Paper 2601.09708
Fast-ThinkAct: Efficient Vision-Language-Action Reasoning via Verbalizable Latent Planning
- Published
- Jan 2026
- Research lab
- NVIDIA
- Citations
- 17
- GitHub
- Not linked
01 In brief
Summary
Fast-ThinkAct is an efficient reasoning framework for Vision-Language-Action (VLA) tasks that compresses lengthy chain-of-thought (CoT) reasoning into compact latent representations.
It uses a teacher-student distillation approach, where a textual teacher VLM (trained with GRPO) provides reasoning traces, and a latent student VLM learns to generate continuous latent tokens.
A verbalizer LLM decodes these latents into text, enabling preference-guided distillation (using DPO-style loss) to favor high-quality reasoning.
Additionally, action-aligned visual plan distillation transfers spatial reasoning by aligning hidden states and using spatial tokens for parallel waypoint prediction.
The student's latent planning conditions a diffusion-based action model for policy learning.
Experiments on LIBERO, SimplerEnv, and RoboTwin2.0 show Fast-ThinkAct achieves up to 89.3% latency reduction (e.g., 805ms vs.
7513ms for ThinkAct-7B) while improving success rates (e.g., 89.7% on LIBERO vs.
84.4% for ThinkAct-7B).
It also outperforms baselines on embodied reasoning benchmarks (EgoPlan-Bench2, RoboVQA, OpenEQA) and demonstrates long-horizon planning, failure recovery, and few-shot adaptation.
Ablations confirm the importance of the verbalizer and distillation losses.
02 From the paper
Abstract
Vision-Language-Action (VLA) tasks require reasoning over complex visual scenes and executing adaptive actions in dynamic environments. While recent studies on reasoning VLAs show that explicit chain-of-thought (CoT) can improve generalization, they suffer from high inference latency due to lengthy reasoning traces. We propose Fast-ThinkAct, an efficient reasoning framework that achieves compact yet performant planning through verbalizable latent reasoning. Fast-ThinkAct learns to reason efficiently with latent CoTs by distilling from a teacher, driven by a preference-guided objective to align manipulation trajectories that transfers both linguistic and visual planning capabilities for embodied control. This enables reasoning-enhanced policy learning that effectively connects compact reasoning to action execution. Extensive experiments across diverse embodied manipulation and reasoning benchmarks demonstrate that Fast-ThinkAct achieves strong performance with up to 89.3% reduced inference latency over state-of-the-art reasoning VLAs, while maintaining effective long-horizon planning, few-shot adaptation, and failure recovery.