ThinkWorld trains robot action reasoning from unlabeled human video
Original title: ThinkWorld: World Imagination Reinforces Chain-of-Dynamics Reasoning
Chain-of-Dynamics tokens let human and robot video supervise the same action-reasoning space.
ThinkWorld is a vision-language-action framework that represents each action thought as a Chain-of-Dynamics, discrete motion tokens linking the current world to a future state. It initializes those tokens from labeled demonstrations, predicts the induced future, and reinforces predictions that match observed outcomes. Unlabeled human and robot video can then supervise the same space. Simulation and real-robot experiments show human-to-robot transfer and generalization to unseen tasks and environments.
- CoD is intended as a shared action representation across embodiments.
- Videos without action labels supply supervision from observed state changes.
- Reinforcement uses imagined world outcomes to strengthen the reasoning.
- The authors report tests in simulation and on real robots.
From the original article
1 min read at arxiv.orgVision-Language-Action (VLA) models have made substantial progress in robotic manipulation, but their generalization remains constrained by the diversity and coverage of action-labeled robot demonstrations. Human videos provide much broader behavioral and environmental experience, while learning from them faces a fundamental trade-off between scalable data and explicit action grounding.
By Chaehyun Kim · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.