DreamAvoid: Critical-Phase Test-Time Dreaming to Avoid Failures in VLA Policies

TL;DR

DreamAvoid is a novel framework that enables vision-language-action models to anticipate and avoid failures during critical phases of manipulation tasks by using test-time dreaming and boundary learning.

Contribution

It introduces a test-time dreaming approach with a Dream Trigger, Action Proposer, and Dream Evaluator to improve failure avoidance in VLA models.

Findings

Significantly reduces failure rates in real-world manipulation tasks.

Improves success rates on simulation benchmarks.

Effectively anticipates failures through boundary-aware dreaming.

Abstract

Vision-Language-Action (VLA) models are often brittle in fine-grained manipulation, where minor action errors during the critical phases can rapidly escalate into irrecoverable failures. Since existing VLA models rely predominantly on successful demonstrations for training, they lack an explicit awareness of failure during these critical phases. To address this, we propose DreamAvoid, a critical-phase test-time dreaming framework that enables VLA models to anticipate and avoid failures. We also introduce an autonomous boundary learning paradigm to refine the system's understanding of the subtle boundary between success and failure. Specifically, we (1) utilize a Dream Trigger to determine whether the execution has entered a critical phase, (2) sample multiple candidate action chunks from the VLA via an Action Proposer, and (3) employ a Dream Evaluator, jointly trained on mixed data (success, failure, and boundary cases), to "dream" the short-horizon futures corresponding to the candidate actions, evaluate their values, and select the optimal action. We conduct extensive evaluations on real-world manipulation tasks and simulation benchmarks. The results demonstrate that DreamAvoid can effectively avoid failures, thereby improving the overall task success rate. Our code is available at https://github.com/XianzheFan/DreamAvoid.