(1212-A) Advancing Autonomous Discovery: Enhancements to the Workcell Execution Interface

Abstract: We present our progress on the development of a fully automated and reconfigurable laboratory for autonomous discovery with the Workflow Execution Interface (WEI), a software application stack and library developed in the Rapid Prototyping Laboratory at Argonne National Laboratory. WEI coordinates the many instruments and services required for automated experiments and autonomous discovery in the physical sciences.

We have designed WEI as a robust, modular, composable, and extensible toolkit, aiming to provide an approachable option for physical scientists who wish to automate their experiments or develop new autonomous discovery capabilities. To do this, we rely on Docker container images and Docker Compose files, as well as our own YAML-based formats for defining Workcells, the group of instruments designed to work in tandem as part of an automated lab, and Workflows, a linear collection of steps to be run on a given workcell to accomplish a scientific task. We also define a set of common REST-based interfaces that allows WEI to control each instrument in the workcell, and for the user to interface with WEI from python code they write implementing their experiment logic. Finally, WEI implements a core scheduler to queue multiple workflows on a single workcell, and opportunistically schedule each action in the workflow sequence to run on the various available modules. This design allows for multi-tenancy and enables better utilization of a workcell’s resources than single-tenant implementations.

Our system design and implementation has been directed and informed by the needs of our collaborations with domain scientists in the fields of biology, material science, and chemistry. We have successfully demonstrated autonomous experimentation for a single-workcell benchmark, and will focus future work on furthering domain science applications and facilitating autonomous experimentation across multiple workcells.