Abstract: There is an opportunity to bring together the power of large-scale robotic laboratories with high-performance computing and artificial intelligence. To lead in this novel area, Argonne is developing an emerging thrust in AI-driven Large-scale Automated Scientific Experimental Laboratories (Self-driving Labs) aimed at accelerating discovery in materials, chemistry, biochemistry, biology and energy. We are moving forward on pilot projects in Self-driving Labs to demonstrate the end-to-end concept with the goal of creating momentum on the concept of AL-based experimental loops that use robotic platforms to carry out high-throughput experiments under the control of a machine learning-based optimization and lead generation system. 

Organizing our efforts at Argonne will rapidly demonstrate capabilities and build crosscutting teams that have experience to drive a broader research agenda in this space. This year, we envision working on a specific set of driver problems, roughly two or three per domain, where a domain is biology, chemistry, materials, polymers, etc. The idea of having two or more per domain is to, from the very beginning, build generalization into the capabilities. The core demonstrations should not be overly optimized for a specific problem, but instead be designed as a general capability that can be easily retargeted to new problems, with addition of new instruments or reagents or with reprogramming, but without the need to build a whole new infrastructure and a whole new software stack for each problem.