Taming VLAs under Robot Execution Errors: Self-Compensation and Stress Testing
Self-compensating VLAs adapt online to robot execution errors, raising task success by over 30 points.
Vision-language-action policies often fail when executed robot motion diverges from the commanded action because of wear, payload, friction, backlash, or compliance. Self-compensating VLA updates the policy at deployment from the residual between commanded actions and executed motion, without task rewards or labels. On RoboStress, a simulation benchmark of seven scenarios built from joint-level friction, backlash, compliance, and gravity-compensation errors, it outperformed base policies and training-time robustness methods. On two physical robot arms with different usage histories, average task success rose by more than 30 percentage points, including on unseen objects.
- Self-compensating VLA adapts online from commanded-versus-executed motion residuals.
- No task rewards or extra labels are required at deployment.
- RoboStress covers seven state- and history-dependent execution-error scenarios.
- Two physical arms gained more than 30 points in average task success.
- Products
- RoboStress
From the original article
1 min read at huggingface.co
Vision-language-action (VLA) policies often fail when a robot's executed motion deviates from their commanded action. Such execution errors arise from the robot's mechanics and operating conditions, such as wear and payload changes. We propose self-compensating VLA, a deployment-time adaptation method that enables a VLA policy to pre-compensate for the robot's execution errors when generating commands.
By Sohyun Lee, Yoonjae Baek, Jaesang Won · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.