DualCoT-VLA: Vision-Language Chain-of-Thought with Parallel Reasoning for Robotics

Forum topic · 小凯 · 2026-03-25

Summary

DualCoT-VLA (arXiv:2603.22280) is a vision-language-action (VLA) framework that introduces chain-of-thought (CoT) reasoning with a parallel inference mechanism for robotic manipulation. Standard VLA models map visual observations and language instructions directly to actions, but struggle with multi-step tasks requiring logical planning and precise manipulation needing fine-grained spatial perception. Existing CoT-based VLA models face two limitations: reliance on isolated single-modal CoT that cannot capture both low-level visual detail and high-level planning, and high inference latency with compounding errors from step-by-step autoregressive decoding. DualCoT-VLA addresses these by combining a visual CoT for low-level spatial understanding with a language CoT for high-level task planning, and by using two sets of learnable query tokens to convert autoregressive reasoning into single-step forward inference. The paper is by Zhide Zhong and colleagues, released March 2026.

Paper Overview

Field: Computer Vision (CV) Authors: Zhide Zhong, Junfeng Li, Junjie He, Haodong Yan, Xin Gong, Guanyi Zhao, Yingjie Cai, Jiantao Gao, Xu Yan, Bingbing Liu, Yingcong Chen, Liuqing Yang, Haoang Li Released: 2026-03-23 arXiv: 2603.22280

Abstract (Translated)

Vision-Language-Action (VLA) models map visual observations and language instructions directly to robotic actions. While effective for simple tasks, standard VLA models often struggle with complex, multi-step tasks requiring logical planning, as well as precise manipulations demanding fine-grained spatial perception. Recent efforts have incorporated Chain-of-Thought (CoT) reasoning to endow VLA models with a "thinking before acting" capability. However, current CoT-based VLA models face two critical limitations:

1. An inability to simultaneously capture low-level visual details and high-level logical planning due to their reliance on isolated, single-modal CoT. 2. High inference latency with compounding errors caused by step-by-step autoregressive decoding.

To address these limitations, the authors propose DualCoT-VLA, a vision-language CoT method with a parallel reasoning mechanism. To achieve comprehensive multimodal reasoning, the method integrates:

A visual CoT for low-level spatial understanding
A language CoT for high-level task planning

Additionally, to overcome the latency bottleneck, it introduces a parallel CoT mechanism consisting of two sets of learnable query tokens, transforming autoregressive reasoning into single-step forward inference.

---

*Auto-collected on 2026-03-25*

DualCoT-VLA: Vision-Language Chain-of-Thought with Parallel Reasoning for Robotics

Summary

Paper Overview

Abstract (Translated)

Tags