Robots learn natural speeds for smoother and faster actions
Retiming Robot Demonstrations with a Learned Speed Prior
Artificial IntelligenceRobotics
Summary
Robot demonstrations can vary a lot in speed depending on the person controlling them, making it tough for robots to learn consistent timing. The authors teach robots to predict a natural speed for each task by learning from many examples. This lets robots perform actions more smoothly and quickly by retiming their movements based on the learned speed. They tested this on many tasks and a real robot, showing better success and faster completion than before.
What this means in practice
- •For robotics engineers: Improve robot task execution by using learned natural speed profiles for smoother and faster movements.
- •For industrial automation teams: Enhance robot teleoperation data by retiming demonstrations to standardize speed, leading to better training of control policies.
- •For robotics hardware manufacturers: Integrate learned speed prior models with robot control layers to deliver improved real-time task execution performance.$Commercial implications: This enables selling robots or control systems that perform faster and more reliably with natural task pacing.
Authors
Seungyong Moon, Deokjae Lee, Hansang Lee, Munkhtselmeg Battsengel, Youngin Kim, Seonghyeon Yun, Sungjin Jeon, SongEun Kim, Jeongmin Park, Hyun Oh Song
Abstract
The pace of a robot demonstration is often determined by the teleoperator rather than by the task itself. As a result, the same motion can appear at widely different speeds across demonstrations, and policies trained on such data may inherit operator-specific timing along with task behavior. We propose to learn a speed prior directly from large-scale demonstration data. Given a task instruction, a motion path and the images along it, it predicts a distribution over execution speed along the path, from which a chosen quantile defines a consistent clock for retiming each demonstration. Existing methods derive the acceleration signal from the very data they accelerate, so each new task costs another fit, annotation, or pair of predefined rates. A single trained speed prior instead covers every task in a corpus. We fit a speed prior on both robot teleoperation data and handheld demonstrations in which a human arm, rather than a robot arm, moves the gripper, and use it to retime robot demonstrations to the handheld pace before fine-tuning a vision-language-action policy. For execution, we pair the policy with a real-time optimal-control layer that tracks predicted action chunks under robot constraints using the predicted pace. On 120 RoboLab tasks, policies fine-tuned on our learned clocks succeed more often and move with up to half the jerk compared with policies fine-tuned on the original demonstration timing and uniformly accelerated to the same speed. On a real robot, our policy stacks cups in 3.50s, 1.72$\times$ faster than the default policy's 6.02s while succeeding as often, in 29 of 30 trials. Videos and further material are available at https://mllab.snu.ac.kr/projects/learned_speed_prior.