Dresden 2017 – wissenschaftliches Programm
CPP 44.9: Vortrag
Mittwoch, 22. März 2017, 17:15–17:30, HÜL 186
Reinforcement learning of artificial microswimmers — •Santiago Muiños-Landin1, Keyan Ghazi-Zahedi2, and Frank Cichos1 — 1Molecular Nanophotonics. University of Leipzig. Institut for Experimental Physics I — 2Information Theory of Cognitive Systems, Max Planck Institute for Mathematics in the Sciences
Reinforcement Learning (RL) is a special area of the Machine Learning discipline which consist in the search of an optimal policy in the context of Markovian Decision Processes (MDP). Learning is based on the interaction of the system with its environment and is guided by sparse rewards. In RL a policy is a function that connects the available actions that an agent can execute with the states where this agent can be located at. MDPs were already proposed as a model for the navigation of natural microswimmers. Here we present now a method that uses this RL in order to achieve an autonomous explorative behavior from a self-thermophoretic microswimmer. We implement it experimentally by photon nudging to reach reinforcement learning of a symmetric microswimmer.