Reinforcement Learning Based Precise Positioning Method for a Millimeters-Sized Omnidirectional Mobile Microrobot

Abstract

A tri-wheeled omnidirectional microrobot actuated by 3mm electromagnetic micromotors is described. Since it is designed for microassembly in a microfactory, the positioning precision becomes a key parameter. However, with traditional PID control, the microrobot positioning precision is not high during a bearing/axis assembly task. This paper presents a reinforcement learning (RL) algorithm based on the on-policy (Sarsa(λ)) method using linear function approximation. The algorithm is used to generate an optimal path by controlling the choice of four moving actions of the microrobot. The aim is to reach the target position with high positioning precision. Simulations show that this RL algorithm is able to greatly improve the positioning precision with regard to the global path optimization.