- Add VoskSTT class to audio_utils.py: offline Vosk STT backend as
low-latency CPU alternative to Whisper for Jetson deployments
- Update audio_pipeline_node.py: stt_backend param ("whisper"/"vosk"),
Vosk loading with Whisper fallback, CPU auto-detection for Whisper,
dual-backend _process_utterance dispatch, STT/<backend> log prefix
- Update audio_pipeline_params.yaml: add stt_backend and vosk_model_path
- Add test/test_audio_pipeline.py: 40 unit tests covering EnergyVAD,
PCM conversion, AudioBuffer, UtteranceSegmenter, VoskSTT, JabraAudioDevice,
AudioMetrics, AudioState
- Integrate into full_stack.launch.py: audio_pipeline at t=5s with
enable_audio_pipeline and audio_stt_backend args
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- Integrate saltybot_docking package into full_stack.launch.py
- Auto-trigger docking when battery drops to 20% (configurable via battery_low_pct)
- Launch docking at t=7s (after sensors, before Nav2)
- Add /saltybot/docking_state publisher (std_msgs/String) for state monitoring
- Update docking_params.yaml:
- battery_low_pct: 15% → 20% per Issue #489
- Add references to Issue #475 for conservative FC+hoverboard speeds
- Docking behavior includes:
- ArUco marker or IR beacon detection for dock location
- Nav2-based approach to pre-dock pose (~1m away)
- Visual servoing final alignment with contact detection
- Auto-undocking on full charge (80%) or command
- Integration with power management for mission interruption/resumption
Co-Authored-By: Claude Haiku 4.5 <noreply@anthropic.com>
Natural language voice command routing with fuzzy matching for speech variations.
Supported Commands:
- Follow me / Come with me
- Stop / Halt / Freeze
- Go home / Return to dock / Charge
- Patrol / Autonomous mode
- Come here / Approach
- Sit / Sit down
- Spin / Rotate / Turn around
- Dance / Groove
- Take photo / Picture / Smile
- What's that / Identify / Recognize
- Battery status / Battery level
Features:
- Fuzzy matching (rapidfuzz token_set_ratio) with 75% threshold
- Multiple pattern support per command for natural variations
- Three routing types: velocity (/cmd_vel), actions (/saltybot/action_command), services
- Command monitoring via /saltybot/voice_command
- Graceful handling of unrecognized speech
Architecture:
- Input: /saltybot/speech/transcribed_text (lowercase text)
- Fuzzy match against 11 command groups with 40+ patterns
- Route to: /cmd_vel (velocity), /saltybot/action_command (actions), or services
Files:
- saltybot_voice_router_node.py: Main router with fuzzy matching
- launch/voice_router.launch.py: Launch configuration
- VOICE_ROUTER_README.md: Usage documentation
Dependencies:
- rapidfuzz: Fuzzy string matching for natural speech handling
- rclpy, std_msgs, geometry_msgs: ROS2 core
Performance: <100ms per command (fuzzy matching + routing)
Co-Authored-By: Claude Haiku 4.5 <noreply@anthropic.com>
- saltybot_object_detection_msgs: DetectedObject, DetectedObjectArray, QueryObjects.srv
- saltybot_object_detection: YOLOv8n TensorRT FP16 node with depth projection
- Message filters for RGB-depth sync, TF2 transform to base_link
- Configurable confidence and class filtering (COCO 80 classes)
- Query service for voice integration ("whats in front of you")
- TensorRT build script with ONNX fallback
- Launch file with parameter configuration
- Full stack integration at t=6s (30 FPS target alongside person tracker)
Co-Authored-By: Claude Haiku 4.5 <noreply@anthropic.com>
