Computable Laboratory Representation Enables Verifiable Robotic Science Workflows

Summary

The paper proposes a computable representation of the physical laboratory to complement machine-readable scientific knowledge. It uses typed research objects, capability-bound operations, and a compositional workflow algebra to express experiments as programs over changing laboratory states. The framework represents dependencies, decisions, iteration, and concurrency explicitly. The authors implemented it in a modular agentic robotic laboratory by connecting formal operations to executable Function Skills. For varied scientific intents, the system generated workflows relative to the laboratory’s available capabilities. Stateful simulation then tracked object transformations and checked operation preconditions and laboratory constraints before dispatch. The authors present this interface as a foundation for connecting agent reasoning with physical transformations in end-to-end autonomous scientific discovery.