# PhysCodeBench: Benchmarking Physics-Aware Symbolic Simulation of 3D Scenes via Self-Corrective Multi-Agent Refinement

arXiv: https://arxiv.org/abs/2604.23580 · HTML: https://mlsanity.com/paper/2604.23580/

## Key facts

- Cited by 10 papers in the map (data through 14 Oct 2026).
- Placed in Multimodal and embodied agents, one of 16 directions in LLM agents and tool use.
- Builds on 31 of its 46 references that are in the map.
- First posted on arXiv 26 Apr 2026.

## Placed in

- LLM agents and tool use / [Multimodal and embodied agents](https://mlsanity.com/agents/multimodal-and-embodied-agents/)

## Builds on (most cited first, 30 of 31)

- [Training language models to follow instructions with human feedback](https://mlsanity.com/paper/2203.02155/) (2022)
- [Evaluating Large Language Models Trained on Code](https://mlsanity.com/paper/2107.03374/) (2021)
- [Direct Preference Optimization: Your Language Model is Secretly a Reward Model](https://mlsanity.com/paper/2305.18290/) (2023)
- [Qwen2.5 Technical Report](https://mlsanity.com/paper/2412.15115/) (2024)
- [GPT-4o System Card](https://mlsanity.com/paper/2410.21276/) (2024)
- [Program Synthesis with Large Language Models](https://mlsanity.com/paper/2108.07732/) (2021)
- [CLIPScore: A Reference-free Evaluation Metric for Image Captioning](https://mlsanity.com/paper/2104.08718/) (2021)
- [RAFT: Recurrent All-Pairs Field Transforms for Optical Flow](https://mlsanity.com/paper/2003.12039/) (2020)
- [Code Llama: Open Foundation Models for Code](https://mlsanity.com/paper/2308.12950/) (2023)
- [AutoGen: Enabling Next-Gen LLM Applications via Multi-Agent Conversation](https://mlsanity.com/paper/2308.08155/) (2023)
- [VBench: Comprehensive Benchmark Suite for Video Generative Models](https://mlsanity.com/paper/2311.17982/) (2023)
- [Code as Policies: Language Model Programs for Embodied Control](https://mlsanity.com/paper/2209.07753/) (2022)
- [Teaching Large Language Models to Self-Debug](https://mlsanity.com/paper/2304.05128/) (2023)
- [ChatDev: Communicative Agents for Software Development](https://mlsanity.com/paper/2307.07924/) (2023)
- [VoxPoser: Composable 3D Value Maps for Robotic Manipulation with Language Models](https://mlsanity.com/paper/2307.05973/) (2023)
- [WizardCoder: Empowering Code Large Language Models with Evol-Instruct](https://mlsanity.com/paper/2306.08568/) (2023)
- [Eureka: Human-Level Reward Design via Coding Large Language Models](https://mlsanity.com/paper/2310.12931/) (2023)
- [CLEVRER: CoLlision Events for Video REpresentation and Reasoning](https://mlsanity.com/paper/1910.01442/) (2019)
- [CodeBLEU: a Method for Automatic Evaluation of Code Synthesis](https://mlsanity.com/paper/2009.10297/) (2020)
- [PDEBENCH: An Extensive Benchmark for Scientific Machine Learning](https://mlsanity.com/paper/2210.07182/) (2022)
- [RoboGen: Towards Unleashing Infinite Data for Automated Robot Learning via Generative Simulation](https://mlsanity.com/paper/2311.01455/) (2023)
- [Physion: Evaluating Physical Prediction from Vision in Humans and Machines](https://mlsanity.com/paper/2106.08261/) (2021)
- [SceneCraft: An LLM Agent for Synthesizing 3D Scene as Blender Code](https://mlsanity.com/paper/2403.01248/) (2024)
- [GenSim: Generating Robotic Simulation Tasks via Large Language Models](https://mlsanity.com/paper/2310.01361/) (2023)
- [Mind's Eye: Grounded Language Model Reasoning through Simulation](https://mlsanity.com/paper/2210.05359/) (2022)
- [PHYBench: Holistic Evaluation of Physical Perception and Reasoning in Large Language Models](https://mlsanity.com/paper/2504.16074/) (2025)
- [CodeChain: Towards Modular Code Generation Through Chain of Self-revisions with Representative Sub-modules](https://mlsanity.com/paper/2310.08992/) (2023)
- [Physion++: Evaluating Physical Scene Understanding that Requires Online Inference of Different Physical Properties](https://mlsanity.com/paper/2306.15668/) (2023)
- [SimGen: Simulator-conditioned Driving Scene Generation](https://mlsanity.com/paper/2406.09386/) (2024)
- [LLMPhy: Parameter-Identifiable Physical Reasoning Combining Large Language Models and Physics Engines](https://mlsanity.com/paper/2411.08027/) (2024)

## Built on by (most cited first, 10 of 10)

- [Code as Worlds: Agentic Discovery of Executable World Representations for Physical Reasoning](https://mlsanity.com/paper/2608.27549/) (2026)
- [MPMWorlds: Material-Point-Method Simulations for Inferring and Extrapolating Physical Dynamics](https://mlsanity.com/paper/2606.01538/) (2026)
- [4DCodeBench: Benchmarking Agents on Inverse Graphics of Dynamic Scenes](https://mlsanity.com/paper/2610.03715/) (2026)
- [ChronoAgentic: A Code-based Multi-Agent World Simulator for Physically Grounded Simulation Construction](https://mlsanity.com/paper/2605.14398/) (2026)
- [PACE-Bench: Benchmarking Physics Adaptation via Code Evolution in Dynamic Environments](https://mlsanity.com/paper/2608.14441/) (2026)
- [Agentic Visual Generation: From Generative Models to Agentic Control](https://mlsanity.com/paper/2609.06758/) (2026)
- [WorldSolver: Can LLM Agents Simulate the Physical Dynamics via Solver Generation?](https://mlsanity.com/paper/2610.08720/) (2026)
- [ExeWorldBench: Benchmarking Code-Based Executable 3D World Modeling](https://mlsanity.com/paper/2610.15166/) (2026)
- [PhySimCode: A Benchmark and Evaluation Method for Physics Video to Code Generation](https://mlsanity.com/paper/2610.15722/) (2026)
- [PhysCodeGym: Benchmarking Coding Agents for Physically Verifiable 4D Reconstruction](https://mlsanity.com/paper/2610.17288/) (2026)

## Abstract (arXiv)

Translating natural-language descriptions of physical phenomena into executable simulation code requires both programming expertise and physical reasoning. Current large language models (LLMs) lack this combination: they frequently produce code that runs but simulates the wrong physics. We introduce PhysCodeBench, the first benchmark for this task, with 1,200 expert-validated examples spanning four physical domains. Its evaluation suite, PhysCodeEval, goes beyond executability and visual fidelity to measure physical correctness directly from the engine state via conservation-law residuals and expert-written assertions, and supports cross-engine evaluation to disentangle physics reasoning from API fluency. As a reference method, we propose the Self-Corrective Multi-Agent Refinement Framework (SMRF), which decouples physics-aware error correction from code generation through specialized agents. This design is motivated by our finding that targeted correction, rather than generic iterative refinement, is the key driver of physical accuracy. SMRF nearly triples the physical-assertion pass rate of the best proprietary baseline (70.6% vs. 23.8%) and retains its advantage under cross-engine transfer.