Like proteins, chains of micrometer-sized droplets with programmable DNA interactions fold into well-defined structures [McMullen et al., Nature 2022]. During folding, the chain passes through a combinatorially large set of intermediate structures, and the pathway it takes decides whether it reaches its target or gets trapped. In recent work, we showed that a reinforcement learning agent can control folding beyond what temperature protocols achieve, much as chaperones assist protein folding. In this project you will use simulations and reinforcement learning to identify which intermediates decide the outcome of folding. You will then extend control to refolding, where the chain switches between structures on demand. This is a step toward reconfigurable micromachines that move, grasp or switch in response to signals.