MANUSH Perception

Role (Doc B §12): Computer vision and sensor fusion.

MANUSH Perception turns raw sensor streams into a consistent understanding of the scene: objects, people, surfaces and free space.

Responsibilities

• Camera pipelines (RGB, depth, stereo, thermal) with calibration and time synchronisation.
• Sensor fusion across cameras, LiDAR, IMU, joint encoders, force/torque and tactile sensors.
• Object detection, segmentation, 6-DoF pose estimation and tracking.
• Human detection and body-pose estimation — the input MANUSH Safety uses for proximity limits.
• Robustness to Indian operating conditions: glare, dust, monsoon rain, low light and crowded scenes.

Building on Perception

Perception modules you write consume synchronised sensor frames and publish typed observations (detections, poses, occupancy). Declare the sensors your skill needs in required_sensors — for example rgb_camera, depth_camera, thermal_camera, lidar.

Train detectors on labelled synthetic data from the MANUSH Sim Synthetic data generation and Synthetic objects environments before collecting real-world data.