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

Key points

• Close-proximity multi-arm manipulation requires collision models that are both geometrically accurate and differentiable enough for real-time optimization.1
• PI-UDF combines analytical forward kinematics with learnable link-geometry embeddings and a shared residual network to predict pairwise inter-arm distances directly from robot configurations.1
• We validate the framework on a real dual-Franka platform through high-speed close-proximity 14-DoF dual-arm swapping, sustained single-arm dynamic evasion, and dynamic-evasion planning configurations with frozen, predicted, and target-switching treatments of the moving arm.1
• Hardware experiments and offline Drake/FCL replay show that PI-UDF provides a differentiable inter-arm clearance estimate suitable for closed-loop collision-aware collaborative robot motion generation.1

Sources (1)

• [1]A Physics-Informed Collision Learning Framework for Collaborative Robot Motion Generation
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 8, 05:45 PM

  We present PI-UDF, a physics-informed unified differentiable framework for body-to-body collision distance prediction between articulated robots.

  Close-proximity multi-arm manipulation requires collision models that are both geometrically accurate and differentiable enough for real-time optimization.

Extractive summary: sentences quoted from the sources.