ThinkingVLA: Interleaved Vision and Language Reasoning for Robotic Manipulation

2026-06-16

AI Summary

Existing vision-language-action models lack explicit reasoning capabilities, rendering them ill-suited for long-horizon robotic manipulation tasks that demand complex planning. This work proposes a unified autoregressive architecture that, for the first time, integrates forward subgoal reasoning with goal-conditioned inverse dynamics inference within a single framework. The model enables deep multimodal reasoning by interleaving the generation of textual and visual representations: it first employs chain-of-thought reasoning to predict future visual states and then retroactively infers action sequences from these predicted images. Evaluated on both simulated and real-world benchmarks, the proposed method substantially outperforms current state-of-the-art approaches, achieving particularly pronounced gains in long-horizon tasks.

Abstract

Most Vision-Language-Action (VLA) models map observations directly to actions without explicit reasoning, limiting their capacity for reasoning-intensive long-horizon tasks. To address this, existing approaches adopt Chain-of-Thought (CoT) reasoning to enable subgoal decomposition and spatial anticipation. However, those methods lack a unified architecture for effective cross-modal reasoning and fail to explicitly include inverse reasoning ability based on the target state. We argue that manipulation planning naturally decomposes into prediction, anticipating the next visual state, and inverse dynamics, inferring the actions to reach it. Bridging both requires a unified autoregressive architecture that interleaves textual and visual reasoning in a single generation process. We propose \textbf{ThinkingVLA}, a generative model that realizes this decomposition within a unified Mixture-of-Transformers architecture. ThinkingVLA consists of a forward CoT that identifies the immediate subgoal and guides the visual forecasting; the predicted image then serves as the target state, grounding an inverse CoT that reasons about spatial relationships and action intent based on the predicted image; and the final action is generated conditioned on this full reasoning context. Extensive experiments on simulation and real-world benchmarks demonstrate that ThinkingVLA consistently outperforms state-of-the-art baselines, with particularly large gains on long-horizon manipulation tasks.

Problem

• Vision-Language-Action
• Chain-of-Thought reasoning
• inverse reasoning
• long-horizon manipulation
• cross-modal reasoning

Innovation

• Vision-Language-Action
• Chain-of-Thought Reasoning
• Inverse Dynamics
• Autoregressive Generation
• Mixture-of-Transformers