Search papers, labs, and topics across Lattice.
This paper introduces VISUALTHINK-VLA, a framework for vision-language-action policies that uses visual intermediate reasoning to improve accuracy and reduce latency. It bootstraps action prediction through a compact visual-evidence interface and employs a selective routing mechanism to learn relevant visual evidence tokens. Experiments across benchmarks and real-robot settings demonstrate that VISUALTHINK-VLA achieves state-of-the-art success rates with sub-second latency, significantly outperforming text-based chain-of-thought methods.
Ditch slow, irrelevant text-based reasoning: VISUALTHINK-VLA uses visual tokens to speed up vision-language-action policies by 22x while boosting accuracy.
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.
