CAPABLE: Capability-Aware Policy Adaptation via Behavioral Latent Encoding

Mohammad Khoshnazar; Mohammad Dehghani Tezerjani; Deyuan Qu; Zhiyuan Gao; Yanxiang Zhan; Jeroen Schafer; Andrew Melnik; Qing Yang

cs.ROOct 8, 2026

CAPABLE: Capability-Aware Policy Adaptation via Behavioral Latent Encoding

Authors: Mohammad Khoshnazar, Mohammad Dehghani Tezerjani, Deyuan Qu, Zhiyuan Gao, Yanxiang Zhan, Jeroen Schafer, Andrew Melnik, Qing Yang, +1 more

Organizations: University of Bremen, Bremen, Germany. · University of North Texas, Denton, TX, USA. · Toyota Motor North America, InfoTech Labs, Mountain View, CA, USA.

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/

Figures & tables

Fig. 1: The illustrated sequence (1) opens the drawer , (2) lifts the plate , and (3) places it inside . A faulty joint (red highlight) prevents the frozen VLA motion from succeeding (red cross); CAPABLE supplies a corrective motion (green check). CAPABLE infers each joint’s remaining capability from its command–response history and the live Jacobian, and adds a bounded residual correction to the frozen VLA action. No fault label or explicit joint-index embedding is supplied, allowing the shared estimator to transfer to an actuator whose failure was not observed during training.

Fig. 2: CAPABLE architecture. Shared encoders map per-joint command–response histories and live kinematics to $z_{t}^{\mathrm{cap}}$ , which FiLM-conditions the residual SAC actor. The bounded arm correction is added to the frozen VLA action; the gripper bypasses RL. Critics, replay and rewards are training-only.

Method	Base VLA	Residual	Oracle fault	Oracle task state	Joint-fact.
Base VLA (no correction)	Yes	No	No	No	No
POMDP-RL	No	No	No	No	No
Classical RR	Yes	No	No	No	Yes
DEFT	No	No	Yes	Yes	No
Global-history SAC	Yes	Yes	No	No	No
CAPABLE (ours)	Yes	Yes	No	No	Yes

TABLE I: Comparison assumptions. Learned methods use the same task pool, fault conditions, budget, and held-out states. Classical RR is non-learning; DEFT receives simulator-derived goals and actuator availability.

Method	$H$	$S$	$U$	$\Delta$ vs. Base
Base VLA	$\mathbf{91.4\!\pm\!0.0}$	$41.6\!\pm\!0.0$	$24.8\!\pm\!0.0$	–
POMDP-RL	$19.6\!\pm\!2.4$	$11.3\!\pm\!2.1$	$10.1\!\pm\!1.9$	$-14.7$
Classical RR	$81.6\!\pm\!0.0$	$61.3\!\pm\!0.0$	$32.9\!\pm\!0.0$	$+8.1$
DEFT	$89.7\!\pm\!1.4$	$70.1\!\pm\!2.3$	$\mathbf{62.8\!\pm\!2.9}$	$\mathbf{+38.0}$
Global-history SAC	$88.1\!\pm\!1.3$	$73.8\!\pm\!2.1$	$41.9\!\pm\!3.2$	$+17.1$
CAPABLE	$90.8\!\pm\!1.0$	$\mathbf{86.7\!\pm\!1.8}$	$59.3\!\pm\!3.0$	$+34.5$

TABLE II: Main 28-task success (%).

H/S/U

: healthy/seen/unseen-

j_{2}

;

\Delta

is versus Base VLA on unseen

j_{2}

. Classical RR is non-learning; DEFT uses oracle fault and task state.

Suite	Base VLA	Global-history SAC	CAPABLE	$\Delta$ vs. Global
LIBERO-10	$8.0\!\pm\!0.0$	$26.5\!\pm\!4.6$	$\mathbf{45.2\!\pm\!4.1}$	$+18.7$
LIBERO-Goal	$40.0\!\pm\!0.0$	$49.0\!\pm\!3.8$	$\mathbf{73.3\!\pm\!4.7}$	$+24.3$
LIBERO-Object	$20.8\!\pm\!0.0$	$52.8\!\pm\!3.3$	$\mathbf{65.6\!\pm\!4.4}$	$+12.8$
LIBERO-Spatial	$41.4\!\pm\!0.0$	$49.5\!\pm\!1.2$	$\mathbf{64.0\!\pm\!3.8}$	$+14.5$
Task-balanced mean	$24.8\!\pm\!0.0$	$41.9\!\pm\!3.2$	$\mathbf{59.3\!\pm\!3.0}$	$+17.4$

TABLE III: Unseen-

j_{2}

success by suite (%).

Method	Persistent lock	Damping	Friction	Partial effectiveness	Range restriction	Late-onset lock
Base VLA	$24.8\!\pm\!0.0$	$45.8\!\pm\!0.0$	$40.6\!\pm\!0.0$	$44.2\!\pm\!0.0$	$32.7\!\pm\!0.0$	$50.4\!\pm\!0.0$
Global-history SAC	$41.9\!\pm\!3.2$	$55.6\!\pm\!1.4$	$51.8\!\pm\!1.6$	$\mathbf{64.1\!\pm\!3.3}$	$\mathbf{54.7\!\pm\!2.8}$	$61.5\!\pm\!3.1$
CAPABLE	$\mathbf{59.3\!\pm\!3.0}$	$\mathbf{67.8\!\pm\!2.9}$	$\mathbf{64.2\!\pm\!3.2}$	$62.3\!\pm\!3.0$	$48.2\!\pm\!3.5$	$\mathbf{72.4\!\pm\!2.7}$
$\Delta$ vs. Global	$+17.4$	$+12.2$	$+12.4$	$-1.8$	$-6.5$	$+10.9$

TABLE IV: Zero-shot cross-fault success on unseen

j_{2}

(%). Training uses only persistent locks; non-lock columns average three severities.

Held-out joint	Global-history SAC	CAPABLE	$\Delta$ vs. Global
$j_{0}$	$55.0\!\pm\!4.4$	$\mathbf{76.7\!\pm\!3.3}$	$+21.7$
$j_{1}$	$39.4\!\pm\!1.3$	$\mathbf{56.2\!\pm\!2.3}$	$+16.8$
$j_{2}$	$46.6\!\pm\!5.2$	$\mathbf{69.3\!\pm\!3.0}$	$+22.7$
$j_{3}$	$39.6\!\pm\!4.6$	$\mathbf{63.6\!\pm\!4.7}$	$+24.0$
$j_{4}$	$44.7\!\pm\!5.1$	$\mathbf{88.1\!\pm\!1.9}$	$+43.4$
$j_{5}$	$29.3\!\pm\!4.2$	$\mathbf{48.2\!\pm\!3.6}$	$+18.9$

TABLE V: Leave-one-actuator-out transfer across six held-out joints on eight LIBERO tasks. Each row uses a separate leave-one-actuator-out training split in which the indicated joint is excluded from fault training and evaluated zero-shot. Values are success rates (%).

Variant	$P$	$H$	$S$	$U$	$\Delta$ vs. Full
Full CAPABLE	$0.82$	$90.8\!\pm\!1.0$	$\mathbf{86.7\!\pm\!1.8}$	$\mathbf{59.3\!\pm\!3.0}$	–
Global-history SAC	$0.82$	$88.1\!\pm\!1.3$	$73.8\!\pm\!2.1$	$41.9\!\pm\!3.2$	$-17.4$
Equal-strata replay	$0.82$	$\mathbf{92.3\!\pm\!0.9}$	$71.5\!\pm\!2.6$	$46.4\!\pm\!3.7$	$-12.9$
No Jacobian query	$0.81$	$89.8\!\pm\!1.4$	$75.9\!\pm\!2.4$	$47.8\!\pm\!3.8$	$-11.5$
No cross-joint attention	$0.71$	$89.2\!\pm\!1.5$	$81.2\!\pm\!2.7$	$46.1\!\pm\!3.6$	$-13.2$
No capability loss	$0.80$	$88.7\!\pm\!1.7$	$72.0\!\pm\!3.1$	$43.7\!\pm\!4.6$	$-15.6$

TABLE VI: 28-task ablations.

P

: parameters (M);

H/S/U

: healthy/seen/unseen-

j_{2}

;

\Delta

is versus full CAPABLE on unseen

j_{2}

Task	Locked joint	Global-history SAC	CAPABLE
Open drawer	$j_{0}$ (seen)	5/10 (50%)	9/10 (90%)
Put the object in the drawer	$j_{2}$ (unseen)	4/10 (40%)	7/10 (70%)
Object rearrangement	$j_{6}$ (seen)	4/10 (40%)	8/10 (80%)
Overall	–	13/30 (43.3%)	24/30 (80.0%)

TABLE VII: Real-robot success under persistent locks, 10 trials per condition. Locks are software-enforced at the joint-reference layer.

Explore similar work

Uncovering Vulnerability of Vision-Language-Action Models under Joint-Level Physical Faults
Jun 9, 2026Minsoo Jo, Taeju Kwon, Junha Chun +2Robotic Control Vision-Language-Action Models
Graduate School of Data Science, Seoul National University
ReCoVLA: VLM-Guided Reward Compilation for Failure Recovery in Vision-Language-Action Policies
Jun 8, 2026Haodi Hu, Chung-Ta Huang, Jing Liu +4Robot Failure Recovery Vision-Language-Action Models
University of Southern California, USA · Mitsubishi Electric Research Laboratories (MERL), USA · Harvard University, USA
RePO-VLA: Recovery-Driven Policy Optimization for Vision-Language-Action Models
May 10, 2026Weijia Liufu, Xiaoyu Guo, Ruiyi Chen +16VLM Robustness Robot Failure Recovery
1Sun Yat-sen University · 2South China University of Technology · 3Peng Cheng Laboratory +3

Jun 9, 2026, cs.ROJ/K move · Enter open · S save

Uncovering Vulnerability of Vision-Language-Action Models under Joint-Level Physical Faults

Minsoo Jo, Taeju Kwon, Junha Chun, Youngjoon Jeong+1

Graduate School of Data Science, Seoul National University

Robotic Control Vision-Language-Action Models Residual Policy Learning Robustness of VLA Models

Open paper AskPDF arXiv

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.

Abstract

Explore similar work

Uncovering Vulnerability of Vision-Language-Action Models under Joint-Level Physical Faults

ReCoVLA: VLM-Guided Reward Compilation for Failure Recovery in Vision-Language-Action Policies

RePO-VLA: Recovery-Driven Policy Optimization for Vision-Language-Action Models