Research map / Large language models / Agents, code and math
Grounding LLMs in Scientific Discovery via Embodied Actions
Bo Zhang, Jinfeng Zhou, Yuxuan Chen, Jianing Yin, Minlie Huang, Hongning Wang
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add Reflective Decision Maker (ablation) | EmbodiedAct (GPT-5.2) · EngDesign Core | average score: 65.0 -> 70.6 | Improved |
| add Runtime Perception engine (ablation) | EmbodiedAct (GPT-5.2) · EngDesign Core | average score: 56.5 -> 70.6 | Improved |
| add Runtime Perception engine (ablation) | EmbodiedAct (GPT-5-mini) · EngDesign Core | average score: 48.5 -> 59.3 | Improved |
| add both Reflective Decision Maker and Runtime Perception (ablation) | EmbodiedAct (GPT-5.2) · EngDesign Core | average score: 51.5 -> 70.6 | Improved |
| add Reflective Decision Maker (ablation) | EmbodiedAct (GPT-5-mini) · SciBench-107 | accuracy (%): 48.0 -> 49.5 | Improved |
| add Runtime Perception engine (ablation) | EmbodiedAct (GPT-5.2) · SciBench-107 | accuracy (%): 50.0 -> 54.2 | Improved |
| use EmbodiedAct instead of CodeAct | GPT-5.2 · EngDesign Core | average score: 55.4 -> 70.6 | Improved |
| use EmbodiedAct instead of CodeAct | GPT-5.2 · EngDesign Extended | average score: 49.4 -> 65.4 | Improved |
| use EmbodiedAct instead of CodeAct | GPT-5.2 · SciBench-107 | accuracy (%): 49.53 -> 54.21 | Improved |
| use EmbodiedAct instead of generative baseline | Qwen3-235B-Instruct · EngDesign Core | average score: 39.6 -> 57.0 | Improved |
| use EmbodiedAct instead of CodeAct | Qwen3-235B-Instruct · SciBench-107 | accuracy (%): 55.14 -> 52.34 | Hurt |
| use EmbodiedAct instead of CodeAct | Qwen3-8B-VL-Instruct · EngDesign Core | average score: 38.9 -> 40.8 | Mixed |
Closest papers by meaning
- PhysGym: Benchmarking LLMs in Interactive Physics Discovery with Controlled Priors (NeurIPS 2025 · similarity 0.89)
- ScienceBoard: Evaluating Multimodal Autonomous Agents in Realistic Scientific Workflows (ICLR 2026 · similarity 0.89)
- LLM and Simulation as Bilevel Optimizers: A New Paradigm to Advance Physical Scientific Discovery (ICML 2024 · similarity 0.88)
- LabUtopia: High-Fidelity Simulation and Hierarchical Benchmark for Scientific Embodied Agents (NeurIPS 2025 · similarity 0.88)
- Embodied Agent Interface: Benchmarking LLMs for Embodied Decision Making (NeurIPS 2024 · similarity 0.88)
- SciAgentGym: Benchmarking Multi-Step Scientific Tool-Use in LLM Agents (ICML 2026 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
