EvoSim is a self‑evolving AI scientist for physical modeling. It leverages discrepancies between experimental observations and model predictions to revise mechanisms and governing equations, while using held‑out experimental data to test physical plausibility. Each exploration updates the knowledge base, skill set, and multi‑agent orchestration, enabling co‑evolution of the model structure and the AI itself.
The system was evaluated on two industrial battery‑modeling tasks:
- Predicting the onset of lithium‑metal plating across 25–45 °C and 2–6 C with a mean absolute error of 1.79% in state of charge.
- Dynamic voltage prediction under vehicle driving conditions achieved a root‑mean‑square error of 7.62 mV, surpassing the accuracy reported for models built by human experts.
Self‑evolution reduced model and physics errors by roughly 36% relative to a baseline, demonstrating enhanced scientific modeling capability. EvoSim turns experimental observations into validated models and accumulates research expertise.
Review