Capability aware adaptation improves robot arm fault recovery online
CAPABLE: Capability-Aware Policy Adaptation via Behavioral Latent Encoding
RoboticsMachine Learning
Summary
Robot arms sometimes don't move as expected because of faults like broken joints. The authors developed CAPABLE, a way for a robot’s control system to figure out which parts are not working well by watching how commands actually affect movement. This system then adjusts commands in real time to help the robot succeed at tasks without needing prior knowledge about the faults. It works better than previous methods and even works on real robot arms.
What this means in practice
- •For robotics engineers: Enable robot arms to adapt in real time to joint faults without requiring fault labels or offline retraining.
- •For robotics system integrators: Improve robot task success rates by integrating capability-aware adaptations that recover from actuator faults across different robot designs.
Authors
Mohammad Khoshnazar, Mohammad Dehghani Tezerjani, Deyuan Qu, Zhiyuan Gao, Yanxiang Zhan, Jeroen Schafer, Andrew Melnik, Qing Yang, Michael Beetz
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/