ThinkAct: Vision-Language-Action Reasoning via Reinforced Visual Latent Planning

TL;DR

ThinkAct introduces a dual-system framework that combines high-level visual reasoning with low-level action execution using reinforced latent planning, significantly improving adaptability and planning in embodied AI tasks.

Contribution

It presents a novel visual latent planning approach that enhances reasoning and planning capabilities in vision-language-action tasks through reinforcement learning.

Findings

Enables few-shot adaptation in complex tasks

Supports long-horizon planning and self-correction

Improves robustness in embodied AI environments

Abstract

Vision-language-action (VLA) reasoning tasks require agents to interpret multimodal instructions, perform long-horizon planning, and act adaptively in dynamic environments. Existing approaches typically train VLA models in an end-to-end fashion, directly mapping inputs to actions without explicit reasoning, which hinders their ability to plan over multiple steps or adapt to complex task variations. In this paper, we propose ThinkAct, a dual-system framework that bridges high-level reasoning with low-level action execution via reinforced visual latent planning. ThinkAct trains a multimodal LLM to generate embodied reasoning plans guided by reinforcing action-aligned visual rewards based on goal completion and trajectory consistency. These reasoning plans are compressed into a visual plan latent that conditions a downstream action model for robust action execution on target environments. Extensive experiments on embodied reasoning and robot manipulation benchmarks demonstrate that ThinkAct enables few-shot adaptation, long-horizon planning, and self-correction behaviors in complex embodied AI tasks.