RoboPace speeds robots while safely handling contact tasks
RoboPace: Contact-Aware Time-Optimal Retiming for Action-Chunk Policies
Robotics
Summary
Robots often learn how to move from watching humans, but humans move at a pace that doesn’t always work well for robots. Fast robot moves can cause mistakes when touching objects, while moving too slowly wastes time. The authors created RoboPace, a method that changes how quickly a robot performs a task without changing the path it takes. It carefully slows down or speeds up depending on when the robot will touch something, making robots finish tasks faster while still being safe and accurate.
What this means in practice
- •For industrial robot operators: Improve robot task speed without sacrificing safety during contact-rich assembly operations by adjusting movement timing in real time.
- •For robotics software engineers: Integrate a retiming layer that automatically adapts policy execution speed based on contact prediction and robot constraints without retraining models.
Authors
Mimo Shirasaka, Takehiko Ohkawa, Takuya Okubo, Nicola Scianca, Tatsuya Matsushima, Kei Ota
Abstract
Robot manipulation data collection has been shifting from teleoperation toward robot-free demonstrations, through interfaces such as the Universal Manipulation Interface (UMI) or directly from human hands. Vision-Language-Action (VLA) policies trained on such data inherit the demonstrator's timing. Yet human timing does not directly transfer to robots: compliant hands tolerate fast contact, whereas robots may overshoot due to actuator and tracking limitations; conversely, robots can move faster in free space. This motivates a unified approach that reconciles execution speed with contact safety. We present RoboPace, an online retiming layer that preserves the policy's geometric path while adapting its timing, respecting the target robot's kinematic and dynamic constraints. It adapts execution speed based on predicted contact, jointly accounting for contact-dependent speed limits and the robot's motion constraints. The method requires no policy retraining and operates in real time. Across three contact-rich tasks on a dual-arm robot, faster uniform execution and physical-limit-only retiming largely fail. RoboPace instead achieves higher overall success than slow uniform execution while completing four of five commands in approximately half the time, retaining the reliability of slow execution without its time cost.