Test-Time Adaptation of Manipulation Policies Under Actuator Degradation

2026-09-28

AI Summary

This study addresses the failure of pretrained robot policies caused by hardware degradation, such as motor overheating. To mitigate this issue, we propose TeAR, a framework that pioneers the use of onboard multimodal telemetry signals—temperature, current, and voltage—for action correction. Specifically, TeAR employs a lightweight Transformer to fuse telemetry features in real time, rectifying the outputs of a frozen policy and enabling general test-time adaptation decoupled from the underlying policy. This approach allows for online compensation of hardware degradation without requiring retraining or fine-tuning. Experimental evaluations demonstrate that TeAR improves simulation success rates by 31.8% over baselines and yields a 10–15% increase in task success on physical robots operating under heated conditions.

Abstract

Robot manipulation policies are usually trained under the assumption that a commanded action produces the same motion as it did during training even after hours of operation. Real hardware violates this assumption as the motors gradually heat up, current saturates near contact, voltage sags under load, thus the same policy action can produce a weaker, delayed, or noisier motion. These conditions are already measured by onboard telemetry, such as joint temperature, motor current, and supply voltage, yet this signal is typically used only for logging or safety checks rather than policy adaptation. We introduce Telemetry-Aware Action Rectification (TeAR), a policy-agnostic method that turns a frozen manipulation policy into a telemetry-conditioned policy by rectifying its outgoing action before it reaches the low-level controller. TeAR learns a lightweight Transformer that combines the proposed action with live actuator telemetry and amplifies, damps, or biases individual action components. We evaluate TeAR across 18 policy-task pairs spanning 8 policy families and 5 manipulation tasks. In an additional paired evaluation with degradation-model mismatch, TeAR achieves 31.8% success, compared with 25.6% for the base policy and 30.6% for an assumed-model inverse. On a physical arm, TeAR improves success under heating by 10-15% without on-robot fine-tuning.

Problem

• Test-Time Adaptation
• Actuator Degradation
• Robot Manipulation
• Telemetry-Aware

Innovation

• Test-Time Adaptation
• Telemetry-Aware Action Rectification
• Actuator Degradation
• Robot Manipulation
• Lightweight Transformer