A dynamic knowledge graph approach to distributed self-driving laboratories

Area of Science:

• Artificial Intelligence
• Robotics
• Cheminformatics
• Digital Twins

Background:

• Scientific discovery is accelerated by resource integration and knowledge sharing across organizations.
• Global challenges necessitate collaborative, globally-distributed scientific solutions.
• The World Avatar project aims to create a comprehensive digital twin using a dynamic knowledge graph.

Purpose of the Study:

• To develop an architecture for distributed self-driving laboratories.
• To enable autonomous agents to manage the design-make-test-analyze experimental workflow.
• To ensure data provenance for findability, accessibility, interoperability, and reusability (FAIR data principles).

Main Methods:

• Utilized ontologies to capture data and material flows within experimental cycles.
• Employed autonomous agents as executable knowledge components for workflow execution.
• Implemented a dynamic knowledge graph to represent and evolve research goals.

Main Results:

• Successfully linked two robots in Cambridge and Singapore for real-time, closed-loop optimization.
• Achieved Pareto front generation for cost-yield optimization of an aldol condensation reaction within three days.
• Demonstrated autonomous evolution of the knowledge graph towards scientific objectives.

Conclusions:

• The developed architecture facilitates distributed, collaborative scientific discovery.
• Self-driving laboratories integrated within a knowledge graph can significantly expedite research.
• This approach offers a scalable solution for addressing complex global scientific challenges.