Humanoid robots learn to selectively replace arm movements during motion control
RedirectMimic: Local Motion Content Redaction for Humanoid Whole-Body Control
Robotics
Summary
Humanoid robots often follow a set of reference motions, but sometimes specific movements need to be replaced without changing the whole behavior. The researchers developed RedirectMimic, a method that allows the robot’s controller to ignore certain arm motions and replace them with new ones while still responding normally to other movements. This method was tested in simulated robot arms and showed better performance in editing the arm motions than previous approaches, without causing the robot to fall or lose balance. RedirectMimic keeps the robot’s original motions accurate while making targeted adjustments.
What this means in practice
- •For robotics engineers: Improve humanoid robot controllers to selectively modify specific arm motions while maintaining balance and overall motion accuracy.
- •For game developers: Create more flexible robot character animations that can locally alter gestures or movements without redoing entire motion sequences.
Authors
Zepeng Wang, Xiaochuan Shi
Abstract
Humanoid motion interfaces should accommodate application-specific movement constraints while retaining useful reference behavior. Local content redaction requires a controller to replace selected movements even when it continues to receive the original reference. Removing the tracking incentive for those movements leaves their replacement unspecified. We introduce RedirectMimic, which separates the reference observed by the actor from the target used to train its execution. A rule-generated edited reference guides tracking rewards during proximal policy optimization, while the deployed observation and action interfaces remain unchanged. We evaluate pose- and speed-defined arm editing in Unitree G1 simulation using a tracker pretrained on 7,509 motion clips. On the overhead-arm task, RedirectMimic reduces core-window content occurrence to 2.7%, compared with 58.9% for reward zeroing and 32.8% for anti-tracking. It achieves 94.2% core-window joint success, combining content suppression, fall-free execution, and local retention. On the speed-defined task, fast outward arm movements occurred at similarly low rates under all three local editing methods. Across both tasks, RedirectMimic achieved the lowest replacement error among these methods and tracked unedited motions with accuracy comparable to that of the pretrained controller. Project page: https://xiaopeng-whu.github.io/RedirectMimic/.