Event
RAPAC-DP: A Pending-Action Compensation Framework for Delayed Execution
1 reports · 1 sources
Tao Wang and co-authors propose RAPAC-DP, a response-aligned pending-action compensation framework for diffusion and flow-based action generators, aimed at enabling such policies to still output complete action chunks aligned with the handover moment under cloud inference latency. The method uses a lightweight pending-action encoder and LoRA adapters to learn compensation, keeps the base policy frozen, and constructs delay-conditioned training samples from latency-free demonstrations; the authors state that no explicit dynamics model or additional delay data is required. The results reported by the authors are: at the maximum evaluated delay, Kinetix retains 81.4% of its latency-free overall performance in simulation, and RoboMimic achieves an average success rate of 63.3% across three tasks; in 450 physical trials on a Franka Panda robotic arm, the average success rate under high delay for static and conveyor pick-and-place tasks increased from 28.9% for RTC to 60.0%.
Report timeline
- RAPAC-DP: Response-Aligned Pending-Action Compensation for Diffusion Policies under Delayed Execution
RAPAC-DP proposes a response-aligned pending-action compensation framework that enables diffusion and flow-based action generators to still output full action chunks aligned with the handover moment under cloud inference latency. The method uses a lightweight pending-action encoder and LoRA adapters to learn the compensation while keeping the base policy frozen, and constructs delay-conditioned training samples from delay-free demonstrations, requiring neither an explicit dynamics model nor additional delay data.
arXiv cs.RO ·