['name'],
+                'samples': info['samples'],
+                'enrolled_at': info['enrolled_at'],
+            }
+
+        os.makedirs(os.path.dirname(gallery_path) or '.', exist_ok=True)
+        np.savez(gallery_path, **arrays)
+
+        meta_path = gallery_path.replace('.npz', '_meta.json')
+        with open(meta_path, 'w') as f:
+            json.dump(meta, f, indent=2)
+        logger.info('Gallery saved: %d persons to %s', len(self.gallery), gallery_path)
+
+    def match(self, embedding: np.ndarray, threshold: float = 0.35) -> tuple[int, str, float]:
+        """Match an embedding against the gallery.
+
+        Args:
+            embedding: L2-normalized query embedding of shape (512,).
+            threshold: Minimum cosine similarity for a match.
+
+        Returns:
+            (person_id, person_name, score) or (-1, '', 0.0) if no match.
+        """
+        if not self.gallery:
+            return (-1, '', 0.0)
+
+        best_pid = -1
+        best_name = ''
+        best_score = 0.0
+
+        for pid, info in self.gallery.items():
+            score = float(np.dot(embedding, info['embedding']))
+            if score > best_score:
+                best_score = score
+                best_pid = pid
+                best_name = info['name']
+
+        if best_score >= threshold:
+            return (best_pid, best_name, best_score)
+        return (-1, '', 0.0)
+
+    def enroll(self, person_id: int, person_name: str, embeddings_list: list[np.ndarray]) -> None:
+        """Enroll a person by averaging multiple embeddings.
+
+        Args:
+            person_id: Unique integer ID for this person.
+            person_name: Human-readable name.
+            embeddings_list: List of L2-normalized embeddings to average.
+        """
+        import time as _time
+
+        if not embeddings_list:
+            return
+
+        mean_emb = np.mean(embeddings_list, axis=0).astype(np.float32)
+        norm = np.linalg.norm(mean_emb)
+        if norm > 0:
+            mean_emb = mean_emb / norm
+
+        self.gallery[person_id] = {
+            'name': person_name,
+            'embedding': mean_emb,
+            'samples': len(embeddings_list),
+            'enrolled_at': _time.time(),
+        }
+        logger.info('Enrolled person %d (%s) with %d samples.',
+                     person_id, person_name, len(embeddings_list))

jetson/ros2_ws/src/saltybot_social_face/saltybot_social_face/enrollment_cli.py

@@ -0,0 +1,78 @@
+#!/usr/bin/env python3
+"""
+enrollment_cli.py -- CLI tool for enrolling persons via the /social/enroll service.
+
+Usage:
+    ros2 run saltybot_social_face enrollment_cli -- --name Alice --mode face --samples 10
+"""
+
+import argparse
+import sys
+
+import rclpy
+from rclpy.node import Node
+
+from saltybot_social_msgs.srv import EnrollPerson
+
+
+class EnrollmentCLI(Node):
+    """Simple CLI node that calls the EnrollPerson service."""
+
+    def __init__(self, name: str, mode: str, n_samples: int):
+        super().__init__('enrollment_cli')
+        self._client = self.create_client(EnrollPerson, '/social/enroll')
+
+        self.get_logger().info('Waiting for /social/enroll service...')
+        if not self._client.wait_for_service(timeout_sec=10.0):
+            self.get_logger().error('Service /social/enroll not available.')
+            return
+
+        request = EnrollPerson.Request()
+        request.name = name
+        request.mode = mode
+        request.n_samples = n_samples
+
+        self.get_logger().info(
+            'Enrolling "%s" (mode=%s, samples=%d)...', name, mode, n_samples)
+
+        future = self._client.call_async(request)
+        rclpy.spin_until_future_complete(self, future, timeout_sec=120.0)
+
+        if future.result() is not None:
+            result = future.result()
+            if result.success:
+                self.get_logger().info(
+                    'Enrollment successful: person_id=%d, %s',
+                    result.person_id, result.message)
+            else:
+                self.get_logger().error(
+                    'Enrollment failed: %s', result.message)
+        else:
+            self.get_logger().error('Enrollment service call timed out or failed.')
+
+
+def main(args=None):
+    """Entry point for enrollment CLI."""
+    parser = argparse.ArgumentParser(description='Enroll a person for face recognition.')
+    parser.add_argument('--name', type=str, required=True,
+                        help='Name of the person to enroll.')
+    parser.add_argument('--mode', type=str, default='face',
+                        choices=['face', 'voice', 'both'],
+                        help='Enrollment mode (default: face).')
+    parser.add_argument('--samples', type=int, default=10,
+                        help='Number of face samples to collect (default: 10).')
+
+    # Parse only known args so ROS2 remapping args pass through
+    parsed, remaining = parser.parse_known_args(args=sys.argv[1:])
+
+    rclpy.init(args=remaining)
+    node = EnrollmentCLI(parsed.name, parsed.mode, parsed.samples)
+    try:
+        pass  # Node does all work in __init__
+    finally:
+        node.destroy_node()
+        rclpy.shutdown()
+
+
+if __name__ == '__main__':
+    main()

jetson/ros2_ws/src/saltybot_social_face/saltybot_social_face/face_gallery.py

@@ -0,0 +1,206 @@
+"""
+face_gallery.py -- Persistent face embedding gallery backed by numpy .npz + JSON.
+
+Thread-safe gallery storage for face recognition.  Embeddings are stored in a
+.npz file, with a sidecar metadata.json containing names, sample counts, and
+enrollment timestamps.  Auto-increment IDs start at 1.
+"""
+
+import json
+import logging
+import os
+import threading
+import time
+from typing import Optional
+
+import numpy as np
+
+logger = logging.getLogger(__name__)
+
+
+class FaceGallery:
+    """Persistent, thread-safe face embedding gallery.
+
+    Args:
+        gallery_dir: Directory for gallery.npz and metadata.json files.
+    """
+
+    def __init__(self, gallery_dir: str):
+        self._gallery_dir = gallery_dir
+        self._npz_path = os.path.join(gallery_dir, 'gallery.npz')
+        self._meta_path = os.path.join(gallery_dir, 'metadata.json')
+        self._gallery: dict[int, dict] = {}
+        self._next_id = 1
+        self._lock = threading.Lock()
+
+    def load(self) -> None:
+        """Load gallery from disk. Populates internal gallery dict."""
+        with self._lock:
+            self._gallery = {}
+            self._next_id = 1
+
+            if not os.path.isfile(self._npz_path):
+                logger.info('No gallery file at %s, starting empty.', self._npz_path)
+                return
+
+            data = np.load(self._npz_path, allow_pickle=False)
+            meta: dict = {}
+            if os.path.isfile(self._meta_path):
+                with open(self._meta_path, 'r') as f:
+                    meta = json.load(f)
+
+            for key in data.files:
+                pid = int(key)
+                embedding = data[key].astype(np.float32)
+                norm = np.linalg.norm(embedding)
+                if norm > 0:
+                    embedding = embedding / norm
+                info = meta.get(str(pid), {})
+                self._gallery[pid] = {
+                    'name': info.get('name', f'person_{pid}'),
+                    'embedding': embedding,
+                    'samples': info.get('samples', 1),
+                    'enrolled_at': info.get('enrolled_at', 0.0),
+                }
+                if pid >= self._next_id:
+                    self._next_id = pid + 1
+
+            logger.info('Gallery loaded: %d persons from %s',
+                        len(self._gallery), self._npz_path)
+
+    def save(self) -> None:
+        """Save gallery to disk (npz + JSON sidecar)."""
+        with self._lock:
+            os.makedirs(self._gallery_dir, exist_ok=True)
+
+            arrays = {}
+            meta = {}
+            for pid, info in self._gallery.items():
+                arrays[str(pid)] = info['embedding']
+                meta[str(pid)] = {
+                    'name': info['name'],
+                    'samples': info['samples'],
+                    'enrolled_at': info['enrolled_at'],
+                }
+
+            np.savez(self._npz_path, **arrays)
+            with open(self._meta_path, 'w') as f:
+                json.dump(meta, f, indent=2)
+
+            logger.info('Gallery saved: %d persons to %s',
+                        len(self._gallery), self._npz_path)
+
+    def add_person(self, name: str, embedding: np.ndarray, samples: int = 1) -> int:
+        """Add a new person to the gallery.
+
+        Args:
+            name: Person's name.
+            embedding: L2-normalized 512-dim embedding.
+            samples: Number of samples used to compute the embedding.
+
+        Returns:
+            Assigned person_id (auto-increment integer).
+        """
+        with self._lock:
+            pid = self._next_id
+            self._next_id += 1
+            emb = embedding.astype(np.float32)
+            norm = np.linalg.norm(emb)
+            if norm > 0:
+                emb = emb / norm
+            self._gallery[pid] = {
+                'name': name,
+                'embedding': emb,
+                'samples': samples,
+                'enrolled_at': time.time(),
+            }
+            logger.info('Added person %d (%s), %d samples.', pid, name, samples)
+            return pid
+
+    def update_name(self, person_id: int, new_name: str) -> bool:
+        """Update a person's name.
+
+        Args:
+            person_id: The ID of the person to update.
+            new_name: New name string.
+
+        Returns:
+            True if the person was found and updated.
+        """
+        with self._lock:
+            if person_id not in self._gallery:
+                return False
+            self._gallery[person_id]['name'] = new_name
+            return True
+
+    def delete_person(self, person_id: int) -> bool:
+        """Remove a person from the gallery.
+
+        Args:
+            person_id: The ID of the person to delete.
+
+        Returns:
+            True if the person was found and removed.
+        """
+        with self._lock:
+            if person_id not in self._gallery:
+                return False
+            del self._gallery[person_id]
+            logger.info('Deleted person %d.', person_id)
+            return True
+
+    def get_all(self) -> list[dict]:
+        """Get all gallery entries.
+
+        Returns:
+            List of dicts with keys: person_id, name, embedding, samples, enrolled_at.
+        """
+        with self._lock:
+            result = []
+            for pid, info in self._gallery.items():
+                result.append({
+                    'person_id': pid,
+                    'name': info['name'],
+                    'embedding': info['embedding'].copy(),
+                    'samples': info['samples'],
+                    'enrolled_at': info['enrolled_at'],
+                })
+            return result
+
+    def match(self, query_embedding: np.ndarray, threshold: float = 0.35) -> tuple[int, str, float]:
+        """Match a query embedding against the gallery using cosine similarity.
+
+        Args:
+            query_embedding: L2-normalized 512-dim embedding.
+            threshold: Minimum cosine similarity for a match.
+
+        Returns:
+            (person_id, name, score) or (-1, '', 0.0) if no match.
+        """
+        with self._lock:
+            if not self._gallery:
+                return (-1, '', 0.0)
+
+            best_pid = -1
+            best_name = ''
+            best_score = 0.0
+
+            query = query_embedding.astype(np.float32)
+            norm = np.linalg.norm(query)
+            if norm > 0:
+                query = query / norm
+
+            for pid, info in self._gallery.items():
+                score = float(np.dot(query, info['embedding']))
+                if score > best_score:
+                    best_score = score
+                    best_pid = pid
+                    best_name = info['name']
+
+            if best_score >= threshold:
+                return (best_pid, best_name, best_score)
+            return (-1, '', 0.0)
+
+    def __len__(self) -> int:
+        with self._lock:
+            return len(self._gallery)

jetson/ros2_ws/src/saltybot_social_face/saltybot_social_face/face_recognition_node.py

@@ -0,0 +1,431 @@
+"""
+face_recognition_node.py -- ROS2 node for SCRFD face detection + ArcFace recognition.
+
+Pipeline:
+  1. Subscribe to /camera/color/image_raw (RealSense D435i color stream).
+  2. Run SCRFD face detection (TensorRT FP16 or ONNX fallback).
+  3. For each detected face, align and extract ArcFace embedding.
+  4. Match embedding against persistent gallery.
+  5. Publish FaceDetectionArray with identified faces.
+
+Services:
+  /social/enroll           -- Enroll a new person (collects N face samples).
+  /social/persons/list     -- List all enrolled persons.
+  /social/persons/delete   -- Delete a person from the gallery.
+  /social/persons/update   -- Update a person's name.
+"""
+
+import time
+
+import numpy as np
+
+import rclpy
+from rclpy.node import Node
+from rclpy.qos import QoSProfile, ReliabilityPolicy, HistoryPolicy, DurabilityPolicy
+
+import cv2
+from cv_bridge import CvBridge
+
+from sensor_msgs.msg import Image
+from builtin_interfaces.msg import Time
+
+from saltybot_social_msgs.msg import (
+    FaceDetection,
+    FaceDetectionArray,
+    FaceEmbedding,
+    FaceEmbeddingArray,
+)
+from saltybot_social_msgs.srv import (
+    EnrollPerson,
+    ListPersons,
+    DeletePerson,
+    UpdatePerson,
+)
+
+from .scrfd_detector import SCRFDDetector
+from .arcface_recognizer import ArcFaceRecognizer
+from .face_gallery import FaceGallery
+
+
+class FaceRecognitionNode(Node):
+    """ROS2 node: SCRFD face detection + ArcFace gallery matching."""
+
+    def __init__(self):
+        super().__init__('face_recognizer')
+        self._bridge = CvBridge()
+        self._frame_count = 0
+        self._fps_t0 = time.monotonic()
+
+        # -- Parameters --------------------------------------------------------
+        self.declare_parameter('scrfd_engine_path',
+                               '/mnt/nvme/saltybot/models/scrfd_2.5g.engine')
+        self.declare_parameter('scrfd_onnx_path',
+                               '/mnt/nvme/saltybot/models/scrfd_2.5g_bnkps.onnx')
+        self.declare_parameter('arcface_engine_path',
+                               '/mnt/nvme/saltybot/models/arcface_r50.engine')
+        self.declare_parameter('arcface_onnx_path',
+                               '/mnt/nvme/saltybot/models/arcface_r50.onnx')
+        self.declare_parameter('gallery_dir', '/mnt/nvme/saltybot/gallery')
+        self.declare_parameter('recognition_threshold', 0.35)
+        self.declare_parameter('publish_debug_image', False)
+        self.declare_parameter('max_faces', 10)
+        self.declare_parameter('scrfd_conf_thresh', 0.5)
+
+        self._recognition_threshold = self.get_parameter('recognition_threshold').value
+        self._pub_debug_flag = self.get_parameter('publish_debug_image').value
+        self._max_faces = self.get_parameter('max_faces').value
+
+        # -- Models ------------------------------------------------------------
+        self._detector = SCRFDDetector(
+            engine_path=self.get_parameter('scrfd_engine_path').value,
+            onnx_path=self.get_parameter('scrfd_onnx_path').value,
+            conf_thresh=self.get_parameter('scrfd_conf_thresh').value,
+        )
+        self._recognizer = ArcFaceRecognizer(
+            engine_path=self.get_parameter('arcface_engine_path').value,
+            onnx_path=self.get_parameter('arcface_onnx_path').value,
+        )
+
+        # -- Gallery -----------------------------------------------------------
+        gallery_dir = self.get_parameter('gallery_dir').value
+        self._gallery = FaceGallery(gallery_dir)
+        self._gallery.load()
+        self.get_logger().info('Gallery loaded: %d persons.', len(self._gallery))
+
+        # -- Enrollment state --------------------------------------------------
+        self._enrolling = None  # {name, samples_needed, collected: [embeddings]}
+
+        # -- QoS profiles ------------------------------------------------------
+        best_effort_qos = QoSProfile(
+            reliability=ReliabilityPolicy.BEST_EFFORT,
+            history=HistoryPolicy.KEEP_LAST,
+            depth=1,
+        )
+        reliable_qos = QoSProfile(
+            reliability=ReliabilityPolicy.RELIABLE,
+            durability=DurabilityPolicy.TRANSIENT_LOCAL,
+            history=HistoryPolicy.KEEP_LAST,
+            depth=1,
+        )
+
+        # -- Subscribers -------------------------------------------------------
+        self.create_subscription(
+            Image,
+            '/camera/color/image_raw',
+            self._on_image,
+            best_effort_qos,
+        )
+
+        # -- Publishers --------------------------------------------------------
+        self._pub_detections = self.create_publisher(
+            FaceDetectionArray, '/social/faces/detections', best_effort_qos)
+        self._pub_embeddings = self.create_publisher(
+            FaceEmbeddingArray, '/social/faces/embeddings', reliable_qos)
+        if self._pub_debug_flag:
+            self._pub_debug_img = self.create_publisher(
+                Image, '/social/faces/debug_image', best_effort_qos)
+
+        # -- Services ----------------------------------------------------------
+        self.create_service(EnrollPerson, '/social/enroll', self._handle_enroll)
+        self.create_service(ListPersons, '/social/persons/list', self._handle_list)
+        self.create_service(DeletePerson, '/social/persons/delete', self._handle_delete)
+        self.create_service(UpdatePerson, '/social/persons/update', self._handle_update)
+
+        # Publish initial gallery state
+        self._publish_gallery_embeddings()
+
+        self.get_logger().info('FaceRecognitionNode ready.')
+
+    # -- Image callback --------------------------------------------------------
+
+    def _on_image(self, msg: Image):
+        """Process incoming camera frame: detect, recognize, publish."""
+        try:
+            bgr = self._bridge.imgmsg_to_cv2(msg, desired_encoding='bgr8')
+        except Exception as e:
+            self.get_logger().error('Image decode error: %s', str(e),
+                                    throttle_duration_sec=5.0)
+            return
+
+        # Detect faces
+        detections = self._detector.detect(bgr)
+
+        # Limit face count
+        if len(detections) > self._max_faces:
+            detections = sorted(detections, key=lambda d: d['score'], reverse=True)
+            detections = detections[:self._max_faces]
+
+        # Build output message
+        det_array = FaceDetectionArray()
+        det_array.header = msg.header
+
+        for det in detections:
+            # Extract embedding and match gallery
+            embedding = self._recognizer.align_and_embed(bgr, det['kps'])
+            pid, pname, score = self._gallery.match(
+                embedding, self._recognition_threshold)
+
+            # Handle enrollment: collect embedding from largest face
+            if self._enrolling is not None:
+                self._enrollment_collect(det, embedding)
+
+            # Build FaceDetection message
+            face_msg = FaceDetection()
+            face_msg.header = msg.header
+            face_msg.face_id = pid
+            face_msg.person_name = pname
+            face_msg.confidence = det['score']
+            face_msg.recognition_score = score
+
+            bbox = det['bbox']
+            face_msg.bbox_x = bbox[0]
+            face_msg.bbox_y = bbox[1]
+            face_msg.bbox_w = bbox[2] - bbox[0]
+            face_msg.bbox_h = bbox[3] - bbox[1]
+
+            kps = det['kps']
+            for i in range(10):
+                face_msg.landmarks[i] = kps[i]
+
+            det_array.faces.append(face_msg)
+
+        self._pub_detections.publish(det_array)
+
+        # Debug image
+        if self._pub_debug_flag and hasattr(self, '_pub_debug_img'):
+            debug_img = self._draw_debug(bgr, detections, det_array.faces)
+            self._pub_debug_img.publish(
+                self._bridge.cv2_to_imgmsg(debug_img, encoding='bgr8'))
+
+        # FPS logging
+        self._frame_count += 1
+        if self._frame_count % 30 == 0:
+            elapsed = time.monotonic() - self._fps_t0
+            fps = 30.0 / elapsed if elapsed > 0 else 0.0
+            self._fps_t0 = time.monotonic()
+            self.get_logger().info(
+                'FPS: %.1f | faces: %d', fps, len(detections))
+
+    # -- Enrollment logic ------------------------------------------------------
+
+    def _enrollment_collect(self, det: dict, embedding: np.ndarray):
+        """Collect an embedding sample during enrollment (largest face only)."""
+        if self._enrolling is None:
+            return
+
+        # Only collect from the largest face (by bbox area)
+        bbox = det['bbox']
+        area = (bbox[2] - bbox[0]) * (bbox[3] - bbox[1])
+
+        if not hasattr(self, '_enroll_best_area'):
+            self._enroll_best_area = 0.0
+            self._enroll_best_embedding = None
+
+        if area > self._enroll_best_area:
+            self._enroll_best_area = area
+            self._enroll_best_embedding = embedding
+
+    def _enrollment_frame_end(self):
+        """Called at end of each frame to finalize enrollment sample collection."""
+        if self._enrolling is None or self._enroll_best_embedding is None:
+            return
+
+        self._enrolling['collected'].append(self._enroll_best_embedding)
+        self._enroll_best_area = 0.0
+        self._enroll_best_embedding = None
+
+        collected = len(self._enrolling['collected'])
+        needed = self._enrolling['samples_needed']
+        self.get_logger().info('Enrollment: %d/%d samples for "%s".',
+                               collected, needed, self._enrolling['name'])
+
+        if collected >= needed:
+            # Finalize enrollment
+            name = self._enrolling['name']
+            embeddings = self._enrolling['collected']
+            mean_emb = np.mean(embeddings, axis=0).astype(np.float32)
+            norm = np.linalg.norm(mean_emb)
+            if norm > 0:
+                mean_emb = mean_emb / norm
+
+            pid = self._gallery.add_person(name, mean_emb, samples=len(embeddings))
+            self._gallery.save()
+            self._publish_gallery_embeddings()
+
+            self.get_logger().info('Enrollment complete: person %d (%s).', pid, name)
+
+            # Store result for the service callback
+            self._enrolling['result_pid'] = pid
+            self._enrolling['done'] = True
+            self._enrolling = None
+
+    # -- Service handlers ------------------------------------------------------
+
+    def _handle_enroll(self, request, response):
+        """Handle EnrollPerson service: start collecting face samples."""
+        name = request.name.strip()
+        if not name:
+            response.success = False
+            response.message = 'Name cannot be empty.'
+            response.person_id = -1
+            return response
+
+        n_samples = request.n_samples if request.n_samples > 0 else 10
+
+        self.get_logger().info('Starting enrollment for "%s" (%d samples).',
+                               name, n_samples)
+
+        # Set enrollment state — frames will collect embeddings
+        self._enrolling = {
+            'name': name,
+            'samples_needed': n_samples,
+            'collected': [],
+            'done': False,
+            'result_pid': -1,
+        }
+        self._enroll_best_area = 0.0
+        self._enroll_best_embedding = None
+
+        # Spin until enrollment is done (blocking service)
+        rate = self.create_rate(10)  # 10 Hz check
+        timeout_sec = n_samples * 2.0 + 10.0  # generous timeout
+        t0 = time.monotonic()
+
+        while not self._enrolling.get('done', False):
+            # Finalize any pending frame collection
+            self._enrollment_frame_end()
+
+            if time.monotonic() - t0 > timeout_sec:
+                self._enrolling = None
+                response.success = False
+                response.message = f'Enrollment timed out after {timeout_sec:.0f}s.'
+                response.person_id = -1
+                return response
+
+            rclpy.spin_once(self, timeout_sec=0.1)
+
+        response.success = True
+        response.message = f'Enrolled "{name}" with {n_samples} samples.'
+        response.person_id = self._enrolling.get('result_pid', -1) if self._enrolling else -1
+
+        # Clean up (already set to None in _enrollment_frame_end on success)
+        return response
+
+    def _handle_list(self, request, response):
+        """Handle ListPersons service: return all gallery entries."""
+        entries = self._gallery.get_all()
+        for entry in entries:
+            emb_msg = FaceEmbedding()
+            emb_msg.person_id = entry['person_id']
+            emb_msg.person_name = entry['name']
+            emb_msg.embedding = entry['embedding'].tolist()
+            emb_msg.sample_count = entry['samples']
+
+            secs = int(entry['enrolled_at'])
+            nsecs = int((entry['enrolled_at'] - secs) * 1e9)
+            emb_msg.enrolled_at = Time(sec=secs, nanosec=nsecs)
+
+            response.persons.append(emb_msg)
+        return response
+
+    def _handle_delete(self, request, response):
+        """Handle DeletePerson service: remove a person from the gallery."""
+        if self._gallery.delete_person(request.person_id):
+            self._gallery.save()
+            self._publish_gallery_embeddings()
+            response.success = True
+            response.message = f'Deleted person {request.person_id}.'
+        else:
+            response.success = False
+            response.message = f'Person {request.person_id} not found.'
+        return response
+
+    def _handle_update(self, request, response):
+        """Handle UpdatePerson service: rename a person."""
+        new_name = request.new_name.strip()
+        if not new_name:
+            response.success = False
+            response.message = 'New name cannot be empty.'
+            return response
+
+        if self._gallery.update_name(request.person_id, new_name):
+            self._gallery.save()
+            self._publish_gallery_embeddings()
+            response.success = True
+            response.message = f'Updated person {request.person_id} to "{new_name}".'
+        else:
+            response.success = False
+            response.message = f'Person {request.person_id} not found.'
+        return response
+
+    # -- Gallery publishing ----------------------------------------------------
+
+    def _publish_gallery_embeddings(self):
+        """Publish current gallery as FaceEmbeddingArray (latched-like)."""
+        entries = self._gallery.get_all()
+        msg = FaceEmbeddingArray()
+        msg.header.stamp = self.get_clock().now().to_msg()
+
+        for entry in entries:
+            emb_msg = FaceEmbedding()
+            emb_msg.person_id = entry['person_id']
+            emb_msg.person_name = entry['name']
+            emb_msg.embedding = entry['embedding'].tolist()
+            emb_msg.sample_count = entry['samples']
+
+            secs = int(entry['enrolled_at'])
+            nsecs = int((entry['enrolled_at'] - secs) * 1e9)
+            emb_msg.enrolled_at = Time(sec=secs, nanosec=nsecs)
+
+            msg.embeddings.append(emb_msg)
+
+        self._pub_embeddings.publish(msg)
+
+    # -- Debug image -----------------------------------------------------------
+
+    def _draw_debug(self, bgr: np.ndarray, detections: list[dict],
+                    face_msgs: list) -> np.ndarray:
+        """Draw bounding boxes, landmarks, and names on the image."""
+        vis = bgr.copy()
+        for det, face_msg in zip(detections, face_msgs):
+            bbox = det['bbox']
+            x1, y1, x2, y2 = int(bbox[0]), int(bbox[1]), int(bbox[2]), int(bbox[3])
+
+            # Color: green if recognized, yellow if unknown
+            if face_msg.face_id >= 0:
+                color = (0, 255, 0)
+                label = f'{face_msg.person_name} ({face_msg.recognition_score:.2f})'
+            else:
+                color = (0, 255, 255)
+                label = f'unknown ({face_msg.confidence:.2f})'
+
+            cv2.rectangle(vis, (x1, y1), (x2, y2), color, 2)
+            cv2.putText(vis, label, (x1, y1 - 8),
+                        cv2.FONT_HERSHEY_SIMPLEX, 0.5, color, 1)
+
+            # Draw landmarks
+            kps = det['kps']
+            for k in range(5):
+                px, py = int(kps[k * 2]), int(kps[k * 2 + 1])
+                cv2.circle(vis, (px, py), 2, (0, 0, 255), -1)
+
+        return vis
+
+
+# -- Entry point ---------------------------------------------------------------
+
+def main(args=None):
+    """ROS2 entry point for face_recognition node."""
+    rclpy.init(args=args)
+    node = FaceRecognitionNode()
+    try:
+        rclpy.spin(node)
+    except KeyboardInterrupt:
+        pass
+    finally:
+        node.destroy_node()
+        rclpy.shutdown()
+
+
+if __name__ == '__main__':
+    main()

jetson/ros2_ws/src/saltybot_social_face/saltybot_social_face/scrfd_detector.py

@@ -0,0 +1,350 @@
+"""
+scrfd_detector.py -- SCRFD face detection with TensorRT FP16 + ONNX fallback.
+
+SCRFD (Sample and Computation Redistribution for Face Detection) produces
+9 output tensors across 3 strides (8, 16, 32), each with score, bbox, and
+keypoint branches.  This module handles anchor generation, bbox/keypoint
+decoding, and NMS to produce final face detections.
+"""
+
+import os
+import logging
+from typing import Optional
+
+import numpy as np
+import cv2
+
+logger = logging.getLogger(__name__)
+
+_STRIDES = [8, 16, 32]
+_NUM_ANCHORS = 2  # anchors per cell per stride
+
+
+# -- Inference backends --------------------------------------------------------
+
+class _TRTBackend:
+    """TensorRT inference engine for SCRFD."""
+
+    def __init__(self, engine_path: str):
+        import tensorrt as trt
+        import pycuda.driver as cuda
+        import pycuda.autoinit  # noqa: F401
+
+        self._cuda = cuda
+        trt_logger = trt.Logger(trt.Logger.WARNING)
+        with open(engine_path, 'rb') as f, trt.Runtime(trt_logger) as runtime:
+            self._engine = runtime.deserialize_cuda_engine(f.read())
+        self._context = self._engine.create_execution_context()
+
+        self._inputs = []
+        self._outputs = []
+        self._output_names = []
+        self._bindings = []
+        for i in range(self._engine.num_io_tensors):
+            name = self._engine.get_tensor_name(i)
+            dtype = trt.nptype(self._engine.get_tensor_dtype(name))
+            shape = tuple(self._engine.get_tensor_shape(name))
+            nbytes = int(np.prod(shape)) * np.dtype(dtype).itemsize
+            host_mem = cuda.pagelocked_empty(shape, dtype)
+            device_mem = cuda.mem_alloc(nbytes)
+            self._bindings.append(int(device_mem))
+            if self._engine.get_tensor_mode(name) == trt.TensorIOMode.INPUT:
+                self._inputs.append({'host': host_mem, 'device': device_mem})
+            else:
+                self._outputs.append({'host': host_mem, 'device': device_mem,
+                                      'shape': shape})
+                self._output_names.append(name)
+        self._stream = cuda.Stream()
+
+    def infer(self, input_data: np.ndarray) -> list[np.ndarray]:
+        """Run inference and return output tensors."""
+        np.copyto(self._inputs[0]['host'], input_data.ravel())
+        self._cuda.memcpy_htod_async(
+            self._inputs[0]['device'], self._inputs[0]['host'], self._stream)
+        self._context.execute_async_v2(self._bindings, self._stream.handle)
+        results = []
+        for out in self._outputs:
+            self._cuda.memcpy_dtoh_async(out['host'], out['device'], self._stream)
+        self._stream.synchronize()
+        for out in self._outputs:
+            results.append(out['host'].reshape(out['shape']).copy())
+        return results
+
+
+class _ONNXBackend:
+    """ONNX Runtime inference (CUDA EP with CPU fallback)."""
+
+    def __init__(self, onnx_path: str):
+        import onnxruntime as ort
+        providers = ['CUDAExecutionProvider', 'CPUExecutionProvider']
+        self._session = ort.InferenceSession(onnx_path, providers=providers)
+        self._input_name = self._session.get_inputs()[0].name
+        self._output_names = [o.name for o in self._session.get_outputs()]
+
+    def infer(self, input_data: np.ndarray) -> list[np.ndarray]:
+        """Run inference and return output tensors."""
+        return self._session.run(None, {self._input_name: input_data})
+
+
+# -- NMS ----------------------------------------------------------------------
+
+def _nms(boxes: np.ndarray, scores: np.ndarray, iou_thresh: float) -> list[int]:
+    """Non-maximum suppression. boxes: [N, 4] as x1,y1,x2,y2."""
+    if len(boxes) == 0:
+        return []
+
+    x1, y1, x2, y2 = boxes[:, 0], boxes[:, 1], boxes[:, 2], boxes[:, 3]
+    areas = (x2 - x1) * (y2 - y1)
+    order = scores.argsort()[::-1]
+    keep = []
+
+    while order.size > 0:
+        i = order[0]
+        keep.append(int(i))
+        if order.size == 1:
+            break
+        xx1 = np.maximum(x1[i], x1[order[1:]])
+        yy1 = np.maximum(y1[i], y1[order[1:]])
+        xx2 = np.minimum(x2[i], x2[order[1:]])
+        yy2 = np.minimum(y2[i], y2[order[1:]])
+        inter = np.maximum(0.0, xx2 - xx1) * np.maximum(0.0, yy2 - yy1)
+        iou = inter / (areas[i] + areas[order[1:]] - inter + 1e-6)
+        remaining = np.where(iou <= iou_thresh)[0]
+        order = order[remaining + 1]
+
+    return keep
+
+
+# -- Anchor generation ---------------------------------------------------------
+
+def _generate_anchors(input_h: int, input_w: int) -> dict[int, np.ndarray]:
+    """Generate anchor centers for each stride.
+
+    Returns dict mapping stride -> array of shape [H*W*num_anchors, 2],
+    where each row is (cx, cy) in input pixel coordinates.
+    """
+    anchors = {}
+    for stride in _STRIDES:
+        feat_h = input_h // stride
+        feat_w = input_w // stride
+        grid_y, grid_x = np.mgrid[:feat_h, :feat_w]
+        centers = np.stack([grid_x.ravel(), grid_y.ravel()], axis=1).astype(np.float32)
+        centers = (centers + 0.5) * stride
+        # Repeat for num_anchors per cell
+        centers = np.tile(centers, (_NUM_ANCHORS, 1))  # [H*W*2, 2]
+        # Interleave properly: [anchor0_cell0, anchor1_cell0, anchor0_cell1, ...]
+        centers = np.repeat(
+            np.stack([grid_x.ravel(), grid_y.ravel()], axis=1).astype(np.float32),
+            _NUM_ANCHORS, axis=0
+        )
+        centers = (centers + 0.5) * stride
+        anchors[stride] = centers
+    return anchors
+
+
+# -- Main detector class -------------------------------------------------------
+
+class SCRFDDetector:
+    """SCRFD face detector with TensorRT FP16 and ONNX fallback.
+
+    Args:
+        engine_path: Path to TensorRT engine file.
+        onnx_path: Path to ONNX model file (used if engine not available).
+        conf_thresh: Minimum confidence for detections.
+        nms_iou: IoU threshold for NMS.
+        input_size: Model input resolution (square).
+    """
+
+    def __init__(
+        self,
+        engine_path: str = '',
+        onnx_path: str = '',
+        conf_thresh: float = 0.5,
+        nms_iou: float = 0.4,
+        input_size: int = 640,
+    ):
+        self._conf_thresh = conf_thresh
+        self._nms_iou = nms_iou
+        self._input_size = input_size
+        self._backend: Optional[_TRTBackend | _ONNXBackend] = None
+        self._anchors = _generate_anchors(input_size, input_size)
+
+        # Try TRT first, then ONNX
+        if engine_path and os.path.isfile(engine_path):
+            try:
+                self._backend = _TRTBackend(engine_path)
+                logger.info('SCRFD TensorRT backend loaded: %s', engine_path)
+                return
+            except Exception as e:
+                logger.warning('SCRFD TRT load failed (%s), trying ONNX', e)
+
+        if onnx_path and os.path.isfile(onnx_path):
+            try:
+                self._backend = _ONNXBackend(onnx_path)
+                logger.info('SCRFD ONNX backend loaded: %s', onnx_path)
+                return
+            except Exception as e:
+                logger.error('SCRFD ONNX load failed: %s', e)
+
+        logger.error('No SCRFD model loaded. Detection will be unavailable.')
+
+    @property
+    def is_loaded(self) -> bool:
+        """Return True if a backend is loaded and ready."""
+        return self._backend is not None
+
+    def detect(self, bgr: np.ndarray) -> list[dict]:
+        """Detect faces in a BGR image.
+
+        Args:
+            bgr: Input image in BGR format, shape (H, W, 3).
+
+        Returns:
+            List of dicts with keys:
+                bbox: [x1, y1, x2, y2] in original image coordinates
+                kps: [x0,y0, x1,y1, ..., x4,y4] — 10 floats, 5 landmarks
+                score: detection confidence
+        """
+        if self._backend is None:
+            return []
+
+        orig_h, orig_w = bgr.shape[:2]
+        tensor, scale, pad_w, pad_h = self._preprocess(bgr)
+        outputs = self._backend.infer(tensor)
+        detections = self._decode_outputs(outputs)
+        detections = self._rescale(detections, scale, pad_w, pad_h, orig_w, orig_h)
+        return detections
+
+    def _preprocess(self, bgr: np.ndarray) -> tuple[np.ndarray, float, int, int]:
+        """Resize to input_size x input_size with letterbox padding, normalize."""
+        h, w = bgr.shape[:2]
+        size = self._input_size
+        scale = min(size / h, size / w)
+        new_w, new_h = int(w * scale), int(h * scale)
+        pad_w = (size - new_w) // 2
+        pad_h = (size - new_h) // 2
+
+        resized = cv2.resize(bgr, (new_w, new_h), interpolation=cv2.INTER_LINEAR)
+        canvas = np.full((size, size, 3), 0, dtype=np.uint8)
+        canvas[pad_h:pad_h + new_h, pad_w:pad_w + new_w] = resized
+
+        # Normalize: subtract 127.5, divide 128.0
+        blob = canvas.astype(np.float32)
+        blob = (blob - 127.5) / 128.0
+        # HWC -> NCHW
+        blob = blob.transpose(2, 0, 1)[np.newaxis]
+        blob = np.ascontiguousarray(blob)
+        return blob, scale, pad_w, pad_h
+
+    def _decode_outputs(self, outputs: list[np.ndarray]) -> list[dict]:
+        """Decode SCRFD 9-output format into face detections.
+
+        SCRFD outputs 9 tensors, 3 per stride (score, bbox, kps):
+            score_8, bbox_8, kps_8, score_16, bbox_16, kps_16, score_32, bbox_32, kps_32
+        """
+        all_scores = []
+        all_bboxes = []
+        all_kps = []
+
+        for idx, stride in enumerate(_STRIDES):
+            score_out = outputs[idx * 3].squeeze()       # [H*W*num_anchors]
+            bbox_out = outputs[idx * 3 + 1].squeeze()    # [H*W*num_anchors, 4]
+            kps_out = outputs[idx * 3 + 2].squeeze()     # [H*W*num_anchors, 10]
+
+            if score_out.ndim == 0:
+                continue
+
+            # Ensure proper shapes
+            if score_out.ndim == 1:
+                n = score_out.shape[0]
+            else:
+                n = score_out.shape[0]
+                score_out = score_out.ravel()
+
+            if bbox_out.ndim == 1:
+                bbox_out = bbox_out.reshape(-1, 4)
+            if kps_out.ndim == 1:
+                kps_out = kps_out.reshape(-1, 10)
+
+            # Filter by confidence
+            mask = score_out > self._conf_thresh
+            if not mask.any():
+                continue
+
+            scores = score_out[mask]
+            bboxes = bbox_out[mask]
+            kps = kps_out[mask]
+
+            anchors = self._anchors[stride]
+            # Trim or pad anchors to match output count
+            if anchors.shape[0] > n:
+                anchors = anchors[:n]
+            elif anchors.shape[0] < n:
+                continue
+
+            anchors = anchors[mask]
+
+            # Decode bboxes: center = anchor + pred[:2]*stride, size = exp(pred[2:])*stride
+            cx = anchors[:, 0] + bboxes[:, 0] * stride
+            cy = anchors[:, 1] + bboxes[:, 1] * stride
+            w = np.exp(bboxes[:, 2]) * stride
+            h = np.exp(bboxes[:, 3]) * stride
+            x1 = cx - w / 2.0
+            y1 = cy - h / 2.0
+            x2 = cx + w / 2.0
+            y2 = cy + h / 2.0
+            decoded_bboxes = np.stack([x1, y1, x2, y2], axis=1)
+
+            # Decode keypoints: kp = anchor + pred * stride
+            decoded_kps = np.zeros_like(kps)
+            for k in range(5):
+                decoded_kps[:, k * 2] = anchors[:, 0] + kps[:, k * 2] * stride
+                decoded_kps[:, k * 2 + 1] = anchors[:, 1] + kps[:, k * 2 + 1] * stride
+
+            all_scores.append(scores)
+            all_bboxes.append(decoded_bboxes)
+            all_kps.append(decoded_kps)
+
+        if not all_scores:
+            return []
+
+        scores = np.concatenate(all_scores)
+        bboxes = np.concatenate(all_bboxes)
+        kps = np.concatenate(all_kps)
+
+        # NMS
+        keep = _nms(bboxes, scores, self._nms_iou)
+
+        results = []
+        for i in keep:
+            results.append({
+                'bbox': bboxes[i].tolist(),
+                'kps': kps[i].tolist(),
+                'score': float(scores[i]),
+            })
+        return results
+
+    def _rescale(
+        self,
+        detections: list[dict],
+        scale: float,
+        pad_w: int,
+        pad_h: int,
+        orig_w: int,
+        orig_h: int,
+    ) -> list[dict]:
+        """Rescale detections from model input space to original image space."""
+        for det in detections:
+            bbox = det['bbox']
+            bbox[0] = max(0.0, (bbox[0] - pad_w) / scale)
+            bbox[1] = max(0.0, (bbox[1] - pad_h) / scale)
+            bbox[2] = min(float(orig_w), (bbox[2] - pad_w) / scale)
+            bbox[3] = min(float(orig_h), (bbox[3] - pad_h) / scale)
+            det['bbox'] = bbox
+
+            kps = det['kps']
+            for k in range(5):
+                kps[k * 2] = (kps[k * 2] - pad_w) / scale
+                kps[k * 2 + 1] = (kps[k * 2 + 1] - pad_h) / scale
+            det['kps'] = kps
+        return detections

jetson/ros2_ws/src/saltybot_social_face/scripts/build_face_trt_engines.py

@@ -0,0 +1,112 @@
+#!/usr/bin/env python3
+"""
+build_face_trt_engines.py -- Build TensorRT FP16 engines for SCRFD and ArcFace.
+
+Converts ONNX model files to optimized TensorRT engines with FP16 precision
+for fast inference on Jetson Orin Nano Super.
+
+Usage:
+    python3 build_face_trt_engines.py \
+        --scrfd-onnx /path/to/scrfd_2.5g_bnkps.onnx \
+        --arcface-onnx /path/to/arcface_r50.onnx \
+        --output-dir /mnt/nvme/saltybot/models \
+        --fp16 --workspace-mb 1024
+"""
+
+import argparse
+import os
+import time
+
+
+def build_engine(onnx_path: str, engine_path: str, fp16: bool, workspace_mb: int):
+    """Build a TensorRT engine from an ONNX model.
+
+    Args:
+        onnx_path: Path to the source ONNX model file.
+        engine_path: Output path for the serialized TensorRT engine.
+        fp16: Enable FP16 precision.
+        workspace_mb: Maximum workspace size in megabytes.
+    """
+    import tensorrt as trt
+
+    logger = trt.Logger(trt.Logger.INFO)
+    builder = trt.Builder(logger)
+    network = builder.create_network(
+        1 << int(trt.NetworkDefinitionCreationFlag.EXPLICIT_BATCH))
+    parser = trt.OnnxParser(network, logger)
+
+    print(f'Parsing ONNX model: {onnx_path}')
+    t0 = time.monotonic()
+
+    with open(onnx_path, 'rb') as f:
+        if not parser.parse(f.read()):
+            for i in range(parser.num_errors):
+                print(f'  ONNX parse error: {parser.get_error(i)}')
+            raise RuntimeError(f'Failed to parse {onnx_path}')
+
+    parse_time = time.monotonic() - t0
+    print(f'  Parsed in {parse_time:.1f}s')
+
+    config = builder.create_builder_config()
+    config.set_memory_pool_limit(trt.MemoryPoolType.WORKSPACE,
+                                  workspace_mb * (1 << 20))
+    if fp16:
+        if builder.platform_has_fast_fp16:
+            config.set_flag(trt.BuilderFlag.FP16)
+            print('  FP16 enabled.')
+        else:
+            print('  Warning: FP16 not supported on this platform, using FP32.')
+
+    print(f'Building engine (this may take several minutes)...')
+    t0 = time.monotonic()
+    serialized = builder.build_serialized_network(network, config)
+    build_time = time.monotonic() - t0
+
+    if serialized is None:
+        raise RuntimeError('Engine build failed.')
+
+    os.makedirs(os.path.dirname(engine_path) or '.', exist_ok=True)
+    with open(engine_path, 'wb') as f:
+        f.write(serialized)
+
+    size_mb = os.path.getsize(engine_path) / (1 << 20)
+    print(f'  Engine saved: {engine_path} ({size_mb:.1f} MB, built in {build_time:.1f}s)')
+
+
+def main():
+    """Main entry point for TRT engine building."""
+    parser = argparse.ArgumentParser(
+        description='Build TensorRT FP16 engines for SCRFD and ArcFace.')
+    parser.add_argument('--scrfd-onnx', type=str, default='',
+                        help='Path to SCRFD ONNX model.')
+    parser.add_argument('--arcface-onnx', type=str, default='',
+                        help='Path to ArcFace ONNX model.')
+    parser.add_argument('--output-dir', type=str,
+                        default='/mnt/nvme/saltybot/models',
+                        help='Output directory for engine files.')
+    parser.add_argument('--fp16', action='store_true', default=True,
+                        help='Enable FP16 precision (default: True).')
+    parser.add_argument('--no-fp16', action='store_false', dest='fp16',
+                        help='Disable FP16 (use FP32 only).')
+    parser.add_argument('--workspace-mb', type=int, default=1024,
+                        help='TRT workspace size in MB (default: 1024).')
+    args = parser.parse_args()
+
+    if not args.scrfd_onnx and not args.arcface_onnx:
+        parser.error('At least one of --scrfd-onnx or --arcface-onnx is required.')
+
+    if args.scrfd_onnx:
+        engine_path = os.path.join(args.output_dir, 'scrfd_2.5g.engine')
+        print(f'\n=== Building SCRFD engine ===')
+        build_engine(args.scrfd_onnx, engine_path, args.fp16, args.workspace_mb)
+
+    if args.arcface_onnx:
+        engine_path = os.path.join(args.output_dir, 'arcface_r50.engine')
+        print(f'\n=== Building ArcFace engine ===')
+        build_engine(args.arcface_onnx, engine_path, args.fp16, args.workspace_mb)
+
+    print('\nDone.')
+
+
+if __name__ == '__main__':
+    main()

jetson/ros2_ws/src/saltybot_social_face/setup.cfg

@@ -0,0 +1,4 @@
+[develop]
+script_dir=$base/lib/saltybot_social_face
+[install]
+install_scripts=$base/lib/saltybot_social_face

jetson/ros2_ws/src/saltybot_social_face/setup.py

@@ -0,0 +1,30 @@
+"""Setup for saltybot_social_face package."""
+
+from setuptools import find_packages, setup
+
+package_name = 'saltybot_social_face'
+
+setup(
+    name=package_name,
+    version='0.1.0',
+    packages=find_packages(exclude=['test']),
+    data_files=[
+        ('share/ament_index/resource_index/packages', ['resource/' + package_name]),
+        ('share/' + package_name, ['package.xml']),
+        ('share/' + package_name + '/launch', ['launch/face_recognition.launch.py']),
+        ('share/' + package_name + '/config', ['config/face_recognition_params.yaml']),
+    ],
+    install_requires=['setuptools'],
+    zip_safe=True,
+    maintainer='seb',
+    maintainer_email='seb@vayrette.com',
+    description='SCRFD face detection and ArcFace recognition for SaltyBot social interactions',
+    license='MIT',
+    tests_require=['pytest'],
+    entry_points={
+        'console_scripts': [
+            'face_recognition = saltybot_social_face.face_recognition_node:main',
+            'enrollment_cli = saltybot_social_face.enrollment_cli:main',
+        ],
+    },
+)

jetson/ros2_ws/src/saltybot_social_msgs/CMakeLists.txt

@@ -0,0 +1,24 @@
+cmake_minimum_required(VERSION 3.8)
+project(saltybot_social_msgs)
+
+find_package(ament_cmake REQUIRED)
+find_package(rosidl_default_generators REQUIRED)
+find_package(std_msgs REQUIRED)
+find_package(geometry_msgs REQUIRED)
+find_package(builtin_interfaces REQUIRED)
+
+rosidl_generate_interfaces(${PROJECT_NAME}
+  "msg/FaceDetection.msg"
+  "msg/FaceDetectionArray.msg"
+  "msg/FaceEmbedding.msg"
+  "msg/FaceEmbeddingArray.msg"
+  "msg/PersonState.msg"
+  "msg/PersonStateArray.msg"
+  "srv/EnrollPerson.srv"
+  "srv/ListPersons.srv"
+  "srv/DeletePerson.srv"
+  "srv/UpdatePerson.srv"
+  DEPENDENCIES std_msgs geometry_msgs builtin_interfaces
+)
+
+ament_package()

jetson/ros2_ws/src/saltybot_social_msgs/msg/FaceDetection.msg

@@ -0,0 +1,10 @@
+std_msgs/Header header
+int32 face_id
+string person_name
+float32 confidence
+float32 recognition_score
+float32 bbox_x
+float32 bbox_y
+float32 bbox_w
+float32 bbox_h
+float32[10] landmarks

jetson/ros2_ws/src/saltybot_social_msgs/msg/FaceDetectionArray.msg

@@ -0,0 +1,2 @@
+std_msgs/Header header
+saltybot_social_msgs/FaceDetection[] faces

jetson/ros2_ws/src/saltybot_social_msgs/msg/FaceEmbedding.msg

@@ -0,0 +1,5 @@
+int32 person_id
+string person_name
+float32[] embedding
+builtin_interfaces/Time enrolled_at
+int32 sample_count

jetson/ros2_ws/src/saltybot_social_msgs/msg/FaceEmbeddingArray.msg

@@ -0,0 +1,2 @@
+std_msgs/Header header
+saltybot_social_msgs/FaceEmbedding[] embeddings

jetson/ros2_ws/src/saltybot_social_msgs/msg/PersonState.msg

@@ -0,0 +1,19 @@
+std_msgs/Header header
+int32 person_id
+string person_name
+int32 face_id
+string speaker_id
+string uwb_anchor_id
+geometry_msgs/Point position
+float32 distance
+float32 bearing_deg
+uint8 state
+uint8 STATE_UNKNOWN=0
+uint8 STATE_APPROACHING=1
+uint8 STATE_ENGAGED=2
+uint8 STATE_TALKING=3
+uint8 STATE_LEAVING=4
+uint8 STATE_ABSENT=5
+float32 engagement_score
+builtin_interfaces/Time last_seen
+int32 camera_id

jetson/ros2_ws/src/saltybot_social_msgs/msg/PersonStateArray.msg

@@ -0,0 +1,3 @@
+std_msgs/Header header
+saltybot_social_msgs/PersonState[] persons
+int32 primary_attention_id

jetson/ros2_ws/src/saltybot_social_msgs/package.xml

@@ -0,0 +1,19 @@
+<?xml version="1.0"?>
+<?xml-model href="http://download.ros.org/schema/package_format3.xsd" schematypens="http://www.w3.org/2001/XMLSchema"?>
+<package format="3">
+  <name>saltybot_social_msgs</name>
+  <version>0.1.0</version>
+  <description>Custom ROS2 messages and services for saltybot social capabilities</description>
+  <maintainer email="seb@vayrette.com">seb</maintainer>
+  <license>MIT</license>
+  <buildtool_depend>ament_cmake</buildtool_depend>
+  <depend>std_msgs</depend>
+  <depend>geometry_msgs</depend>
+  <depend>builtin_interfaces</depend>
+  <build_depend>rosidl_default_generators</build_depend>
+  <exec_depend>rosidl_default_runtime</exec_depend>
+  <member_of_group>rosidl_interface_packages</member_of_group>
+  <export>
+    <build_type>ament_cmake</build_type>
+  </export>
+</package>

jetson/ros2_ws/src/saltybot_social_msgs/srv/DeletePerson.srv

@@ -0,0 +1,4 @@
+int32 person_id
+---
+bool success
+string message

jetson/ros2_ws/src/saltybot_social_msgs/srv/EnrollPerson.srv

@@ -0,0 +1,7 @@
+string name
+string mode
+int32 n_samples
+---
+bool success
+string message
+int32 person_id

jetson/ros2_ws/src/saltybot_social_msgs/srv/ListPersons.srv

@@ -0,0 +1,2 @@
+---
+saltybot_social_msgs/FaceEmbedding[] persons

jetson/ros2_ws/src/saltybot_social_msgs/srv/UpdatePerson.srv

@@ -0,0 +1,5 @@
+int32 person_id
+string new_name
+---
+bool success
+string message

jetson/ros2_ws/src/saltybot_social_nav/config/social_nav_params.yaml

@@ -0,0 +1,22 @@
+social_nav:
+  ros__parameters:
+    follow_mode: 'shadow'
+    follow_distance: 1.2
+    lead_distance: 2.0
+    orbit_radius: 1.5
+    max_linear_speed: 1.0
+    max_linear_speed_fast: 5.5
+    max_angular_speed: 1.0
+    goal_tolerance: 0.3
+    routes_dir: '/mnt/nvme/saltybot/routes'
+    home_x: 0.0
+    home_y: 0.0
+    map_resolution: 0.05
+    obstacle_inflation_cells: 3
+
+midas_depth:
+  ros__parameters:
+    onnx_path: '/mnt/nvme/saltybot/models/midas_small.onnx'
+    engine_path: '/mnt/nvme/saltybot/models/midas_small.engine'
+    process_rate: 5.0
+    output_scale: 1.0

jetson/ros2_ws/src/saltybot_social_nav/launch/social_nav.launch.py

@@ -0,0 +1,57 @@
+"""Launch file for saltybot social navigation."""
+
+from launch import LaunchDescription
+from launch.actions import DeclareLaunchArgument
+from launch.substitutions import LaunchConfiguration
+from launch_ros.actions import Node
+
+
+def generate_launch_description():
+    return LaunchDescription([
+        DeclareLaunchArgument('follow_mode', default_value='shadow',
+                              description='Follow mode: shadow/lead/side/orbit/loose/tight'),
+        DeclareLaunchArgument('follow_distance', default_value='1.2',
+                              description='Follow distance in meters'),
+        DeclareLaunchArgument('max_linear_speed', default_value='1.0',
+                              description='Max linear speed (m/s)'),
+        DeclareLaunchArgument('routes_dir',
+                              default_value='/mnt/nvme/saltybot/routes',
+                              description='Directory for saved routes'),
+
+        Node(
+            package='saltybot_social_nav',
+            executable='social_nav',
+            name='social_nav',
+            output='screen',
+            parameters=[{
+                'follow_mode': LaunchConfiguration('follow_mode'),
+                'follow_distance': LaunchConfiguration('follow_distance'),
+                'max_linear_speed': LaunchConfiguration('max_linear_speed'),
+                'routes_dir': LaunchConfiguration('routes_dir'),
+            }],
+        ),
+
+        Node(
+            package='saltybot_social_nav',
+            executable='midas_depth',
+            name='midas_depth',
+            output='screen',
+            parameters=[{
+                'onnx_path': '/mnt/nvme/saltybot/models/midas_small.onnx',
+                'engine_path': '/mnt/nvme/saltybot/models/midas_small.engine',
+                'process_rate': 5.0,
+                'output_scale': 1.0,
+            }],
+        ),
+
+        Node(
+            package='saltybot_social_nav',
+            executable='waypoint_teacher',
+            name='waypoint_teacher',
+            output='screen',
+            parameters=[{
+                'routes_dir': LaunchConfiguration('routes_dir'),
+                'recording_interval': 0.5,
+            }],
+        ),
+    ])

jetson/ros2_ws/src/saltybot_social_nav/package.xml

@@ -0,0 +1,28 @@
+<?xml version="1.0"?>
+<?xml-model href="http://download.ros.org/schema/package_format3.xsd" schematypens="http://www.w3.org/2001/XMLSchema"?>
+<package format="3">
+  <name>saltybot_social_nav</name>
+  <version>0.1.0</version>
+  <description>Social navigation for saltybot: follow modes, waypoint teaching, A* avoidance, MiDaS depth</description>
+  <maintainer email="seb@vayrette.com">seb</maintainer>
+  <license>MIT</license>
+
+  <depend>rclpy</depend>
+  <depend>std_msgs</depend>
+  <depend>geometry_msgs</depend>
+  <depend>nav_msgs</depend>
+  <depend>sensor_msgs</depend>
+  <depend>cv_bridge</depend>
+  <depend>tf2_ros</depend>
+  <depend>tf2_geometry_msgs</depend>
+  <depend>saltybot_social_msgs</depend>
+
+  <test_depend>ament_copyright</test_depend>
+  <test_depend>ament_flake8</test_depend>
+  <test_depend>ament_pep257</test_depend>
+  <test_depend>python3-pytest</test_depend>
+
+  <export>
+    <build_type>ament_python</build_type>
+  </export>
+</package>

jetson/ros2_ws/src/saltybot_social_nav/saltybot_social_nav/astar.py

@@ -0,0 +1,82 @@
+"""astar.py -- A* path planner for saltybot social navigation."""
+
+import heapq
+import numpy as np
+
+
+def astar(grid: np.ndarray, start: tuple, goal: tuple,
+          obstacle_val: int = 100) -> list | None:
+    """
+    A* on a 2D occupancy grid (row, col indexing).
+
+    Args:
+        grid: 2D numpy array, values 0=free, >=obstacle_val=obstacle
+        start: (row, col) start cell
+        goal: (row, col) goal cell
+        obstacle_val: cells with value >= obstacle_val are blocked
+
+    Returns:
+        List of (row, col) tuples from start to goal, or None if no path.
+    """
+    rows, cols = grid.shape
+
+    def h(a, b):
+        return abs(a[0] - b[0]) + abs(a[1] - b[1])  # Manhattan heuristic
+
+    open_set = []
+    heapq.heappush(open_set, (h(start, goal), 0, start))
+    came_from = {}
+    g_score = {start: 0}
+
+    # 8-directional movement
+    neighbors_delta = [
+        (-1, -1), (-1, 0), (-1, 1),
+        (0, -1),           (0, 1),
+        (1, -1),  (1, 0),  (1, 1),
+    ]
+
+    while open_set:
+        _, cost, current = heapq.heappop(open_set)
+
+        if current == goal:
+            path = []
+            while current in came_from:
+                path.append(current)
+                current = came_from[current]
+            path.append(start)
+            return list(reversed(path))
+
+        if cost > g_score.get(current, float('inf')):
+            continue
+
+        for dr, dc in neighbors_delta:
+            nr, nc = current[0] + dr, current[1] + dc
+            if not (0 <= nr < rows and 0 <= nc < cols):
+                continue
+            if grid[nr, nc] >= obstacle_val:
+                continue
+            move_cost = 1.414 if (dr != 0 and dc != 0) else 1.0
+            new_g = g_score[current] + move_cost
+            neighbor = (nr, nc)
+            if new_g < g_score.get(neighbor, float('inf')):
+                g_score[neighbor] = new_g
+                f = new_g + h(neighbor, goal)
+                came_from[neighbor] = current
+                heapq.heappush(open_set, (f, new_g, neighbor))
+
+    return None  # No path found
+
+
+def inflate_obstacles(grid: np.ndarray, inflation_radius_cells: int) -> np.ndarray:
+    """Inflate obstacles for robot footprint safety."""
+    from scipy.ndimage import binary_dilation
+
+    obstacle_mask = grid >= 50
+    kernel = np.ones(
+        (2 * inflation_radius_cells + 1, 2 * inflation_radius_cells + 1),
+        dtype=bool,
+    )
+    inflated = binary_dilation(obstacle_mask, structure=kernel)
+    result = grid.copy()
+    result[inflated] = 100
+    return result

jetson/ros2_ws/src/saltybot_social_nav/saltybot_social_nav/follow_modes.py

@@ -0,0 +1,82 @@
+"""follow_modes.py -- Follow mode geometry for saltybot social navigation."""
+
+import math
+from enum import Enum
+import numpy as np
+
+
+class FollowMode(Enum):
+    SHADOW = 'shadow'    # stay directly behind at follow_distance
+    LEAD   = 'lead'      # move ahead of person by lead_distance
+    SIDE   = 'side'      # stay to the right (or left) at side_offset
+    ORBIT  = 'orbit'     # circle around person at orbit_radius
+    LOOSE  = 'loose'     # general follow, larger tolerance
+    TIGHT  = 'tight'     # close follow, small tolerance
+
+
+def compute_shadow_target(person_pos, person_bearing_deg, follow_dist=1.2):
+    """Target position: behind person along their movement direction."""
+    bearing_rad = math.radians(person_bearing_deg + 180.0)
+    tx = person_pos[0] + follow_dist * math.sin(bearing_rad)
+    ty = person_pos[1] + follow_dist * math.cos(bearing_rad)
+    return (tx, ty, person_pos[2])
+
+
+def compute_lead_target(person_pos, person_bearing_deg, lead_dist=2.0):
+    """Target position: ahead of person."""
+    bearing_rad = math.radians(person_bearing_deg)
+    tx = person_pos[0] + lead_dist * math.sin(bearing_rad)
+    ty = person_pos[1] + lead_dist * math.cos(bearing_rad)
+    return (tx, ty, person_pos[2])
+
+
+def compute_side_target(person_pos, person_bearing_deg, side_dist=1.0, right=True):
+    """Target position: to the right (or left) of person."""
+    sign = 1.0 if right else -1.0
+    bearing_rad = math.radians(person_bearing_deg + sign * 90.0)
+    tx = person_pos[0] + side_dist * math.sin(bearing_rad)
+    ty = person_pos[1] + side_dist * math.cos(bearing_rad)
+    return (tx, ty, person_pos[2])
+
+
+def compute_orbit_target(person_pos, orbit_angle_deg, orbit_radius=1.5):
+    """Target on circle of radius orbit_radius around person."""
+    angle_rad = math.radians(orbit_angle_deg)
+    tx = person_pos[0] + orbit_radius * math.sin(angle_rad)
+    ty = person_pos[1] + orbit_radius * math.cos(angle_rad)
+    return (tx, ty, person_pos[2])
+
+
+def compute_loose_target(person_pos, robot_pos, follow_dist=2.0, tolerance=0.8):
+    """Only move if farther than follow_dist + tolerance."""
+    dx = person_pos[0] - robot_pos[0]
+    dy = person_pos[1] - robot_pos[1]
+    dist = math.hypot(dx, dy)
+    if dist <= follow_dist + tolerance:
+        return robot_pos
+    # Target at follow_dist behind person (toward robot)
+    scale = (dist - follow_dist) / dist
+    return (robot_pos[0] + dx * scale, robot_pos[1] + dy * scale, person_pos[2])
+
+
+def compute_tight_target(person_pos, follow_dist=0.6):
+    """Close follow: stay very near person."""
+    return (person_pos[0], person_pos[1] - follow_dist, person_pos[2])
+
+
+MODE_VOICE_COMMANDS = {
+    'shadow': FollowMode.SHADOW,
+    'follow me': FollowMode.SHADOW,
+    'behind me': FollowMode.SHADOW,
+    'lead': FollowMode.LEAD,
+    'go ahead': FollowMode.LEAD,
+    'lead me': FollowMode.LEAD,
+    'side': FollowMode.SIDE,
+    'stay beside': FollowMode.SIDE,
+    'orbit': FollowMode.ORBIT,
+    'circle me': FollowMode.ORBIT,
+    'loose': FollowMode.LOOSE,
+    'give me space': FollowMode.LOOSE,
+    'tight': FollowMode.TIGHT,
+    'stay close': FollowMode.TIGHT,
+}

jetson/ros2_ws/src/saltybot_social_nav/saltybot_social_nav/midas_depth_node.py

@@ -0,0 +1,231 @@
+"""
+midas_depth_node.py -- MiDaS monocular depth estimation for saltybot.
+
+Uses MiDaS_small via ONNX Runtime or TensorRT FP16.
+Provides relative depth estimates for cameras without active depth (CSI cameras).
+
+Publishes /social/depth/cam{i}/image  (sensor_msgs/Image, float32, relative depth)
+"""
+
+import os
+import numpy as np
+import rclpy
+from rclpy.node import Node
+from rclpy.qos import QoSProfile, ReliabilityPolicy, HistoryPolicy
+from sensor_msgs.msg import Image
+from cv_bridge import CvBridge
+
+# MiDaS_small input size
+_MIDAS_H = 256
+_MIDAS_W = 256
+# ImageNet normalization
+_MEAN = np.array([0.485, 0.456, 0.406], dtype=np.float32)
+_STD = np.array([0.229, 0.224, 0.225], dtype=np.float32)
+
+
+class _TRTBackend:
+    """TensorRT inference backend for MiDaS."""
+
+    def __init__(self, engine_path: str, logger):
+        self._logger = logger
+        try:
+            import tensorrt as trt
+            import pycuda.driver as cuda
+            import pycuda.autoinit  # noqa: F401
+
+            self._cuda = cuda
+            rt_logger = trt.Logger(trt.Logger.WARNING)
+            with open(engine_path, 'rb') as f:
+                engine = trt.Runtime(rt_logger).deserialize_cuda_engine(f.read())
+            self._context = engine.create_execution_context()
+
+            # Allocate buffers
+            self._d_input = cuda.mem_alloc(1 * 3 * _MIDAS_H * _MIDAS_W * 4)
+            self._d_output = cuda.mem_alloc(1 * _MIDAS_H * _MIDAS_W * 4)
+            self._h_output = np.empty((_MIDAS_H, _MIDAS_W), dtype=np.float32)
+            self._stream = cuda.Stream()
+            self._logger.info(f'TRT engine loaded: {engine_path}')
+        except Exception as e:
+            raise RuntimeError(f'TRT init failed: {e}')
+
+    def infer(self, input_tensor: np.ndarray) -> np.ndarray:
+        self._cuda.memcpy_htod_async(
+            self._d_input, input_tensor.ravel(), self._stream)
+        self._context.execute_async_v2(
+            bindings=[int(self._d_input), int(self._d_output)],
+            stream_handle=self._stream.handle)
+        self._cuda.memcpy_dtoh_async(
+            self._h_output, self._d_output, self._stream)
+        self._stream.synchronize()
+        return self._h_output.copy()
+
+
+class _ONNXBackend:
+    """ONNX Runtime inference backend for MiDaS."""
+
+    def __init__(self, onnx_path: str, logger):
+        self._logger = logger
+        try:
+            import onnxruntime as ort
+            providers = ['CUDAExecutionProvider', 'CPUExecutionProvider']
+            self._session = ort.InferenceSession(onnx_path, providers=providers)
+            self._input_name = self._session.get_inputs()[0].name
+            self._logger.info(f'ONNX model loaded: {onnx_path}')
+        except Exception as e:
+            raise RuntimeError(f'ONNX init failed: {e}')
+
+    def infer(self, input_tensor: np.ndarray) -> np.ndarray:
+        result = self._session.run(None, {self._input_name: input_tensor})
+        return result[0].squeeze()
+
+
+class MiDaSDepthNode(Node):
+    """MiDaS monocular depth estimation node."""
+
+    def __init__(self):
+        super().__init__('midas_depth')
+
+        # Parameters
+        self.declare_parameter('onnx_path',
+                               '/mnt/nvme/saltybot/models/midas_small.onnx')
+        self.declare_parameter('engine_path',
+                               '/mnt/nvme/saltybot/models/midas_small.engine')
+        self.declare_parameter('camera_topics', [
+            '/surround/cam0/image_raw',
+            '/surround/cam1/image_raw',
+            '/surround/cam2/image_raw',
+            '/surround/cam3/image_raw',
+        ])
+        self.declare_parameter('output_scale', 1.0)
+        self.declare_parameter('process_rate', 5.0)
+
+        onnx_path = self.get_parameter('onnx_path').value
+        engine_path = self.get_parameter('engine_path').value
+        self._camera_topics = self.get_parameter('camera_topics').value
+        self._output_scale = self.get_parameter('output_scale').value
+        process_rate = self.get_parameter('process_rate').value
+
+        # Initialize inference backend (TRT preferred, ONNX fallback)
+        self._backend = None
+        if os.path.exists(engine_path):
+            try:
+                self._backend = _TRTBackend(engine_path, self.get_logger())
+            except RuntimeError:
+                self.get_logger().warn('TRT failed, trying ONNX fallback')
+        if self._backend is None and os.path.exists(onnx_path):
+            try:
+                self._backend = _ONNXBackend(onnx_path, self.get_logger())
+            except RuntimeError:
+                self.get_logger().error('Both TRT and ONNX backends failed')
+        if self._backend is None:
+            self.get_logger().error(
+                'No MiDaS model found. Depth estimation disabled.')
+
+        self._bridge = CvBridge()
+
+        # Latest frames per camera (round-robin processing)
+        self._latest_frames = [None] * len(self._camera_topics)
+        self._current_cam_idx = 0
+
+        # QoS for camera subscriptions
+        cam_qos = QoSProfile(
+            reliability=ReliabilityPolicy.BEST_EFFORT,
+            history=HistoryPolicy.KEEP_LAST,
+            depth=1,
+        )
+
+        # Subscribe to each camera topic
+        self._cam_subs = []
+        for i, topic in enumerate(self._camera_topics):
+            sub = self.create_subscription(
+                Image, topic,
+                lambda msg, idx=i: self._on_image(msg, idx),
+                cam_qos)
+            self._cam_subs.append(sub)
+
+        # Publishers: one per camera
+        self._depth_pubs = []
+        for i in range(len(self._camera_topics)):
+            pub = self.create_publisher(
+                Image, f'/social/depth/cam{i}/image', 10)
+            self._depth_pubs.append(pub)
+
+        # Timer: round-robin across cameras
+        timer_period = 1.0 / process_rate
+        self._timer = self.create_timer(timer_period, self._timer_callback)
+
+        self.get_logger().info(
+            f'MiDaS depth node started: {len(self._camera_topics)} cameras '
+            f'@ {process_rate} Hz')
+
+    def _on_image(self, msg: Image, cam_idx: int):
+        """Cache latest frame for each camera."""
+        self._latest_frames[cam_idx] = msg
+
+    def _preprocess(self, bgr: np.ndarray) -> np.ndarray:
+        """Preprocess BGR image to MiDaS input tensor [1,3,256,256]."""
+        import cv2
+        rgb = cv2.cvtColor(bgr, cv2.COLOR_BGR2RGB)
+        resized = cv2.resize(rgb, (_MIDAS_W, _MIDAS_H),
+                             interpolation=cv2.INTER_LINEAR)
+        normalized = (resized.astype(np.float32) / 255.0 - _MEAN) / _STD
+        # HWC -> CHW -> NCHW
+        tensor = normalized.transpose(2, 0, 1)[np.newaxis, ...]
+        return tensor.astype(np.float32)
+
+    def _infer(self, tensor: np.ndarray) -> np.ndarray:
+        """Run inference, returns [256,256] float32 relative inverse depth."""
+        if self._backend is None:
+            return np.zeros((_MIDAS_H, _MIDAS_W), dtype=np.float32)
+        return self._backend.infer(tensor)
+
+    def _postprocess(self, raw: np.ndarray, orig_shape: tuple) -> np.ndarray:
+        """Resize depth back to original image shape, apply output_scale."""
+        import cv2
+        h, w = orig_shape[:2]
+        depth = cv2.resize(raw, (w, h), interpolation=cv2.INTER_LINEAR)
+        depth = depth * self._output_scale
+        return depth
+
+    def _timer_callback(self):
+        """Process one camera per tick (round-robin)."""
+        if not self._camera_topics:
+            return
+
+        idx = self._current_cam_idx
+        self._current_cam_idx = (idx + 1) % len(self._camera_topics)
+
+        msg = self._latest_frames[idx]
+        if msg is None:
+            return
+
+        try:
+            bgr = self._bridge.imgmsg_to_cv2(msg, desired_encoding='bgr8')
+        except Exception as e:
+            self.get_logger().warn(f'cv_bridge error cam{idx}: {e}')
+            return
+
+        tensor = self._preprocess(bgr)
+        raw_depth = self._infer(tensor)
+        depth_map = self._postprocess(raw_depth, bgr.shape)
+
+        # Publish as float32 Image
+        depth_msg = self._bridge.cv2_to_imgmsg(depth_map, encoding='32FC1')
+        depth_msg.header = msg.header
+        self._depth_pubs[idx].publish(depth_msg)
+
+
+def main(args=None):
+    rclpy.init(args=args)
+    node = MiDaSDepthNode()
+    try:
+        rclpy.spin(node)
+    except KeyboardInterrupt:
+        pass
+    finally:
+        node.destroy_node()
+        rclpy.shutdown()
+
+
+if __name__ == '__main__':
+    main()

jetson/ros2_ws/src/saltybot_social_nav/saltybot_social_nav/social_nav_node.py

@@ -0,0 +1,584 @@
+"""
+social_nav_node.py -- Social navigation node for saltybot.
+
+Orchestrates person following with multiple modes, voice commands,
+waypoint teaching/replay, and A* obstacle avoidance.
+
+Follow modes:
+  shadow  -- stay directly behind at follow_distance
+  lead    -- move ahead of person
+  side    -- stay beside (default right)
+  orbit   -- circle around person
+  loose   -- relaxed follow with deadband
+  tight   -- close follow
+
+Waypoint teaching:
+  Voice command "teach route <name>" -> record mode ON
+  Voice command "stop teaching" -> save route
+  Voice command "replay route <name>" -> playback
+
+Voice commands:
+  "follow me" / "shadow" -> SHADOW mode
+  "lead me" / "go ahead" -> LEAD mode
+  "stay beside" -> SIDE mode
+  "orbit" -> ORBIT mode
+  "give me space" -> LOOSE mode
+  "stay close" -> TIGHT mode
+  "stop" / "halt" -> STOP
+  "go home" -> navigate to home waypoint
+  "teach route <name>" -> start recording
+  "stop teaching" -> finish recording
+  "replay route <name>" -> playback recorded route
+"""
+
+import math
+import time
+import re
+from collections import deque
+
+import numpy as np
+import rclpy
+from rclpy.node import Node
+from rclpy.qos import QoSProfile, ReliabilityPolicy, HistoryPolicy
+from geometry_msgs.msg import Twist, PoseStamped
+from nav_msgs.msg import OccupancyGrid, Odometry
+from std_msgs.msg import String, Int32
+
+from .follow_modes import (
+    FollowMode, MODE_VOICE_COMMANDS,
+    compute_shadow_target, compute_lead_target, compute_side_target,
+    compute_orbit_target, compute_loose_target, compute_tight_target,
+)
+from .astar import astar, inflate_obstacles
+from .waypoint_teacher import WaypointRoute, WaypointReplayer
+
+# Try importing social msgs; fallback gracefully
+try:
+    from saltybot_social_msgs.msg import PersonStateArray
+    _HAS_SOCIAL_MSGS = True
+except ImportError:
+    _HAS_SOCIAL_MSGS = False
+
+# Proportional controller gains
+_K_ANG = 2.0   # angular gain
+_K_LIN = 0.8   # linear gain
+_HIGH_SPEED_THRESHOLD = 3.0  # m/s person velocity triggers fast mode
+_PREDICT_AHEAD_S = 0.3       # seconds to extrapolate position
+_TEACH_MIN_DIST = 0.5        # meters between recorded waypoints
+
+
+class SocialNavNode(Node):
+    """Main social navigation orchestrator."""
+
+    def __init__(self):
+        super().__init__('social_nav')
+
+        # -- Parameters --
+        self.declare_parameter('follow_mode', 'shadow')
+        self.declare_parameter('follow_distance', 1.2)
+        self.declare_parameter('lead_distance', 2.0)
+        self.declare_parameter('orbit_radius', 1.5)
+        self.declare_parameter('max_linear_speed', 1.0)
+        self.declare_parameter('max_linear_speed_fast', 5.5)
+        self.declare_parameter('max_angular_speed', 1.0)
+        self.declare_parameter('goal_tolerance', 0.3)
+        self.declare_parameter('routes_dir', '/mnt/nvme/saltybot/routes')
+        self.declare_parameter('home_x', 0.0)
+        self.declare_parameter('home_y', 0.0)
+        self.declare_parameter('map_resolution', 0.05)
+        self.declare_parameter('obstacle_inflation_cells', 3)
+
+        self._follow_mode = FollowMode(
+            self.get_parameter('follow_mode').value)
+        self._follow_distance = self.get_parameter('follow_distance').value
+        self._lead_distance = self.get_parameter('lead_distance').value
+        self._orbit_radius = self.get_parameter('orbit_radius').value
+        self._max_lin = self.get_parameter('max_linear_speed').value
+        self._max_lin_fast = self.get_parameter('max_linear_speed_fast').value
+        self._max_ang = self.get_parameter('max_angular_speed').value
+        self._goal_tol = self.get_parameter('goal_tolerance').value
+        self._routes_dir = self.get_parameter('routes_dir').value
+        self._home_x = self.get_parameter('home_x').value
+        self._home_y = self.get_parameter('home_y').value
+        self._map_resolution = self.get_parameter('map_resolution').value
+        self._inflation_cells = self.get_parameter(
+            'obstacle_inflation_cells').value
+
+        # -- State --
+        self._robot_x = 0.0
+        self._robot_y = 0.0
+        self._robot_yaw = 0.0
+        self._target_person_pos = None   # (x, y, z)
+        self._target_person_bearing = 0.0
+        self._target_person_id = -1
+        self._person_history = deque(maxlen=5)  # for velocity estimation
+        self._stopped = False
+        self._go_home = False
+
+        # Occupancy grid for A*
+        self._occ_grid = None
+        self._occ_origin = (0.0, 0.0)
+        self._occ_resolution = 0.05
+        self._astar_path = None
+
+        # Orbit state
+        self._orbit_angle = 0.0
+
+        # Waypoint teaching / replay
+        self._teaching = False
+        self._current_route = None
+        self._last_teach_x = None
+        self._last_teach_y = None
+        self._replayer = None
+
+        # -- QoS profiles --
+        best_effort_qos = QoSProfile(
+            reliability=ReliabilityPolicy.BEST_EFFORT,
+            history=HistoryPolicy.KEEP_LAST, depth=1)
+        reliable_qos = QoSProfile(
+            reliability=ReliabilityPolicy.RELIABLE,
+            history=HistoryPolicy.KEEP_LAST, depth=1)
+
+        # -- Subscriptions --
+        if _HAS_SOCIAL_MSGS:
+            self.create_subscription(
+                PersonStateArray, '/social/persons',
+                self._on_persons, best_effort_qos)
+        else:
+            self.get_logger().warn(
+                'saltybot_social_msgs not found; '
+                'using /person/target fallback')
+
+        self.create_subscription(
+            PoseStamped, '/person/target',
+            self._on_person_target, best_effort_qos)
+        self.create_subscription(
+            String, '/social/speech/command',
+            self._on_voice_command, 10)
+        self.create_subscription(
+            String, '/social/speech/transcript',
+            self._on_transcript, 10)
+        self.create_subscription(
+            OccupancyGrid, '/map',
+            self._on_map, reliable_qos)
+        self.create_subscription(
+            Odometry, '/odom',
+            self._on_odom, best_effort_qos)
+        self.create_subscription(
+            Int32, '/social/attention/target_id',
+            self._on_target_id, 10)
+
+        # -- Publishers --
+        self._cmd_vel_pub = self.create_publisher(
+            Twist, '/cmd_vel', best_effort_qos)
+        self._mode_pub = self.create_publisher(
+            String, '/social/nav/mode', reliable_qos)
+        self._target_pub = self.create_publisher(
+            PoseStamped, '/social/nav/target_pos', 10)
+        self._status_pub = self.create_publisher(
+            String, '/social/nav/status', best_effort_qos)
+
+        # -- Main loop timer (20 Hz) --
+        self._timer = self.create_timer(0.05, self._control_loop)
+
+        self.get_logger().info(
+            f'Social nav started: mode={self._follow_mode.value}, '
+            f'dist={self._follow_distance}m')
+
+    # ----------------------------------------------------------------
+    # Subscriptions
+    # ----------------------------------------------------------------
+
+    def _on_persons(self, msg):
+        """Handle PersonStateArray from social perception."""
+        target_id = msg.primary_attention_id
+        if self._target_person_id >= 0:
+            target_id = self._target_person_id
+
+        for p in msg.persons:
+            if p.person_id == target_id or (
+                    target_id < 0 and len(msg.persons) > 0):
+                pos = (p.position.x, p.position.y, p.position.z)
+                self._update_person_position(pos, p.bearing_deg)
+                break
+
+    def _on_person_target(self, msg: PoseStamped):
+        """Fallback: single person target pose."""
+        pos = (msg.pose.position.x, msg.pose.position.y,
+               msg.pose.position.z)
+        # Estimate bearing from quaternion yaw
+        q = msg.pose.orientation
+        yaw = math.atan2(2.0 * (q.w * q.z + q.x * q.y),
+                         1.0 - 2.0 * (q.y * q.y + q.z * q.z))
+        self._update_person_position(pos, math.degrees(yaw))
+
+    def _update_person_position(self, pos, bearing_deg):
+        """Update person tracking state and record history."""
+        now = time.time()
+        self._target_person_pos = pos
+        self._target_person_bearing = bearing_deg
+        self._person_history.append((now, pos[0], pos[1]))
+
+    def _on_odom(self, msg: Odometry):
+        """Update robot pose from odometry."""
+        self._robot_x = msg.pose.pose.position.x
+        self._robot_y = msg.pose.pose.position.y
+        q = msg.pose.pose.orientation
+        self._robot_yaw = math.atan2(
+            2.0 * (q.w * q.z + q.x * q.y),
+            1.0 - 2.0 * (q.y * q.y + q.z * q.z))
+
+    def _on_map(self, msg: OccupancyGrid):
+        """Cache occupancy grid for A* planning."""
+        w, h = msg.info.width, msg.info.height
+        data = np.array(msg.data, dtype=np.int8).reshape((h, w))
+        # Convert -1 (unknown) to free (0) for planning
+        data[data < 0] = 0
+        self._occ_grid = data.astype(np.int32)
+        self._occ_origin = (msg.info.origin.position.x,
+                            msg.info.origin.position.y)
+        self._occ_resolution = msg.info.resolution
+
+    def _on_target_id(self, msg: Int32):
+        """Switch target person."""
+        self._target_person_id = msg.data
+        self.get_logger().info(f'Target person ID set to {msg.data}')
+
+    def _on_voice_command(self, msg: String):
+        """Handle discrete voice commands for mode switching."""
+        cmd = msg.data.strip().lower()
+
+        if cmd in ('stop', 'halt'):
+            self._stopped = True
+            self._replayer = None
+            self._publish_status('STOPPED')
+            return
+
+        if cmd in ('resume', 'go', 'start'):
+            self._stopped = False
+            self._publish_status('RESUMED')
+            return
+
+        matched = MODE_VOICE_COMMANDS.get(cmd)
+        if matched:
+            self._follow_mode = matched
+            self._stopped = False
+            mode_msg = String()
+            mode_msg.data = self._follow_mode.value
+            self._mode_pub.publish(mode_msg)
+            self._publish_status(f'MODE: {self._follow_mode.value}')
+
+    def _on_transcript(self, msg: String):
+        """Handle free-form voice transcripts for route teaching."""
+        text = msg.data.strip().lower()
+
+        # "teach route <name>"
+        m = re.match(r'teach\s+route\s+(\w+)', text)
+        if m:
+            name = m.group(1)
+            self._teaching = True
+            self._current_route = WaypointRoute(name)
+            self._last_teach_x = self._robot_x
+            self._last_teach_y = self._robot_y
+            self._publish_status(f'TEACHING: {name}')
+            self.get_logger().info(f'Recording route: {name}')
+            return
+
+        # "stop teaching"
+        if 'stop teaching' in text:
+            if self._teaching and self._current_route:
+                self._current_route.save(self._routes_dir)
+                self._publish_status(
+                    f'SAVED: {self._current_route.name} '
+                    f'({len(self._current_route.waypoints)} pts)')
+                self.get_logger().info(
+                    f'Route saved: {self._current_route.name}')
+            self._teaching = False
+            self._current_route = None
+            return
+
+        # "replay route <name>"
+        m = re.match(r'replay\s+route\s+(\w+)', text)
+        if m:
+            name = m.group(1)
+            try:
+                route = WaypointRoute.load(self._routes_dir, name)
+                self._replayer = WaypointReplayer(route)
+                self._stopped = False
+                self._publish_status(f'REPLAY: {name}')
+                self.get_logger().info(f'Replaying route: {name}')
+            except FileNotFoundError:
+                self._publish_status(f'ROUTE NOT FOUND: {name}')
+            return
+
+        # "go home"
+        if 'go home' in text:
+            self._go_home = True
+            self._stopped = False
+            self._publish_status('GO HOME')
+            return
+
+        # Also try mode commands from transcript
+        for phrase, mode in MODE_VOICE_COMMANDS.items():
+            if phrase in text:
+                self._follow_mode = mode
+                self._stopped = False
+                mode_msg = String()
+                mode_msg.data = self._follow_mode.value
+                self._mode_pub.publish(mode_msg)
+                self._publish_status(f'MODE: {self._follow_mode.value}')
+                return
+
+    # ----------------------------------------------------------------
+    # Control loop
+    # ----------------------------------------------------------------
+
+    def _control_loop(self):
+        """Main 20Hz control loop."""
+        # Record waypoint if teaching
+        if self._teaching and self._current_route:
+            self._maybe_record_waypoint()
+
+        # Publish zero velocity if stopped
+        if self._stopped:
+            self._publish_cmd_vel(0.0, 0.0)
+            return
+
+        # Determine navigation target
+        target = self._get_nav_target()
+        if target is None:
+            self._publish_cmd_vel(0.0, 0.0)
+            return
+
+        tx, ty, tz = target
+
+        # Publish debug target
+        self._publish_target_pose(tx, ty, tz)
+
+        # Check if arrived
+        dist_to_target = math.hypot(tx - self._robot_x, ty - self._robot_y)
+        if dist_to_target < self._goal_tol:
+            self._publish_cmd_vel(0.0, 0.0)
+            if self._go_home:
+                self._go_home = False
+                self._publish_status('HOME REACHED')
+            return
+
+        # Try A* path if map available
+        if self._occ_grid is not None:
+            path_target = self._plan_astar(tx, ty)
+            if path_target:
+                tx, ty = path_target
+
+        # Determine speed limit
+        max_lin = self._max_lin
+        person_vel = self._estimate_person_velocity()
+        if person_vel > _HIGH_SPEED_THRESHOLD:
+            max_lin = self._max_lin_fast
+
+        # Compute and publish cmd_vel
+        lin, ang = self._compute_cmd_vel(
+            self._robot_x, self._robot_y, self._robot_yaw,
+            tx, ty, max_lin)
+        self._publish_cmd_vel(lin, ang)
+
+    def _get_nav_target(self):
+        """Determine current navigation target based on mode/state."""
+        # Route replay takes priority
+        if self._replayer and not self._replayer.is_done:
+            self._replayer.check_arrived(self._robot_x, self._robot_y)
+            wp = self._replayer.current_waypoint()
+            if wp:
+                return (wp.x, wp.y, wp.z)
+            else:
+                self._replayer = None
+                self._publish_status('REPLAY DONE')
+                return None
+
+        # Go home
+        if self._go_home:
+            return (self._home_x, self._home_y, 0.0)
+
+        # Person following
+        if self._target_person_pos is None:
+            return None
+
+        # Predict person position ahead for high-speed tracking
+        px, py, pz = self._predict_person_position()
+        bearing = self._target_person_bearing
+        robot_pos = (self._robot_x, self._robot_y, 0.0)
+
+        if self._follow_mode == FollowMode.SHADOW:
+            return compute_shadow_target(
+                (px, py, pz), bearing, self._follow_distance)
+        elif self._follow_mode == FollowMode.LEAD:
+            return compute_lead_target(
+                (px, py, pz), bearing, self._lead_distance)
+        elif self._follow_mode == FollowMode.SIDE:
+            return compute_side_target(
+                (px, py, pz), bearing, self._follow_distance)
+        elif self._follow_mode == FollowMode.ORBIT:
+            self._orbit_angle = (self._orbit_angle + 1.0) % 360.0
+            return compute_orbit_target(
+                (px, py, pz), self._orbit_angle, self._orbit_radius)
+        elif self._follow_mode == FollowMode.LOOSE:
+            return compute_loose_target(
+                (px, py, pz), robot_pos, self._follow_distance)
+        elif self._follow_mode == FollowMode.TIGHT:
+            return compute_tight_target(
+                (px, py, pz), self._follow_distance)
+
+        return (px, py, pz)
+
+    def _predict_person_position(self):
+        """Extrapolate person position using velocity from recent samples."""
+        if self._target_person_pos is None:
+            return (0.0, 0.0, 0.0)
+
+        px, py, pz = self._target_person_pos
+
+        if len(self._person_history) >= 3:
+            # Use last 3 samples for velocity estimation
+            t0, x0, y0 = self._person_history[-3]
+            t1, x1, y1 = self._person_history[-1]
+            dt = t1 - t0
+            if dt > 0.01:
+                vx = (x1 - x0) / dt
+                vy = (y1 - y0) / dt
+                speed = math.hypot(vx, vy)
+                if speed > _HIGH_SPEED_THRESHOLD:
+                    px += vx * _PREDICT_AHEAD_S
+                    py += vy * _PREDICT_AHEAD_S
+
+        return (px, py, pz)
+
+    def _estimate_person_velocity(self) -> float:
+        """Estimate person speed from recent position history."""
+        if len(self._person_history) < 2:
+            return 0.0
+        t0, x0, y0 = self._person_history[-2]
+        t1, x1, y1 = self._person_history[-1]
+        dt = t1 - t0
+        if dt < 0.01:
+            return 0.0
+        return math.hypot(x1 - x0, y1 - y0) / dt
+
+    def _plan_astar(self, target_x, target_y):
+        """Run A* on occupancy grid, return next waypoint in world coords."""
+        grid = self._occ_grid
+        res = self._occ_resolution
+        ox, oy = self._occ_origin
+
+        # World to grid
+        def w2g(wx, wy):
+            return (int((wy - oy) / res), int((wx - ox) / res))
+
+        start = w2g(self._robot_x, self._robot_y)
+        goal = w2g(target_x, target_y)
+
+        rows, cols = grid.shape
+        if not (0 <= start[0] < rows and 0 <= start[1] < cols):
+            return None
+        if not (0 <= goal[0] < rows and 0 <= goal[1] < cols):
+            return None
+
+        inflated = inflate_obstacles(grid, self._inflation_cells)
+        path = astar(inflated, start, goal)
+
+        if path and len(path) > 1:
+            # Follow a lookahead point (3 steps ahead or end)
+            lookahead_idx = min(3, len(path) - 1)
+            r, c = path[lookahead_idx]
+            wx = ox + c * res + res / 2.0
+            wy = oy + r * res + res / 2.0
+            self._astar_path = path
+            return (wx, wy)
+
+        return None
+
+    def _compute_cmd_vel(self, rx, ry, ryaw, tx, ty, max_lin):
+        """Proportional controller: compute linear and angular velocity."""
+        dx = tx - rx
+        dy = ty - ry
+        dist = math.hypot(dx, dy)
+        angle_to_target = math.atan2(dy, dx)
+        angle_error = angle_to_target - ryaw
+
+        # Normalize angle error to [-pi, pi]
+        while angle_error > math.pi:
+            angle_error -= 2.0 * math.pi
+        while angle_error < -math.pi:
+            angle_error += 2.0 * math.pi
+
+        angular_vel = _K_ANG * angle_error
+        angular_vel = max(-self._max_ang,
+                          min(self._max_ang, angular_vel))
+
+        # Reduce linear speed when turning hard
+        angle_factor = max(0.0, 1.0 - abs(angle_error) / (math.pi / 2.0))
+        linear_vel = _K_LIN * dist * angle_factor
+        linear_vel = max(0.0, min(max_lin, linear_vel))
+
+        return (linear_vel, angular_vel)
+
+    # ----------------------------------------------------------------
+    # Waypoint teaching
+    # ----------------------------------------------------------------
+
+    def _maybe_record_waypoint(self):
+        """Record waypoint if robot moved > _TEACH_MIN_DIST."""
+        if self._last_teach_x is None:
+            self._last_teach_x = self._robot_x
+            self._last_teach_y = self._robot_y
+
+        dist = math.hypot(
+            self._robot_x - self._last_teach_x,
+            self._robot_y - self._last_teach_y)
+
+        if dist >= _TEACH_MIN_DIST:
+            yaw_deg = math.degrees(self._robot_yaw)
+            self._current_route.add(
+                self._robot_x, self._robot_y, 0.0, yaw_deg)
+            self._last_teach_x = self._robot_x
+            self._last_teach_y = self._robot_y
+
+    # ----------------------------------------------------------------
+    # Publishers
+    # ----------------------------------------------------------------
+
+    def _publish_cmd_vel(self, linear: float, angular: float):
+        twist = Twist()
+        twist.linear.x = linear
+        twist.angular.z = angular
+        self._cmd_vel_pub.publish(twist)
+
+    def _publish_target_pose(self, x, y, z):
+        msg = PoseStamped()
+        msg.header.stamp = self.get_clock().now().to_msg()
+        msg.header.frame_id = 'map'
+        msg.pose.position.x = x
+        msg.pose.position.y = y
+        msg.pose.position.z = z
+        self._target_pub.publish(msg)
+
+    def _publish_status(self, status: str):
+        msg = String()
+        msg.data = status
+        self._status_pub.publish(msg)
+        self.get_logger().info(f'Nav status: {status}')
+
+
+def main(args=None):
+    rclpy.init(args=args)
+    node = SocialNavNode()
+    try:
+        rclpy.spin(node)
+    except KeyboardInterrupt:
+        pass
+    finally:
+        node.destroy_node()
+        rclpy.shutdown()
+
+
+if __name__ == '__main__':
+    main()

jetson/ros2_ws/src/saltybot_social_nav/saltybot_social_nav/waypoint_teacher.py

@@ -0,0 +1,91 @@
+"""waypoint_teacher.py -- Record and replay waypoint routes."""
+
+import json
+import time
+import math
+from pathlib import Path
+from dataclasses import dataclass, asdict
+
+
+@dataclass
+class Waypoint:
+    x: float
+    y: float
+    z: float
+    yaw_deg: float
+    timestamp: float
+    label: str = ''
+
+
+class WaypointRoute:
+    """A named sequence of waypoints."""
+
+    def __init__(self, name: str):
+        self.name = name
+        self.waypoints: list[Waypoint] = []
+        self.created_at = time.time()
+
+    def add(self, x, y, z, yaw_deg, label=''):
+        self.waypoints.append(Waypoint(x, y, z, yaw_deg, time.time(), label))
+
+    def to_dict(self):
+        return {
+            'name': self.name,
+            'created_at': self.created_at,
+            'waypoints': [asdict(w) for w in self.waypoints],
+        }
+
+    @classmethod
+    def from_dict(cls, d):
+        route = cls(d['name'])
+        route.created_at = d.get('created_at', 0)
+        route.waypoints = [Waypoint(**w) for w in d['waypoints']]
+        return route
+
+    def save(self, routes_dir: str):
+        Path(routes_dir).mkdir(parents=True, exist_ok=True)
+        path = Path(routes_dir) / f'{self.name}.json'
+        path.write_text(json.dumps(self.to_dict(), indent=2))
+
+    @classmethod
+    def load(cls, routes_dir: str, name: str):
+        path = Path(routes_dir) / f'{name}.json'
+        return cls.from_dict(json.loads(path.read_text()))
+
+    @staticmethod
+    def list_routes(routes_dir: str) -> list[str]:
+        d = Path(routes_dir)
+        if not d.exists():
+            return []
+        return [p.stem for p in d.glob('*.json')]
+
+
+class WaypointReplayer:
+    """Iterates through waypoints, returning next target."""
+
+    def __init__(self, route: WaypointRoute, arrival_radius: float = 0.3):
+        self._route = route
+        self._idx = 0
+        self._arrival_radius = arrival_radius
+
+    def current_waypoint(self) -> Waypoint | None:
+        if self._idx < len(self._route.waypoints):
+            return self._route.waypoints[self._idx]
+        return None
+
+    def check_arrived(self, robot_x, robot_y) -> bool:
+        wp = self.current_waypoint()
+        if wp is None:
+            return False
+        dist = math.hypot(robot_x - wp.x, robot_y - wp.y)
+        if dist < self._arrival_radius:
+            self._idx += 1
+            return True
+        return False
+
+    @property
+    def is_done(self) -> bool:
+        return self._idx >= len(self._route.waypoints)
+
+    def reset(self):
+        self._idx = 0

jetson/ros2_ws/src/saltybot_social_nav/saltybot_social_nav/waypoint_teacher_node.py

@@ -0,0 +1,135 @@
+"""
+waypoint_teacher_node.py -- Standalone waypoint teacher ROS2 node.
+
+Listens to /social/speech/transcript for "teach route <name>" and "stop teaching".
+Records robot pose at configurable intervals. Saves/loads routes via WaypointRoute.
+"""
+
+import math
+
+import rclpy
+from rclpy.node import Node
+from rclpy.qos import QoSProfile, ReliabilityPolicy, HistoryPolicy
+from nav_msgs.msg import Odometry
+from std_msgs.msg import String
+
+from .waypoint_teacher import WaypointRoute
+
+
+class WaypointTeacherNode(Node):
+    """Standalone waypoint teaching node."""
+
+    def __init__(self):
+        super().__init__('waypoint_teacher')
+
+        self.declare_parameter('routes_dir', '/mnt/nvme/saltybot/routes')
+        self.declare_parameter('recording_interval', 0.5)  # meters
+
+        self._routes_dir = self.get_parameter('routes_dir').value
+        self._interval = self.get_parameter('recording_interval').value
+
+        self._teaching = False
+        self._route = None
+        self._last_x = None
+        self._last_y = None
+        self._robot_x = 0.0
+        self._robot_y = 0.0
+        self._robot_yaw = 0.0
+
+        best_effort_qos = QoSProfile(
+            reliability=ReliabilityPolicy.BEST_EFFORT,
+            history=HistoryPolicy.KEEP_LAST, depth=1)
+
+        self.create_subscription(
+            Odometry, '/odom', self._on_odom, best_effort_qos)
+        self.create_subscription(
+            String, '/social/speech/transcript',
+            self._on_transcript, 10)
+
+        self._status_pub = self.create_publisher(
+            String, '/social/waypoint/status', 10)
+
+        # Record timer at 10Hz (check distance)
+        self._timer = self.create_timer(0.1, self._record_tick)
+
+        self.get_logger().info(
+            f'Waypoint teacher ready (interval={self._interval}m, '
+            f'dir={self._routes_dir})')
+
+    def _on_odom(self, msg: Odometry):
+        self._robot_x = msg.pose.pose.position.x
+        self._robot_y = msg.pose.pose.position.y
+        q = msg.pose.pose.orientation
+        self._robot_yaw = math.atan2(
+            2.0 * (q.w * q.z + q.x * q.y),
+            1.0 - 2.0 * (q.y * q.y + q.z * q.z))
+
+    def _on_transcript(self, msg: String):
+        text = msg.data.strip().lower()
+
+        import re
+        m = re.match(r'teach\s+route\s+(\w+)', text)
+        if m:
+            name = m.group(1)
+            self._route = WaypointRoute(name)
+            self._teaching = True
+            self._last_x = self._robot_x
+            self._last_y = self._robot_y
+            self._pub_status(f'RECORDING: {name}')
+            self.get_logger().info(f'Recording route: {name}')
+            return
+
+        if 'stop teaching' in text:
+            if self._teaching and self._route:
+                self._route.save(self._routes_dir)
+                n = len(self._route.waypoints)
+                self._pub_status(
+                    f'SAVED: {self._route.name} ({n} waypoints)')
+                self.get_logger().info(
+                    f'Route saved: {self._route.name} ({n} pts)')
+            self._teaching = False
+            self._route = None
+            return
+
+        if 'list routes' in text:
+            routes = WaypointRoute.list_routes(self._routes_dir)
+            self._pub_status(f'ROUTES: {", ".join(routes) or "(none)"}')
+
+    def _record_tick(self):
+        if not self._teaching or self._route is None:
+            return
+
+        if self._last_x is None:
+            self._last_x = self._robot_x
+            self._last_y = self._robot_y
+
+        dist = math.hypot(
+            self._robot_x - self._last_x,
+            self._robot_y - self._last_y)
+
+        if dist >= self._interval:
+            yaw_deg = math.degrees(self._robot_yaw)
+            self._route.add(self._robot_x, self._robot_y, 0.0, yaw_deg)
+            self._last_x = self._robot_x
+            self._last_y = self._robot_y
+
+    def _pub_status(self, text: str):
+        msg = String()
+        msg.data = text
+        self._status_pub.publish(msg)
+
+
+def main(args=None):
+    rclpy.init(args=args)
+    node = WaypointTeacherNode()
+    try:
+        rclpy.spin(node)
+    except KeyboardInterrupt:
+        pass
+    finally:
+        node.destroy_node()
+        rclpy.shutdown()
+
+
+if __name__ == '__main__':
+    main()

jetson/ros2_ws/src/saltybot_social_nav/scripts/build_midas_trt_engine.py

@@ -0,0 +1,80 @@
+#!/usr/bin/env python3
+"""
+build_midas_trt_engine.py -- Build TensorRT FP16 engine for MiDaS_small from ONNX.
+
+Usage:
+    python3 build_midas_trt_engine.py \
+        --onnx /mnt/nvme/saltybot/models/midas_small.onnx \
+        --engine /mnt/nvme/saltybot/models/midas_small.engine \
+        --fp16
+
+Requires: tensorrt, pycuda
+"""
+
+import argparse
+import os
+import sys
+
+
+def build_engine(onnx_path: str, engine_path: str, fp16: bool = True):
+    try:
+        import tensorrt as trt
+    except ImportError:
+        print('ERROR: tensorrt not found. Install TensorRT first.')
+        sys.exit(1)
+
+    logger = trt.Logger(trt.Logger.INFO)
+    builder = trt.Builder(logger)
+    network = builder.create_network(
+        1 << int(trt.NetworkDefinitionCreationFlag.EXPLICIT_BATCH))
+    parser = trt.OnnxParser(network, logger)
+
+    print(f'Parsing ONNX model: {onnx_path}')
+    with open(onnx_path, 'rb') as f:
+        if not parser.parse(f.read()):
+            for i in range(parser.num_errors):
+                print(f'  ONNX parse error: {parser.get_error(i)}')
+            sys.exit(1)
+
+    config = builder.create_builder_config()
+    config.set_memory_pool_limit(trt.MemoryPoolType.WORKSPACE, 1 << 30)  # 1GB
+
+    if fp16 and builder.platform_has_fast_fp16:
+        config.set_flag(trt.BuilderFlag.FP16)
+        print('FP16 mode enabled')
+    elif fp16:
+        print('WARNING: FP16 not supported on this platform, using FP32')
+
+    print('Building TensorRT engine (this may take several minutes)...')
+    engine_bytes = builder.build_serialized_network(network, config)
+    if engine_bytes is None:
+        print('ERROR: Failed to build engine')
+        sys.exit(1)
+
+    os.makedirs(os.path.dirname(engine_path) or '.', exist_ok=True)
+    with open(engine_path, 'wb') as f:
+        f.write(engine_bytes)
+
+    size_mb = len(engine_bytes) / (1024 * 1024)
+    print(f'Engine saved: {engine_path} ({size_mb:.1f} MB)')
+
+
+def main():
+    parser = argparse.ArgumentParser(
+        description='Build TensorRT FP16 engine for MiDaS_small')
+    parser.add_argument('--onnx', required=True,
+                        help='Path to MiDaS ONNX model')
+    parser.add_argument('--engine', required=True,
+                        help='Output TRT engine path')
+    parser.add_argument('--fp16', action='store_true', default=True,
+                        help='Enable FP16 (default: True)')
+    parser.add_argument('--fp32', action='store_true',
+                        help='Force FP32 (disable FP16)')
+    args = parser.parse_args()
+
+    fp16 = not args.fp32
+    build_engine(args.onnx, args.engine, fp16=fp16)
+
+
+if __name__ == '__main__':
+    main()

jetson/ros2_ws/src/saltybot_social_nav/setup.cfg

@@ -0,0 +1,4 @@
+[develop]
+script_dir=$base/lib/saltybot_social_nav
+[install]
+install_scripts=$base/lib/saltybot_social_nav

jetson/ros2_ws/src/saltybot_social_nav/setup.py

@@ -0,0 +1,31 @@
+from setuptools import setup
+import os
+from glob import glob
+
+package_name = 'saltybot_social_nav'
+
+setup(
+    name=package_name,
+    version='0.1.0',
+    packages=[package_name],
+    data_files=[
+        ('share/ament_index/resource_index/packages', ['resource/' + package_name]),
+        ('share/' + package_name, ['package.xml']),
+        (os.path.join('share', package_name, 'launch'), glob('launch/*.py')),
+        (os.path.join('share', package_name, 'config'), glob('config/*.yaml')),
+    ],
+    install_requires=['setuptools'],
+    zip_safe=True,
+    maintainer='seb',
+    maintainer_email='seb@vayrette.com',
+    description='Social navigation for saltybot: follow modes, waypoint teaching, A* avoidance, MiDaS depth',
+    license='MIT',
+    tests_require=['pytest'],
+    entry_points={
+        'console_scripts': [
+            'social_nav = saltybot_social_nav.social_nav_node:main',
+            'midas_depth = saltybot_social_nav.midas_depth_node:main',
+            'waypoint_teacher = saltybot_social_nav.waypoint_teacher_node:main',
+        ],
+    },
+)

jetson/ros2_ws/src/saltybot_social_nav/test/test_copyright.py

@@ -0,0 +1,12 @@
+# Copyright 2026 SaltyLab
+# Licensed under MIT
+
+from ament_copyright.main import main
+import pytest
+
+
+@pytest.mark.copyright
+@pytest.mark.linter
+def test_copyright():
+    rc = main(argv=['.', 'test'])
+    assert rc == 0, 'Found errors'

jetson/ros2_ws/src/saltybot_social_nav/test/test_flake8.py

@@ -0,0 +1,14 @@
+# Copyright 2026 SaltyLab
+# Licensed under MIT
+
+from ament_flake8.main import main_with_errors
+import pytest
+
+
+@pytest.mark.flake8
+@pytest.mark.linter
+def test_flake8():
+    rc, errors = main_with_errors(argv=[])
+    assert rc == 0, \
+        'Found %d code style errors / warnings:\n' % len(errors) + \
+        '\n'.join(errors)

jetson/ros2_ws/src/saltybot_social_nav/test/test_pep257.py

@@ -0,0 +1,12 @@
+# Copyright 2026 SaltyLab
+# Licensed under MIT
+
+from ament_pep257.main import main
+import pytest
+
+
+@pytest.mark.pep257
+@pytest.mark.linter
+def test_pep257():
+    rc = main(argv=['.', 'test'])
+    assert rc == 0, 'Found code style errors / warnings'

jetson/ros2_ws/src/saltybot_uwb/config/uwb_config.yaml

@@ -1,57 +1,24 @@
-# uwb_config.yaml — MaUWB ESP32-S3 DW3000 UWB follow-me system
+# uwb_config.yaml — MaUWB ESP32-S3 DW3000 UWB integration (Issue #90)
-#
-# Hardware layout:
-#   Anchor-0 (port side)      → USB port_a, y = +anchor_separation/2
-#   Anchor-1 (starboard side) → USB port_b, y = -anchor_separation/2
-#   Tag on person             → belt clip, ~0.9m above ground
 #
 # Run with:
 #   ros2 launch saltybot_uwb uwb.launch.py
-# Override at launch:
-#   ros2 launch saltybot_uwb uwb.launch.py port_a:=/dev/ttyUSB2
 
-# ── Serial ports ──────────────────────────────────────────────────────────────
+port_a:    /dev/uwb-anchor0
-# Set udev rules to get stable symlinks:
+port_b:    /dev/uwb-anchor1
-#   /dev/uwb-anchor0  → port_a
+baudrate:  115200
-#   /dev/uwb-anchor1  → port_b
-# (See jetson/docs/pinout.md for udev setup)
-port_a:    /dev/uwb-anchor0      # Anchor-0 (port)
-port_b:    /dev/uwb-anchor1      # Anchor-1 (starboard)
-baudrate:  115200                 # MaUWB default — do not change
 
-# ── Anchor geometry ────────────────────────────────────────────────────────────
+anchor_separation: 0.25
-# anchor_separation: centre-to-centre distance between anchors (metres)
+anchor_height:     0.80
-#   Must match physical mounting. Larger = more accurate lateral resolution.
+tag_height:        0.90
-anchor_separation: 0.25          # metres (25cm)
 
-# anchor_height: height of anchors above ground (metres)
+range_timeout_s: 1.0
-#   Orin stem mount ≈ 0.80m on the saltybot platform
+max_range_m:     8.0
-anchor_height: 0.80              # metres
+min_range_m:     0.05
 
-# tag_height: height of person's belt-clip tag above ground (metres)
-tag_height: 0.90                 # metres (adjust per user)
-
-# ── Range validity ─────────────────────────────────────────────────────────────
-# range_timeout_s: stale anchor — excluded from triangulation after this gap
-range_timeout_s: 1.0             # seconds
-
-# max_range_m: discard ranges beyond this (DW3000 indoor practical limit ≈8m)
-max_range_m: 8.0                 # metres
-
-# min_range_m: discard ranges below this (likely multipath artefacts)
-min_range_m: 0.05                # metres
-
-# ── Kalman filter ──────────────────────────────────────────────────────────────
-# kf_process_noise: Q scalar — how dynamic the person's motion is
-#   Higher → faster response, more jitter
 kf_process_noise: 0.1
-
-# kf_meas_noise: R scalar — how noisy the UWB ranging is
-#   DW3000 indoor accuracy ≈ 10cm RMS → 0.1m → R ≈ 0.01
-#   Use 0.3 to be conservative on first deployment
 kf_meas_noise:    0.3
 
-# ── Publish rate ──────────────────────────────────────────────────────────────
+range_rate:   100.0
-# Should match or exceed the AT+RANGE? poll rate from both anchors.
+bearing_rate: 10.0
-# 20Hz means 50ms per cycle; each anchor query takes ~10ms → headroom ok.
+
-publish_rate: 20.0               # Hz
+enrolled_tag_ids: [""]

jetson/ros2_ws/src/saltybot_uwb/launch/uwb.launch.py

@@ -1,10 +1,14 @@
 """
-uwb.launch.py — Launch UWB driver node for MaUWB ESP32-S3 follow-me.
+uwb.launch.py — Launch UWB driver node for MaUWB ESP32-S3 DW3000 (Issue #90).
+
+Topics:
+  /uwb/ranges   100 Hz raw TWR ranges
+  /uwb/bearing   10 Hz Kalman-fused bearing
+  /uwb/target    10 Hz triangulated PoseStamped (backwards compat)
 
 Usage:
   ros2 launch saltybot_uwb uwb.launch.py
-  ros2 launch saltybot_uwb uwb.launch.py port_a:=/dev/ttyUSB2 port_b:=/dev/ttyUSB3
+  ros2 launch saltybot_uwb uwb.launch.py enrolled_tag_ids:="['0xDEADBEEF']"
-  ros2 launch saltybot_uwb uwb.launch.py anchor_separation:=0.30 publish_rate:=10.0
 """
 
 import os
@@ -31,7 +35,9 @@ def generate_launch_description():
         DeclareLaunchArgument("min_range_m",        default_value="0.05"),
         DeclareLaunchArgument("kf_process_noise",   default_value="0.1"),
         DeclareLaunchArgument("kf_meas_noise",      default_value="0.3"),
-        DeclareLaunchArgument("publish_rate",       default_value="20.0"),
+        DeclareLaunchArgument("range_rate",         default_value="100.0"),
+        DeclareLaunchArgument("bearing_rate",       default_value="10.0"),
+        DeclareLaunchArgument("enrolled_tag_ids",   default_value="['']"),
 
         Node(
             package="saltybot_uwb",
@@ -52,7 +58,9 @@ def generate_launch_description():
                     "min_range_m":       LaunchConfiguration("min_range_m"),
                     "kf_process_noise":  LaunchConfiguration("kf_process_noise"),
                     "kf_meas_noise":     LaunchConfiguration("kf_meas_noise"),
-                    "publish_rate":      LaunchConfiguration("publish_rate"),
+                    "range_rate":        LaunchConfiguration("range_rate"),
+                    "bearing_rate":      LaunchConfiguration("bearing_rate"),
+                    "enrolled_tag_ids":  LaunchConfiguration("enrolled_tag_ids"),
                 },
             ],
         ),

jetson/ros2_ws/src/saltybot_uwb/saltybot_uwb/ranging_math.py

@@ -29,6 +29,26 @@ Returns (x_t, y_t); caller should treat negative x_t as 0.
 import math
 
 
+# ── Bearing helper ────────────────────────────────────────────────────────────
+
+def bearing_from_pos(x: float, y: float) -> float:
+    """
+    Compute horizontal bearing to a point (x, y) in base_link.
+
+    Parameters
+    ----------
+    x : forward distance (metres)
+    y : lateral offset (metres, positive = left of robot / CCW)
+
+    Returns
+    -------
+    bearing_rad : bearing in radians, range -π .. +π
+                  positive = target to the left (CCW)
+                  0 = directly ahead
+    """
+    return math.atan2(y, x)
+
+
 # ── Triangulation ─────────────────────────────────────────────────────────────
 
 def triangulate_2anchor(

jetson/ros2_ws/src/saltybot_uwb/saltybot_uwb/uwb_driver_node.py

@@ -1,32 +1,52 @@
 """
-uwb_driver_node.py — ROS2 node for MaUWB ESP32-S3 DW3000 follow-me system.
+uwb_driver_node.py — ROS2 node for MaUWB ESP32-S3 DW3000 UWB integration.
 
 Hardware
 ────────
-  • 2× MaUWB ESP32-S3 DW3000 anchors on robot stem (USB → Orin Nano)
+  • 2× MaUWB ESP32-S3 DW3000 anchors on robot stem (USB → Orin)
     - Anchor-0: port side  (y = +sep/2)
     - Anchor-1: starboard  (y = -sep/2)
-  • 1× MaUWB tag on person (belt clip)
+  • 1× MaUWB tag per enrolled person (belt clip)
 
 AT command interface (115200 8N1)
 ──────────────────────────────────
-  Query:   AT+RANGE?\r\n
+  Query:
-  Response (from anchors):
+    AT+RANGE?\r\n
-    +RANGE:<anchor_id>,<range_mm>[,<rssi>]\r\n
 
-  Config:
+  Response (from anchors, TWR protocol):
-    AT+anchor_tag=ANCHOR\r\n  — set module as anchor
+    +RANGE:<anchor_id>,<range_mm>[,<rssi>[,<tag_addr>]]\r\n
-    AT+anchor_tag=TAG\r\n     — set module as tag
+
+  Tag pairing (optional — targets a specific enrolled tag):
+    AT+RANGE_ADDR=<tag_addr>\r\n   → anchor only ranges with that tag
 
 Publishes
 ─────────
-  /uwb/target  (geometry_msgs/PoseStamped) — triangulated person position in base_link
+  /uwb/ranges   (saltybot_uwb_msgs/UwbRangeArray)  100 Hz — raw anchor ranges
-  /uwb/ranges  (saltybot_uwb_msgs/UwbRangeArray) — raw ranges from both anchors
+  /uwb/bearing  (saltybot_uwb_msgs/UwbBearing)      10 Hz — Kalman-fused bearing
+  /uwb/target   (geometry_msgs/PoseStamped)          10 Hz — triangulated position
+                                                              (kept for backwards compat)
 
-Safety
+Tag pairing
-──────
+───────────
-  If a range is stale (> range_timeout_s), that anchor is excluded from
+  Set enrolled_tag_ids to a list of tag address strings (e.g. ["0x1234ABCD"]).
-  triangulation. If both anchors are stale, /uwb/target is not published.
+  When non-empty, ranges from unrecognised tags are silently discarded.
+  The matched tag address is stamped in UwbRange.tag_id and UwbBearing.tag_id.
+  When enrolled_tag_ids is empty, all ranges are accepted (tag_id = "").
+
+Parameters
+──────────
+  port_a, port_b          serial ports for anchor-0 / anchor-1
+  baudrate                115200 (default)
+  anchor_separation       centre-to-centre anchor spacing (m)
+  anchor_height           anchor mounting height (m)
+  tag_height              person tag height (m)
+  range_timeout_s         stale-anchor threshold (s)
+  max_range_m / min_range_m  validity window (m)
+  kf_process_noise        Kalman Q scalar
+  kf_meas_noise           Kalman R scalar
+  range_rate              Hz — /uwb/ranges publish rate  (default 100)
+  bearing_rate            Hz — /uwb/bearing publish rate (default 10)
+  enrolled_tag_ids        list[str] — accepted tag addresses; [] = accept all
 
 Usage
 ─────
@@ -44,8 +64,8 @@ from rclpy.node import Node
 from geometry_msgs.msg import PoseStamped
 from std_msgs.msg import Header
 
-from saltybot_uwb_msgs.msg import UwbRange, UwbRangeArray
+from saltybot_uwb_msgs.msg import UwbRange, UwbRangeArray, UwbBearing
-from saltybot_uwb.ranging_math import triangulate_2anchor, KalmanFilter2D
+from saltybot_uwb.ranging_math import triangulate_2anchor, KalmanFilter2D, bearing_from_pos
 
 try:
     import serial
@@ -54,26 +74,31 @@ except ImportError:
     _SERIAL_AVAILABLE = False
 
 
-# Regex: +RANGE:<id>,<mm>  or  +RANGE:<id>,<mm>,<rssi>
+# +RANGE:<id>,<mm>  or  +RANGE:<id>,<mm>,<rssi>  or  +RANGE:<id>,<mm>,<rssi>,<tag>
 _RANGE_RE = re.compile(
-    r"\+RANGE\s*:\s*(\d+)\s*,\s*(\d+)(?:\s*,\s*(-?[\d.]+))?",
+    r"\+RANGE\s*:\s*(\d+)\s*,\s*(\d+)"
+    r"(?:\s*,\s*(-?[\d.]+)"
+    r"(?:\s*,\s*([\w:x]+))?"
+    r")?",
     re.IGNORECASE,
 )
 
 
 class SerialReader(threading.Thread):
     """
-    Background thread: polls an anchor's UART, fires callback on every
+    Background thread: polls one anchor's UART at maximum TWR rate,
-    valid +RANGE response.
+    fires callback on every valid +RANGE response.
+    Supports optional tag pairing via AT+RANGE_ADDR command.
     """
 
-    def __init__(self, port, baudrate, anchor_id, callback, logger):
+    def __init__(self, port, baudrate, anchor_id, callback, logger, tag_addr=None):
         super().__init__(daemon=True)
         self._port      = port
         self._baudrate  = baudrate
         self._anchor_id = anchor_id
         self._callback  = callback
         self._logger    = logger
+        self._tag_addr  = tag_addr
         self._running   = False
         self._ser       = None
 
@@ -86,7 +111,10 @@ class SerialReader(threading.Thread):
                 )
                 self._logger.info(
                     f"Anchor-{self._anchor_id}: opened {self._port}"
+                    + (f" paired with tag {self._tag_addr}" if self._tag_addr else "")
                 )
+                if self._tag_addr:
+                    self._send_pairing_cmd()
                 self._read_loop()
             except Exception as exc:
                 self._logger.warn(
@@ -96,12 +124,24 @@ class SerialReader(threading.Thread):
                     self._ser.close()
                 time.sleep(2.0)
 
+    def _send_pairing_cmd(self):
+        """Configure the anchor to range only with the paired tag."""
+        try:
+            cmd = f"AT+RANGE_ADDR={self._tag_addr}\r\n".encode("ascii")
+            self._ser.write(cmd)
+            time.sleep(0.1)
+            self._logger.info(
+                f"Anchor-{self._anchor_id}: sent tag pairing {self._tag_addr}"
+            )
+        except Exception as exc:
+            self._logger.warn(
+                f"Anchor-{self._anchor_id}: pairing cmd failed: {exc}"
+            )
+
     def _read_loop(self):
         while self._running:
             try:
-                # Query the anchor
                 self._ser.write(b"AT+RANGE?\r\n")
-                # Read up to 10 lines waiting for a +RANGE response
                 for _ in range(10):
                     raw = self._ser.readline()
                     if not raw:
@@ -111,13 +151,14 @@ class SerialReader(threading.Thread):
                     if m:
                         range_mm = int(m.group(2))
                         rssi     = float(m.group(3)) if m.group(3) else 0.0
-                        self._callback(self._anchor_id, range_mm, rssi)
+                        tag_addr = m.group(4) if m.group(4) else ""
+                        self._callback(self._anchor_id, range_mm, rssi, tag_addr)
                         break
             except Exception as exc:
                 self._logger.warn(
                     f"Anchor-{self._anchor_id} read error: {exc}"
                 )
-                break  # trigger reconnect
+                break
 
     def stop(self):
         self._running = False
@@ -130,9 +171,8 @@ class UwbDriverNode(Node):
     def __init__(self):
         super().__init__("uwb_driver")
 
-        # ── Parameters ────────────────────────────────────────────────────────
+        self.declare_parameter("port_a",             "/dev/uwb-anchor0")
-        self.declare_parameter("port_a",           "/dev/ttyUSB0")
+        self.declare_parameter("port_b",             "/dev/uwb-anchor1")
-        self.declare_parameter("port_b",           "/dev/ttyUSB1")
         self.declare_parameter("baudrate",           115200)
         self.declare_parameter("anchor_separation",  0.25)
         self.declare_parameter("anchor_height",      0.80)
@@ -142,36 +182,40 @@ class UwbDriverNode(Node):
         self.declare_parameter("min_range_m",        0.05)
         self.declare_parameter("kf_process_noise",   0.1)
         self.declare_parameter("kf_meas_noise",      0.3)
-        self.declare_parameter("publish_rate",     20.0)
+        self.declare_parameter("range_rate",         100.0)
+        self.declare_parameter("bearing_rate",       10.0)
+        self.declare_parameter("enrolled_tag_ids",   [""])
 
         self._p = self._load_params()
 
-        # ── State (protected by lock) ──────────────────────────────────────
+        raw_ids = self.get_parameter("enrolled_tag_ids").value
+        self._enrolled_tags = [t.strip() for t in raw_ids if t.strip()]
+        paired_tag = self._enrolled_tags[0] if self._enrolled_tags else None
+
         self._lock   = threading.Lock()
-        self._ranges       = {}    # anchor_id → (range_m, rssi, timestamp)
+        self._ranges: dict = {}
         self._kf     = KalmanFilter2D(
             process_noise=self._p["kf_process_noise"],
             measurement_noise=self._p["kf_meas_noise"],
-            dt=1.0 / self._p["publish_rate"],
+            dt=1.0 / self._p["bearing_rate"],
         )
 
-        # ── Publishers ────────────────────────────────────────────────────
+        self._ranges_pub  = self.create_publisher(UwbRangeArray, "/uwb/ranges",  10)
-        self._target_pub = self.create_publisher(
+        self._bearing_pub = self.create_publisher(UwbBearing,    "/uwb/bearing", 10)
-            PoseStamped, "/uwb/target", 10)
+        self._target_pub  = self.create_publisher(PoseStamped,   "/uwb/target",  10)
-        self._ranges_pub = self.create_publisher(
-            UwbRangeArray, "/uwb/ranges", 10)
 
-        # ── Serial readers ────────────────────────────────────────────────
         if _SERIAL_AVAILABLE:
             self._reader_a = SerialReader(
                 self._p["port_a"], self._p["baudrate"],
                 anchor_id=0, callback=self._range_cb,
                 logger=self.get_logger(),
+                tag_addr=paired_tag,
             )
             self._reader_b = SerialReader(
                 self._p["port_b"], self._p["baudrate"],
                 anchor_id=1, callback=self._range_cb,
                 logger=self.get_logger(),
+                tag_addr=paired_tag,
             )
             self._reader_a.start()
             self._reader_b.start()
@@ -180,19 +224,21 @@ class UwbDriverNode(Node):
                 "pyserial not installed — running in simulation mode (no serial I/O)"
             )
 
-        # ── Publish timer ─────────────────────────────────────────────────
+        self._range_timer = self.create_timer(
-        self._timer = self.create_timer(
+            1.0 / self._p["range_rate"], self._range_publish_cb
-            1.0 / self._p["publish_rate"], self._publish_cb
+        )
+        self._bearing_timer = self.create_timer(
+            1.0 / self._p["bearing_rate"], self._bearing_publish_cb
         )
 
         self.get_logger().info(
             f"UWB driver ready  sep={self._p['anchor_separation']}m "
             f"ports={self._p['port_a']},{self._p['port_b']} "
-            f"rate={self._p['publish_rate']}Hz"
+            f"range={self._p['range_rate']}Hz "
+            f"bearing={self._p['bearing_rate']}Hz "
+            f"enrolled_tags={self._enrolled_tags or ['<any>']}"
         )
 
-    # ── Helpers ───────────────────────────────────────────────────────────────
-
     def _load_params(self):
         return {
             "port_a":            self.get_parameter("port_a").value,
@@ -206,90 +252,106 @@ class UwbDriverNode(Node):
             "min_range_m":       self.get_parameter("min_range_m").value,
             "kf_process_noise":  self.get_parameter("kf_process_noise").value,
             "kf_meas_noise":     self.get_parameter("kf_meas_noise").value,
-            "publish_rate":      self.get_parameter("publish_rate").value,
+            "range_rate":        self.get_parameter("range_rate").value,
+            "bearing_rate":      self.get_parameter("bearing_rate").value,
         }
 
-    # ── Callbacks ─────────────────────────────────────────────────────────────
+    def _is_enrolled(self, tag_addr: str) -> bool:
+        if not self._enrolled_tags:
+            return True
+        return tag_addr in self._enrolled_tags
 
-    def _range_cb(self, anchor_id: int, range_mm: int, rssi: float):
+    def _range_cb(self, anchor_id: int, range_mm: int, rssi: float, tag_addr: str):
-        """Called from serial reader threads — thread-safe update."""
+        if not self._is_enrolled(tag_addr):
+            return
         range_m = range_mm / 1000.0
         p = self._p
         if range_m < p["min_range_m"] or range_m > p["max_range_m"]:
             return
         with self._lock:
-            self._ranges[anchor_id] = (range_m, rssi, time.monotonic())
+            self._ranges[anchor_id] = (range_m, rssi, tag_addr, time.monotonic())
 
-    def _publish_cb(self):
+    def _range_publish_cb(self):
+        """100 Hz: publish current raw ranges as UwbRangeArray."""
         now     = time.monotonic()
         timeout = self._p["range_timeout_s"]
-        sep     = self._p["anchor_separation"]
-
         with self._lock:
-            # Collect valid (non-stale) ranges
+            valid = {
-            valid = {}
+                aid: entry
-            for aid, (r, rssi, t) in self._ranges.items():
+                for aid, entry in self._ranges.items()
-                if now - t <= timeout:
+                if (now - entry[3]) <= timeout
-                    valid[aid] = (r, rssi, t)
+            }
-
-        # Build and publish UwbRangeArray regardless (even if partial)
         hdr = Header()
         hdr.stamp    = self.get_clock().now().to_msg()
         hdr.frame_id = "base_link"
-
         arr = UwbRangeArray()
         arr.header = hdr
-        for aid, (r, rssi, _) in valid.items():
+        for aid, (r, rssi, tag_id, _) in valid.items():
             entry            = UwbRange()
             entry.header     = hdr
             entry.anchor_id  = aid
             entry.range_m    = float(r)
             entry.raw_mm     = int(round(r * 1000.0))
             entry.rssi       = float(rssi)
+            entry.tag_id     = tag_id
             arr.ranges.append(entry)
         self._ranges_pub.publish(arr)
 
-        # Need both anchors to triangulate
+    def _bearing_publish_cb(self):
-        if 0 not in valid or 1 not in valid:
+        """10 Hz: Kalman predict+update, publish fused bearing."""
+        now     = time.monotonic()
+        timeout = self._p["range_timeout_s"]
+        sep     = self._p["anchor_separation"]
+        with self._lock:
+            valid = {
+                aid: entry
+                for aid, entry in self._ranges.items()
+                if (now - entry[3]) <= timeout
+            }
+        if not valid:
             return
-
+        both_fresh  = 0 in valid and 1 in valid
+        confidence  = 1.0 if both_fresh else 0.5
+        active_tag  = valid[next(iter(valid))][2]
+        dt = 1.0 / self._p["bearing_rate"]
+        self._kf.predict(dt=dt)
+        if both_fresh:
             r0 = valid[0][0]
             r1 = valid[1][0]
-
             try:
                 x_t, y_t = triangulate_2anchor(
-                r0=r0,
+                    r0=r0, r1=r1, sep=sep,
-                r1=r1,
-                sep=sep,
                     anchor_z=self._p["anchor_height"],
                     tag_z=self._p["tag_height"],
                 )
             except (ValueError, ZeroDivisionError) as exc:
                 self.get_logger().warn(f"Triangulation error: {exc}")
                 return
-
-        # Kalman filter update
-        dt = 1.0 / self._p["publish_rate"]
-        self._kf.predict(dt=dt)
             self._kf.update(x_t, y_t)
         kx, ky = self._kf.position()
-
+        bearing = bearing_from_pos(kx, ky)
-        # Publish PoseStamped in base_link
+        range_m = math.sqrt(kx * kx + ky * ky)
+        hdr = Header()
+        hdr.stamp    = self.get_clock().now().to_msg()
+        hdr.frame_id = "base_link"
+        brg_msg             = UwbBearing()
+        brg_msg.header      = hdr
+        brg_msg.bearing_rad = float(bearing)
+        brg_msg.range_m     = float(range_m)
+        brg_msg.confidence  = float(confidence)
+        brg_msg.tag_id      = active_tag
+        self._bearing_pub.publish(brg_msg)
         pose = PoseStamped()
         pose.header = hdr
         pose.pose.position.x = kx
         pose.pose.position.y = ky
         pose.pose.position.z = 0.0
-        # Orientation: face the person (yaw = atan2(y, x))
+        yaw = bearing
-        yaw = math.atan2(ky, kx)
         pose.pose.orientation.z = math.sin(yaw / 2.0)
         pose.pose.orientation.w = math.cos(yaw / 2.0)
-
         self._target_pub.publish(pose)
 
 
-# ── Entry point ───────────────────────────────────────────────────────────────
-
 def main(args=None):
     rclpy.init(args=args)
     node = UwbDriverNode()

jetson/ros2_ws/src/saltybot_uwb/test/test_ranging_math.py

@@ -7,7 +7,7 @@ No ROS2 / serial / GPU dependencies — runs with plain pytest.
 import math
 import pytest
 
-from saltybot_uwb.ranging_math import triangulate_2anchor, KalmanFilter2D
+from saltybot_uwb.ranging_math import triangulate_2anchor, KalmanFilter2D, bearing_from_pos
 
 
 # ── triangulate_2anchor ───────────────────────────────────────────────────────
@@ -172,3 +172,47 @@ class TestKalmanFilter2D:
         x, y = kf.position()
         assert not math.isnan(x)
         assert not math.isnan(y)
+
+
+# ── bearing_from_pos ──────────────────────────────────────────────────────────
+
+class TestBearingFromPos:
+
+    def test_directly_ahead_zero_bearing(self):
+        """Person directly ahead: x=2, y=0 → bearing ≈ 0."""
+        b = bearing_from_pos(2.0, 0.0)
+        assert abs(b) < 0.001
+
+    def test_left_gives_positive_bearing(self):
+        """Person to the left (y>0): bearing should be positive."""
+        b = bearing_from_pos(1.0, 1.0)
+        assert b > 0.0
+        assert abs(b - math.pi / 4.0) < 0.001
+
+    def test_right_gives_negative_bearing(self):
+        """Person to the right (y<0): bearing should be negative."""
+        b = bearing_from_pos(1.0, -1.0)
+        assert b < 0.0
+        assert abs(b + math.pi / 4.0) < 0.001
+
+    def test_directly_left_ninety_degrees(self):
+        """Person directly to the left: x=0, y=1 → bearing = π/2."""
+        b = bearing_from_pos(0.0, 1.0)
+        assert abs(b - math.pi / 2.0) < 0.001
+
+    def test_directly_right_minus_ninety_degrees(self):
+        """Person directly to the right: x=0, y=-1 → bearing = -π/2."""
+        b = bearing_from_pos(0.0, -1.0)
+        assert abs(b + math.pi / 2.0) < 0.001
+
+    def test_range_pi_to_minus_pi(self):
+        """Bearing is always in -π..+π."""
+        for x in [-2.0, -0.1, 0.1, 2.0]:
+            for y in [-2.0, -0.1, 0.1, 2.0]:
+                b = bearing_from_pos(x, y)
+                assert -math.pi <= b <= math.pi
+
+    def test_no_nan_for_tiny_distance(self):
+        """Very close target should not produce NaN."""
+        b = bearing_from_pos(0.001, 0.001)
+        assert not math.isnan(b)

jetson/ros2_ws/src/saltybot_uwb_msgs/CMakeLists.txt

@@ -8,6 +8,7 @@ find_package(rosidl_default_generators REQUIRED)
 rosidl_generate_interfaces(${PROJECT_NAME}
   "msg/UwbRange.msg"
   "msg/UwbRangeArray.msg"
+  "msg/UwbBearing.msg"
   DEPENDENCIES std_msgs
 )
 

jetson/ros2_ws/src/saltybot_uwb_msgs/msg/UwbBearing.msg

@@ -0,0 +1,18 @@
+# UwbBearing.msg — 10 Hz Kalman-fused bearing estimate from dual-anchor UWB.
+#
+# bearing_rad : horizontal bearing to tag in base_link frame (radians)
+#               positive = tag to the left (CCW), negative = right (CW)
+#               range: -π .. +π
+# range_m     : Kalman-filtered horizontal range to tag (metres)
+# confidence  : quality indicator
+#                 1.0 = both anchors fresh
+#                 0.5 = single anchor only (bearing less reliable)
+#                 0.0 = stale / no data (message not published in this state)
+# tag_id      : enrolled tag identifier that produced this estimate
+
+std_msgs/Header header
+
+float32 bearing_rad
+float32 range_m
+float32 confidence
+string  tag_id

jetson/ros2_ws/src/saltybot_uwb_msgs/msg/UwbRange.msg

@@ -4,6 +4,7 @@
 # range_m   : measured horizontal range in metres (after height correction)
 # raw_mm    : raw TWR range from AT+RANGE? response, millimetres
 # rssi      : received signal strength (dBm), 0 if not reported by module
+# tag_id    : enrolled tag identifier; empty string if tag pairing is disabled
 
 std_msgs/Header header
 
@@ -11,3 +12,4 @@ uint8   anchor_id
 float32 range_m
 uint32  raw_mm
 float32 rssi
+string  tag_id