Think Less, Act Early: Reinforced Latent Reasoning with Early Exit in Vision-Language-Action Models

2026-06-13

AI Summary

This work addresses the limitations of existing vision-language-action (VLA) models, which rely on explicit chain-of-thought reasoning that incurs high computational overhead and is prone to error propagation in multi-step tasks. The authors propose AVA-VLA, a novel framework that formulates reasoning as a sequence of latent, unobserved variables, thereby eliminating explicit textual generation. AVA-VLA integrates a reinforcement learning–driven denoising mechanism with a state-confidence–based adaptive early-exit strategy to dynamically balance reasoning depth and execution efficiency. Evaluated on the LIBERO benchmark, the method achieves an average success rate of 98.3% while accelerating inference by 6× compared to explicit chain-of-thought approaches, substantially enhancing both the efficiency and robustness of decision-making.

Abstract

Existing Vision-Language-Action (VLA) models predominantly rely on explicit Chain-of-Thought (CoT) reasoning to bridge perception and action. While effective, this paradigm suffers from high computational costs and error propagation in multi-step tasks. In this paper, we propose Adaptive Variable Alignment VLA (AVA-VLA), a novel Latent Reasoning VLA framework that models reasoning as a sequence of unobservable latent variables, bypassing the need for explicit text generation. However, latent trajectories are inherently susceptible to noise interference and misalignment with downstream objectives. To address this, we introduce a Reinforcement Learning-based Denoising mechanism that treats latent state generation as a sequential decision process, optimizing reasoning trajectories via task-level rewards. Furthermore, we incorporate an Early-Exit Strategy that adaptively terminates reasoning based on state confidence, enabling a dynamic trade-off between depth and efficiency. Extensive experiments on embodied decision benchmarks demonstrate that AVA-VLA achieves a 6x inference speedup over explicit CoT methods while attaining a 98.3% average success rate on LIBERO, improving both efficiency and long-horizon stability over full-reasoning baselines.

Problem

• Vision-Language-Action
• Chain-of-Thought reasoning
• computational cost
• error propagation
• multi-step tasks

Innovation

• Latent Reasoning
• Reinforcement Learning
• Early Exit
• Vision-Language-Action Models
• Chain-of-Thought