Vision language action policies show similar but distinct robot hand movements
Outcome-Conditioned End-Effector Geometry Across Vision-Language-Action Policies
RoboticsArtificial Intelligence
Summary
Different vision-language-action policies can complete the same robot manipulation tasks but try very different ways with the robot hand. The authors compared how close these hand movements are when the tasks succeed or fail. They found that when both policies succeed, their hand movements are quite similar, but still show some clear differences that mean one policy cannot simply replace the other. Failures lead to larger differences, and factors like starting conditions or visual distractions also change how similar the movements are. The study helps understand the relationship between success in tasks and the physical motions of robots doing them.
What this means in practice
- •For robotics engineers: Identify differences in robot hand motions across planning methods to improve multi-policy robotic manipulation systems.
- •For warehouse automation teams: Evaluate robot control policies to ensure consistent behavior in picking and placing tasks under varying conditions.
Authors
Xingyu Lin, Zhuang Li, Zhongrun Wu, Shouquan Zhou, Dehui Du
Abstract
Vision-language-action (VLA) policies solve the same manipulation task through different action interfaces, but task success alone does not establish whether their physical executions agree. We study cross-policy end-effector geometry in 15,000 closed-loop LIBERO rollouts from four policies. The primary clean-condition analysis forms 3,600 configuration-matched, and therefore dependent, policy pairs. Both-success pairs have a median normalized dynamic time warping distance of 0.0120 m versus 0.0380 m when exactly one policy succeeds. This ordering holds in every task, every policy pair, and nine sampling and band-limited representations; however, the ratio varies severalfold across representations, so we report the direction rather than a fixed multiple. Both-failure pairs are more separated again but rest on thin, uneven support, so we report them as exploratory. Within successful executions, partner replacements separate more across tasks than across initial states. A matched baseline still reveals measurable, heterogeneous residual policy differences, so a low cross-policy distance does not imply interchangeability. Successful executions sit about as far from same-task demonstrations as those demonstrations sit from each other, compatible with task-associated geometry without separating training-data overlap from task constraints. A common 72-action window preserves the ordering but reduces its magnitude; endpoint and duration adjustment likewise leaves a positive mixed-outcome coefficient relative to both-success pairs, though its magnitude is specification-dependent. Under composite visual stress, policy rankings and pair composition change together.