The process of materials development, which includes molecular design, chemical synthesis and analysis, is time-consuming since it generally operates with trial-and-error method. When it comes to the molecular discovery, high-performance computing and artificial intelligence (AI) have been innovating the materials exploration. However, the experiments still serve as a major rate-limiting stage as the level of automation is low. In order to overcome the issue, an autonomous materials development (AMD) platform has been proposed by integrating AIs and robots to run the whole procedure from synthetic planning to experiment autonomously within a closed-loop workflow.<br/>The AMD platform consists of various AI models as well as six hardware modules of transfer robot, pantry, dispensing, reaction of six parallel slots, sample-preparation, and analysis. The complex architecture requires a dedicated on-line coordinator for efficient and harmonious operation. In this respect, we developed a novel scheduling framework based on multi-agent scheduling (MAS) concept. In the MAS approach, all hardware resources are defined as independent agents and each agent executes a given task autonomously according to its own decision. Thus, the proposed framework enables real-time dynamic and flexible job management.<br/>The performance of the scheduler was validated on the AMD platform in terms of repeatability, stability, and efficiency by synthesizing various organic molecules. The platform found synthetic recipes whose reaction yields are better than known references with the help of deep learning and Bayesian optimization. The scheduler communicated with the AI models continuously and the closed-loop feedback was successfully accomplished each other by running multiple recipes simultaneously. Furthermore, it drove the platform stably for several days and alleviated the arduous labor of chemists. We expect that the scheduler can be a crucial element in the self-driving platforms and eventually help scientists gain more time to immerse themselves in more creative research works.

Scheduling framework for AI-driven autonomous materials development platform

Fall 2022 abstracts, posters, and presentations will launch in mid-September.

ACS Fall 2022

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.