AC-VLA: Robust Out-of-Distribution Action Execution via Compositional Learning
Robotics
Summary
The authors study Vision-Language-Action (VLA) models used for robotic tasks and find they struggle when asked to do new combinations of known actions. They identify two problems: models memorize whole action sequences instead of understanding smaller skills, and they focus too much on close-up visuals rather than the bigger scene. To fix this, they propose AC-VLA, which breaks down instructions into smaller parts and teaches the model to pay attention globally by masking some close-up views during certain actions. Their method improves performance on new, combined tasks by about 28% without hurting normal task success, and works with existing model setups.
Authors
Xiaojiang Peng, Kai Peng, Jie Lu, Zheng Lian, Zitong YU, Xiaobo Wang
Abstract
Vision-Language-Action (VLA) models excel at end-to-end robotic manipulation but struggle with out-of-distribution (OOD) generalization when familiar sub-tasks are recombined in unseen configurations. We identify two mutually reinforcing failure modes: \emph{trajectory overfitting}, where models overfit to holistic trajectory patterns rather than compositional sub-skill semantics; and \emph{perceptual shortcut}, where action tokens over-rely on wrist-view textures at the expense of global spatial grounding. To address both, we introduce \textbf{AC-VLA}, a plug-and-play Action Compositional learning framework comprising two architecture-agnostic components: \textbf{(i)} a compositional learning module that uses an LLM-driven instruction decomposer and a proprioceptive trajectory aligner to generate dense sub-task supervision, followed by mixed training on complete demonstrations and decomposed data to endow the model with compositional generalization; and \textbf{(ii)} a state-conditioned asymmetric masking strategy that suppresses wrist-view inputs during closed-gripper phases, enforcing global semantic grounding. All components are architectural modification-free and directly integrable into any VLA backbone. Instantiated on $π_{0.5}$ and evaluated on LIBERO and LIBERO-OOD benchmarks, AC-VLA achieves a ~28% absolute improvement on compositional OOD tasks while maintaining near-perfect in-distribution performance.