(88g) Autonomous Optimization of Robotic Process Parameters for Accurate Experiments

In the form of self-driving labs, automation and machine learning provide powerful tools to accelerate the pace at which we can make discoveries, and optimize processes and materials. By minimizing human involvement in performing routine tasks, self-driving labs can liberate researchers to generate hypotheses and answer important scientific questions.

Designing a self-driving lab requires balancing generality and accuracy. For example, self-driving labs designed to dose a wide range of sample quantities struggle to do so accurately for very small amounts. As a result, this limits the lab’s ability to carry out experiments that require accurate dosing of small quantities of , for example, catalysts. It would therefore be highly desirable to be able to extend the operating range of robotic systems by finding optimal process parameters, potentially even exceeding manufacturer specifications. However, tuning the process parameters to find the optimal conditions is quite tedious, and significantly encroaches on valuable researcher time. By establishing fully autonomous experiments for process parameter optimization, we can remove the necessity for human involvement and maximize research efficiency.

I will discuss how a closed-loop system integrating a multi-objective Bayesian optimization algorithm and a robotic synthesis platform can be trained to dose small quantities with good accuracy and precision. Building a self-driving lab requires delicately balancing generality, specificity and equipment cost. For instance, the manufacturer specified accuracy of an instrument may limit exploration of future process of experiment parameters, while a robot capable of accuracy in a wide operating range is typically very expensive. In conjunction with a machine learning algorithm, the robot can learn to optimize its own process parameters to achieve accurate results.