Using Machine Learning to Train Smart Gravitactic Particles
Trajectories of a naive (orange) and smart particle (colored) that are trying to move upward in a complex flow field. While the naive particle simply always seeks to orient itself upward to move up against gravity, the smart particle learns a complex series of orientations (shown in the legend) that it should take in order to ascend much further.Recently there has been interest in understanding if artificial microswimmers can be designed that respond intelligentally to their environment for navigation purposes such as targeted drug delivery. Here, I used reinforcement learning to teach a set of active particles how to maximize their vertical ascent in an otherwise complex flow (i.e. gravitaxis).
