camera_vision is reserved for when an RGB camera enters the sensor stack. Its job will be to subscribe to the raw camera topic, run lightweight image preprocessing (resize, color conversion, optional CNN feature extraction), and publish a compact feature vector the policy can consume.
Current status: not implemented. The current indoor rangefinder stack (TF-Luna + VL53L0X array + PMW3901 optical flow) doesn’t depend on a camera. A camera becomes interesting for two use cases: SAR thermal imaging (where the camera is the primary sensor, not auxiliary) and ML-based object detection for higher-level mission logic. Neither is in the H2 2026 scope — see roadmap.
Planned interfaces:
- Input:
/drone/camera/raw(sensor_msgs/Image). - Output:
/drone/camera/features— compact feature vector (exact schema TBD; depends on the model).
Where to go next
- Sensor-processing nodes hub — sibling nodes
- Roadmap H2 2026 — when this work might land
- SAR business case — the use case that motivates serious camera work