Search papers, labs, and topics across Lattice.
This study investigates the vulnerabilities of Vision-Language-Action (VLA) models when deployed in robotic systems facing joint-level physical faults, such as actuator degradation and increased friction. The authors reveal that these faults significantly disrupt the action-to-motion interface, leading to heterogeneous task success rates that cannot be solely explained by physical infeasibility. To address this issue, they introduce the Joint-level Physical-fault Aware Residual Calibrator (J-PARC), which enhances robustness against these faults while maintaining performance in fault-free conditions.
VLA models can suffer drastic performance drops due to joint-level faults, even when actions are physically feasible.
Deploying Vision-Language-Action (VLA) models in real robotic systems requires robustness not only to semantic and perceptual variations, but also to embodiment-side faults that change how actions are physically realized. Real robots can experience joint-level changes caused by actuator degradation, hardware faults, safety limits, collision damage, or wear-induced friction. These faults are critical because they alter the action-to-motion interface of a policy, disrupting the learned closed-loop relationship between commanded actions, realized motion, and subsequent observations. In this work, we study realistic joint-level physical faults and show that VLA models are vulnerable when predicted actions are executed through a perturbed robot body. Our analysis reveals joint-dependent effects, with heterogeneous degradation in task success across affected joints. We also show that performance drops cannot be attributed solely to physical infeasibility, since feasible faults such as increased joint friction can still substantially reduce success rates and induce closed-loop execution mismatch. Motivated by these findings, we propose Joint-level Physical-fault Aware Residual Calibrator (J-PARC), a lightweight residual calibration framework built on top of a frozen VLA policy. J-PARC infers a latent joint-fault regime from recent joint dynamics and conditions a shared residual calibrator on this regime, enabling adaptive action correction across faulty joints. Experiments show that J-PARC improves robustness under joint-level faults while preserving fault-free environment performance.
