AC^2-VLA: Action-Context-Aware Adaptive Computation in Vision-Language-Action Models for Efficient Robotic Manipulation

TL;DR

AC^2-VLA introduces an action-context-aware adaptive computation framework that significantly reduces inference latency and computational cost in vision-language-action models for robotic manipulation, maintaining high task success rates.

Contribution

The paper proposes a novel adaptive computation framework that conditions on action context, enabling efficient inference in VLA models for robotics, with a new training scheme for structured sparsification.

Findings

Achieves up to 1.79× speedup in inference.

Reduces FLOPs to 29.4% of dense baseline.

Maintains comparable task success rates.

Abstract

Vision-Language-Action (VLA) models have demonstrated strong performance in robotic manipulation, yet their closed-loop deployment is hindered by the high latency and compute cost of repeatedly running large vision-language backbones at every timestep. We observe that VLA inference exhibits structured redundancies across temporal, spatial, and depth dimensions, and that most existing efficiency methods ignore action context, despite its central role in embodied tasks. To address this gap, we propose Action-Context-aware Adaptive Computation for VLA models (AC^2-VLA), a unified framework that conditions computation on current visual observations, language instructions, and previous action states. Based on this action-centric context, AC^2-VLA adaptively performs cognition reuse across timesteps, token pruning, and selective execution of model components within a unified mechanism. To train the adaptive policy, we introduce an action-guided self-distillation scheme that preserves the behavior of the dense VLA policy while enabling structured sparsification that transfers across tasks and settings. Extensive experiments on robotic manipulation benchmarks show that AC^2-VLA achieves up to a 1.79\times speedup while reducing FLOPs to 29.4% of the dense baseline, with comparable task success.