# PACE-Bench: Benchmarking Physics Adaptation via Code Evolution in Dynamic Environments

arXiv: https://arxiv.org/abs/2608.14441 · HTML: https://mlsanity.com/paper/2608.14441/

## Key facts

- Cited by 2 papers in the map (data through 14 Oct 2026).
- Builds on 18 of its 28 references that are in the map.
- First posted on arXiv 14 Aug 2026.

## Builds on (most cited first, 18 of 18)

- [ExpeL: LLM Agents Are Experiential Learners](https://mlsanity.com/paper/2308.10144/) (2023)
- [Agentic Context Engineering: Evolving Contexts for Self-Improving Language Models](https://mlsanity.com/paper/2510.04618/) (2025)
- [RAGEN: Understanding Self-Evolution in LLM Agents via Multi-Turn Reinforcement Learning](https://mlsanity.com/paper/2504.20073/) (2025)
- [Absolute Zero: Reinforced Self-play Reasoning with Zero Data](https://mlsanity.com/paper/2505.03335/) (2025)
- [A Comprehensive Survey of Self-Evolving AI Agents: A New Paradigm Bridging Foundation Models and Lifelong Agentic Systems](https://mlsanity.com/paper/2508.07407/) (2025)
- [Learning to Discover at Test Time](https://mlsanity.com/paper/2601.16175/) (2026)
- [CausalWorld: A Robotic Manipulation Benchmark for Causal Structure and Transfer Learning](https://mlsanity.com/paper/2010.04296/) (2020)
- [LifelongAgentBench: Evaluating LLM Agents as Lifelong Learners](https://mlsanity.com/paper/2505.11942/) (2025)
- [SkillLearnBench: Benchmarking Continual Learning Methods for Agent Skill Generation on Real-World Tasks](https://mlsanity.com/paper/2604.20087/) (2026)
- [StreamBench: Towards Benchmarking Continuous Improvement of Language Agents](https://mlsanity.com/paper/2406.08747/) (2024)
- [NewtonBench: Benchmarking Generalizable Scientific Law Discovery in LLM Agents](https://mlsanity.com/paper/2510.07172/) (2025)
- [CARL: A Benchmark for Contextual and Adaptive Reinforcement Learning](https://mlsanity.com/paper/2110.02102/) (2021)
- [Toward Engineering AGI: Benchmarking the Engineering Design Capabilities of LLMs](https://mlsanity.com/paper/2509.16204/) (2025)
- [AutoEnv: Automated Environments for Measuring Cross-Environment Agent Learning](https://mlsanity.com/paper/2511.19304/) (2025)
- [Frontier-Eng: Benchmarking Self-Evolving Agents on Real-World Engineering Tasks with Generative Optimization](https://mlsanity.com/paper/2604.12290/) (2026)
- [EvoAgentBench: Benchmarking Agent Self-Evolution via Ability Transfer](https://mlsanity.com/paper/2607.05202/) (2026)
- [PhysCodeBench: Benchmarking Physics-Aware Symbolic Simulation of 3D Scenes via Self-Corrective Multi-Agent Refinement](https://mlsanity.com/paper/2604.23580/) (2026)
- [SE-Bench: Benchmarking Self-Evolution with Knowledge Internalization](https://mlsanity.com/paper/2602.04811/) (2026)

## Built on by (most cited first, 2 of 2)

- [In-Context Learning for Robots: Methods and Applications](https://mlsanity.com/paper/2609.36012/) (2026)
- [ExeWorldBench: Benchmarking Code-Based Executable 3D World Modeling](https://mlsanity.com/paper/2610.15166/) (2026)

## Abstract (arXiv)

Self-evolving agents improve future behavior from interaction experience, yet existing evaluations typically optimize under fixed execution conditions and do not test recovery after those conditions change. To address this gap, we introduce PACE-Bench (Physics Adaptation via Code Evolution), a simulator-grounded benchmark of 144 source-to-target adaptation pairs across six physics domains. Each pair links a source environment to a mutated target environment with the same goal and interface. A code-driven design that succeeds in the source fails in the target, where agents must iteratively adapt it into a working target design using diagnostic sandbox feedback within a limited attempt budget. We compare ten self-evolving methods from four paradigms. The benchmark remains far from saturated: Reflexion + Qwen3-14B succeeds on only 35.9% of full-benchmark pairs, while GPT-5.5 solves 66.7% of the Statics subset under the full budget. Together, these results show that simulator-grounded reflection is more reliable than unverified self-revision, while memory anchors agents to early designs and broad tree search explores without converging. Even revealing exact physical changes does not raise the performance ceiling, pointing to mechanism redesign rather than parameter inference as the central bottleneck. Data and code are available at https://github.com/thunlp/PACE-Bench.