VisualThink-VLA: Visual Intermediate Reasoning for Effective and Low-Latency Vision-Language-Action Policies

2026-05-28

AI Summary

Existing vision–language–action policies rely on textual intermediate reasoning, which is prone to informational interference and incurs high latency, making them unsuitable for real-time embodied control. This work proposes the first framework based on visual intermediate reasoning, replacing textual chain-of-thought with a compact visual evidence interface to guide action prediction. This approach preserves spatial precision while eliminating the computational overhead of text decoding, and incorporates a selective routing mechanism to enhance efficiency. The authors introduce VisualEvidence-Kit, a supervised resource comprising the VisualEvidence-Agent and VisualEvidence-Set datasets. Experiments demonstrate that the method achieves state-of-the-art success rates across multiple simulated and real-world robotic benchmarks, reducing inference latency from several seconds to sub-second levels—for instance, accelerating inference by 22.8× on BridgeData V2 (from 8.377 s to 0.367 s).

Abstract

Recent work has begun to equip vision-language-action (VLA) policies with explicit intermediate reasoning. In embodied control, however, textual chain-of-thought is a poor fit: irrelevant or weakly textual information can interfere with action prediction, while autoregressive text decoding adds too much latency for real-time closed-loop execution. We present VISUALTHINK-VLA, a visual intermediate-reasoning framework for accurate, low-latency VLA policies. Our bootstrapping philosophy is to guide action with effective visual thinking: VISUALTHINK-VLA bootstraps action prediction through a compact visual-evidence interface that preserves spatial precision while avoiding decoding overhead. Besides, to further improve performance and efficiency, VISUALTHINK-VLA adopts a tailored selective routing mechanism to learn the visual evidence tokens, enabling low-latency inference while preserving high-capacity specialization. We also introduce VisualEvidence-Kit, a supervision-and-audit resource centered on a VisualEvidence-Agent that constructs a 754.7k VLA instructions VisualEvidence-Set for route supervision and counterfactual faithfulness tests. Across multiple benchmarks and real-robot evaluation, VISUALTHINK-VLA achieves the highest success rate on most benchmarks while reducing the multi-second latency of reasoning-augmented baselines to the sub-second regime. For example, on BridgeData V2, it reduces step latency from 8.377,s with ECoT to 0.367,s, achieving a 22.8 times speedup.

Problem

• vision-language-action
• intermediate reasoning
• latency
• embodied control
• real-time execution

Innovation

• visual intermediate reasoning
• vision-language-action policies
• selective routing
• low-latency inference
• VisualEvidence-Kit