From All To A Subset Harmonic-Based Optimal Motion Planning in Constrained Workspaces using Reinforcement Learning

Abstract

In this thesis, a novel solution is presented for optimal motion planning in static, fully-known environments. Our approach allows a robot to safely navigate towards any destination within a subset of its workspace by using a parametric controller based on Artificial Potential Field (APF) theory. Optimization is achieved through the application of Reinforcement Learning (RL) techniques. More specifically, the parameters of the underlying potential field are adjusted appropriately with a policy gradient algorithm in order to minimize a cost function. The proposed method is not without its limitations, but can still be a valuable addition to the arsenal of established motion planning approaches.