Developing a Machine-Learning-based Robotic System for Mixing Solvents

Abstract

Mixing solutions and evaluating their concentrations are common and often critical tasks in laboratory experiments. Traditionally, these tasks have been conducted though manual optimization and by sampling across all conditions. Here, we present an alternative approach adopting a robotic system integrated with Bayesian optimization whereby solvents are automatically and autonomously mixed to the desired concentration. The solution control system has a robot-controlled pipette and an optical transmission setup that makes both the liquid injections and measurement evaluations. A demonstration achieved the target solvent concentration after several iterations. The methodology can be applied to various research scenarios, and the presented system can be extended to sophisticated applications. In this technical note, we provide details of the system for installation in other laboratories.