A Physics-Informed Collision Learning Framework for Collaborative Robot Motion Generation

This research proposes PI-UDF, a differentiable framework for predicting inter-arm collision distances in multi-arm robotic systems. It integrates analytical forward kinematics with learnable link-geometry embeddings and a shared residual network to enable real-time collision-aware motion planning. The framework is validated on a dual-Frank, demonstrating its effectiveness in high-speed close-proximity manipulation tasks.