VLA-ATTC: Adaptive Test-Time Compute for VLA Models with Relative Action Critic Model

2026-05-01

AI Summary

This work addresses the tendency of existing Vision-Language-Action (VLA) models to generate suboptimal or catastrophic actions in complex or ambiguous scenarios due to the absence of deliberative reasoning mechanisms. To mitigate this limitation, the authors propose the VLA-ATTC framework, which incorporates an uncertainty-based “cognitive clutch” to dynamically trigger test-time computation (TTC). The framework further introduces a relative action critic that evaluates candidate actions through pairwise comparisons, circumventing the instability inherent in absolute value estimation. Coupled with efficient action sampling and an automated preference data generation pipeline that requires no human annotations, the approach enables adaptive inference. Evaluated on the LIBERO-LONG benchmark, VLA-ATTC reduces the failure rate of the current state-of-the-art model PI0.5 by over 50%, substantially enhancing decision robustness.

Abstract

Vision-Language-Action (VLA) models have demonstrated remarkable capabilities and generalization in embodied manipulation. However, their decision-making relies on a fast, instinctive process that lacks deliberation. This strategy often leads to suboptimal or catastrophic actions when facing complex or ambiguous scenarios that require greater consideration. In this paper, we introduce \textbf{VLA-ATTC}, a framework that endows VLA models with adaptive test-time compute (TTC). VLA-ATTC employs an uncertainty-based ``cognitive clutch'' to dynamically transition from reflexive execution to a TTC deliberation phase when necessary. During TTC phase, a novel \textbf{Relative Action Critic} (RAC) model identifies the optimal action from generated candidates via pairwise comparisons. This relative mechanism replaces unstable absolute value estimation, significantly simplifying the learning objective. Furthermore, we introduce an efficient sampling strategy to amortize computational costs and an automated data pipeline that curates preference pairs without manual annotation. On the LIBERO-LONG benchmark, VLA-ATTC reduces the failure rate of the SOTA model PI0.5 by over 50\%. We will open-source all the code and weights.

Problem

• Vision-Language-Action
• test-time compute
• deliberation
• embodied manipulation
• decision-making

Innovation

• Adaptive Test-Time Compute
• Relative Action Critic
• Vision-Language-Action Models
• Uncertainty-based Cognitive Clutch
• Preference Learning