An agent that couples a VLM policy with world-model rollouts in a learned physics-code action space outperforms GPT-5 and existing RL/world-model baselines on 200 games and transfers zero-shot to unseen games.
Do as i can, not as i say: Grounding language in robotic affordances
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.AI 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
IPR-1: Interactive Physical Reasoner
An agent that couples a VLM policy with world-model rollouts in a learned physics-code action space outperforms GPT-5 and existing RL/world-model baselines on 200 games and transfers zero-shot to unseen games.