CAPABLE: Capability-Aware Policy Adaptation via Behavioral Latent Encoding

2026-10-08

AI Summary

This study addresses the vulnerability of vision-language-action policies to joint failures by proposing a unified capability-aware adaptation framework. The method integrates temporal encoders, Jacobian grounding, cross-joint attention, and self-supervised physical prediction mechanisms to infer actual joint capabilities without requiring failure labels or privileged information, enabling online recovery through conditioned residual reinforcement learning. Experimental results on the LIBERO benchmark demonstrate that this framework improves the success rate from 24.8% to 59.3%, surpassing baselines by 17.4 percentage points. Furthermore, it achieves zero-shot fault generalization across tasks and actuators, with its effectiveness validated on real-world robots.

Abstract

Vision-language-action (VLA) policies assume the embodiment on which they were trained and can fail when a joint fault changes how commanded actions are physically executed. Existing fault-recovery methods often require task-specific retraining, fault labels, explicit diagnosis, or privileged embodiment information. We introduce CAPABLE, a unified capability-aware adaptation framework for frozen VLAs that integrates self-supervised capability inference with residual reinforcement learning. CAPABLE infers capability, how much of the commanded motion each joint actually realizes and how that motion contributes to end-effector behavior, online from command-response history and kinematics using a temporal encoder shared across joints, Jacobian grounding, cross-joint attention, and self-supervised physical prediction. The resulting representation conditions a residual policy that adds bounded corrections to the VLA arm action without fault labels or faulty-joint identifiers. Across 28 LIBERO tasks, CAPABLE raises success on an actuator excluded from fault training from 24.8% to 59.3%, outperforming a parameter-matched global-history baseline by 17.4 points while preserving healthy performance. Leave-one-actuator-out experiments across six joints show that this transfer is not specific to one actuator, and additional evaluations characterize transfer to unseen fault families and demonstrate recovery on a physical Franka Panda. https://capable-vla.github.io/

Problem

• Vision-language-action policies
• Joint fault recovery
• Embodiment adaptation
• Fault-tolerant control

Innovation

• Vision-Language-Action (VLA)
• Capability-Aware Adaptation
• Self-Supervised Learning
• Residual Reinforcement Learning
• Fault Recovery