HumAIN is a framework for social robot navigation that injects implicit human cues directly into the planning loop. A transformer-based teacher combines historical images, skeletal keypoints, robot state, and the robot’s goal to learn human-aware representations for future trajectory planning. This knowledge is distilled into a lightweight student model that uses fewer inputs for real-time inference. The authors report an average 29.8% improvement across trajectory-prediction metrics over state-of-the-art baselines. The approach targets resource-constrained robotic platforms while attempting to preserve sensitivity to whole-body social signals.
No heat snapshots are available in the last 24 hours.