Reproduce “Experiment-as-Code Labs: A Declarative Stack for AI-Driven Scientific Discovery” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Zhenning Yang, Yuhan Chen, Patrick Tser Jern Kon, Tongyuan Miao, Hongyi Lin, Venkat Viswanathan, Danai Koutra, Ang Chen
Researchers from the University of Michigan introduce Experiment-as-Code (EaC) Labs, a declarative three-layer stack designed to enable AI agents to safely and reproducibly control real-world physical laboratories. The system was successfully deployed in an autonomous battery lab, demonstrating enhanced workflow composition, efficiency gains through selective experimental specification, and scientifically validated results in an ionic conductivity experiment.
Key insights
- Adopting a declarative approach, similar to Infrastructure-as-Code, is crucial for standardizing experimental definitions, enhancing reproducibility, and abstracting the complexities of physical lab instruments for AI agents.
- A unified, real-time lab-state model is essential for the Execution layer to enforce safety, handle device failures, manage calibration drift, and ensure reliable experiment outcomes.
- Intelligent orchestration of heterogeneous, shared resources, informed by the lab-state model, can significantly improve experimental throughput, reduce waiting times, and avoid redundant work in autonomous labs.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.