7 Commits

Author SHA1 Message Date
5043578934 feat(social): speech pipeline + LLM conversation + TTS + orchestrator (#81 #83 #85 #89)
Issue #81 — Speech pipeline:
- speech_pipeline_node.py: OpenWakeWord "hey_salty" → Silero VAD → faster-whisper
  STT (Orin GPU, <500ms wake-to-transcript) → ECAPA-TDNN speaker diarization
- speech_utils.py: pcm16↔float32, EnergyVad, UtteranceSegmenter (pre-roll, max-
  duration), cosine speaker identification — all pure Python, no ROS2/GPU needed
- Publishes /social/speech/transcript (SpeechTranscript) + /social/speech/vad_state

Issue #83 — Conversation engine:
- conversation_node.py: llama-cpp-python GGUF (Phi-3-mini Q4_K_M, 20 GPU layers),
  streaming token output, per-person sliding-window context (4K tokens), summary
  compression, SOUL.md system prompt, group mode
- llm_context.py: PersonContext, ContextStore (JSON persistence), build_llama_prompt
  (ChatML format), context compression via LLM summarization
- Publishes /social/conversation/response (ConversationResponse, partial + final)

Issue #85 — Streaming TTS:
- tts_node.py: Piper ONNX streaming synthesis, sentence-by-sentence first-chunk
  streaming (<200ms to first audio), sounddevice USB speaker playback, volume control
- tts_utils.py: split_sentences, pcm16_to_wav_bytes, chunk_pcm, apply_volume, strip_ssml

Issue #89 — Pipeline orchestrator:
- orchestrator_node.py: IDLE→LISTENING→THINKING→SPEAKING state machine, GPU memory
  watchdog (throttle at <2GB free), rolling latency stats (p50/p95 per stage),
  VAD watchdog (alert if speech pipeline hangs), /social/orchestrator/state JSON pub
- social_bot.launch.py: brings up all 4 nodes with TimerAction delays

New messages: SpeechTranscript.msg, VadState.msg, ConversationResponse.msg
Config YAMLs: speech_params, conversation_params, tts_params, orchestrator_params
Tests: 58 tests (28 speech_utils + 30 llm_context/tts_utils), all passing

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
2026-03-02 08:23:19 -05:00
fa0162fadc feat(social): multi-modal tracking fusion — UWB+camera Kalman filter (Issue #92)
New packages:
  saltybot_social_msgs   — FusedTarget.msg custom message
  saltybot_social_tracking — 4-state Kalman fusion node

saltybot_social_tracking/tracking_fusion_node.py
  Subscribes to /uwb/target (PoseStamped, ~10 Hz) and /person/target
  (PoseStamped, ~30 Hz) and publishes /social/tracking/fused_target
  (FusedTarget) at 20 Hz.

  Source arbitration:
    • "fused"     — both UWB and camera are fresh; confidence-weighted blend
    • "uwb"       — UWB fresh, camera stale
    • "camera"    — camera fresh, UWB stale
    • "predicted" — all sources stale; KF coasts for up to predict_timeout (3 s)

  Kalman filter (kalman_tracker.py):
    State [x, y, vx, vy] with discrete Wiener acceleration noise model
    (process_noise=3.0 m/s²) sized for EUC speeds (20-30 km/h, ≈5.5-8.3 m/s).
    Separate UWB (0.20 m) and camera (0.12 m) measurement noise.
    Velocity estimate converges after ~3 s of 10 Hz UWB measurements.

  Confidence model (source_arbiter.py):
    Per-source confidence = quality × max(0, 1 - age/timeout).
    Composite confidence accounts for KF positional uncertainty and
    is capped at 0.4 during dead-reckoning ("predicted") mode.

Tests: 58/58 pass (no ROS2 runtime required).

Note: saltybot_social_msgs here adds FusedTarget.msg; PR #98
(Issue #84) adds PersonalityState.msg + QueryMood.srv to the same
package. The maintainer should squash-merge #98 first and rebase
this branch on top of it before merging to avoid the package.xml
conflict.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
2026-03-01 23:59:10 -05:00
44771751e2 feat(social): personality system — SOUL.md persona, mood engine, relationship DB (Issue #84)
New packages:
- saltybot_social_msgs: PersonalityState.msg + QueryMood.srv custom interfaces
- saltybot_social_personality: full personality node

Features:
- SOUL.md YAML/Markdown persona file: name, humor_level (0-10), sass_level (0-10),
  base_mood, per-tier greeting templates, mood prefix strings
- Hot-reload: SoulWatcher polls SOUL.md every reload_interval seconds, applies
  changes live without restarting the node
- Per-person relationship memory in SQLite: score, interaction_count,
  first/last_seen, learned preferences (JSON), full interaction log
- Mood engine (pure functions): happy | curious | annoyed | playful
  driven by relationship score, interaction count, recent event window (120s)
- Greeting personalisation: stranger | regular | favorite tiers
  keyed on interaction count thresholds from SOUL.md
- Publishes /social/personality/state (PersonalityState) at publish_rate Hz
- /social/personality/query_mood (QueryMood) service for on-demand mood query
- Full ROS2 dynamic reconfigure: soul_file, db_path, reload_interval, publish_rate
- 52 unit tests, no ROS2 runtime required

ROS2 interfaces:
  Sub: /social/person_detected  (std_msgs/String JSON)
  Pub: /social/personality/state (saltybot_social_msgs/PersonalityState)
  Srv: /social/personality/query_mood (saltybot_social_msgs/QueryMood)
2026-03-01 23:56:05 -05:00
d6a6965af6 Merge pull request 'feat(social): person enrollment system #87' (#95) from sl-perception/social-enrollment into main 2026-03-01 23:55:16 -05:00
5143e5bfac feat(social): Issue #86 — physical expression + motor attention
ESP32-C3 NeoPixel sketch (esp32/social_expression/social_expression.ino):
  - Adafruit NeoPixel + ArduinoJson, serial JSON protocol 115200 8N1
  - Mood→colour: happy=green, curious=blue, annoyed=red, playful=rainbow
  - Idle breathing animation (sine-modulated warm white)
  - Auto-falls to idle after IDLE_TIMEOUT_MS (3 s) with no command

ROS2 saltybot_social_msgs (new package):
  - Mood.msg — {mood, intensity}
  - Person.msg — {track_id, bearing_rad, distance_m, confidence, is_speaking, source}
  - PersonArray.msg — {persons[], active_id}

ROS2 saltybot_social (new package):
  - expression_node: subscribes /social/mood → JSON serial to ESP32-C3
    reconnects on port error; sends idle frame after idle_timeout_s
  - attention_node: subscribes /social/persons → /cmd_vel rotation-only
    proportional control with dead zone; prefers active speaker, falls
    back to highest-confidence person; lost-target idle after 2 s
  - launch/social.launch.py — combined launch
  - config YAML for both nodes with documented parameters
  - test/test_attention.py — 15 pytest-only unit tests

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
2026-03-01 23:35:59 -05:00
5c4f18e46c feat(social): person enrollment system — SQLite gallery + voice trigger (Issue #87)
- saltybot_social_msgs: 6 msg + 5 srv definitions for social interaction
- saltybot_social_enrollment: enrollment_node + enrollment_cli
- PersonDB: thread-safe SQLite-backed gallery (embeddings, voice samples)
- Voice-triggered enrollment via "remember me my name is X" phrase
- CLI: enroll/list/delete/rename via ros2 run
- Services: /social/enroll, /social/persons/list|delete|update
- Gallery sync from /social/faces/embeddings topic
2026-03-01 23:32:26 -05:00
84790412d6 feat(social): multi-modal person state tracker (Issue #82) 2026-03-01 23:08:22 -05:00