LLMPhy uses iterative LLM-generated programs executed in physics engines to solve continuous parameter estimation and discrete scene layout problems, outperforming prior black-box methods on three new zero-shot physical reasoning datasets.
The last attempt modified the parameters to: - ’sliding-friction’: 0.2 - ’armature’: 0.3 - ’stiffness’: 0.4 - ’damping’: 6.5
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2024 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
LLMPhy: Parameter-Identifiable Physical Reasoning Combining Large Language Models and Physics Engines
LLMPhy uses iterative LLM-generated programs executed in physics engines to solve continuous parameter estimation and discrete scene layout problems, outperforming prior black-box methods on three new zero-shot physical reasoning datasets.