A Review of Path Planning Based on IQL and DQN

Abstract

In the field of robot navigation, Path Planning is a very important problem. Reasonable Path Planning can greatly improve the efficiency of transportation and ensure the safety of robots. The traditional Path Planning method solves the problem of an optimal path to some extent, but it is far from enough. With Machine Learning becoming a hot topic, Path Planning using Reinforcement Learning and Deep Reinforcement Learning has been studied. Q-learning, as a basic algorithm of Reinforcement Learning, has been applied for a long time and has been improved by combining multiple algorithms. And Deep Q-network, a classical algorithm of Deep Reinforcement Learning, has been used to solve complex problems which traditional Reinforcement Learning cannot solve, particularly in Path Planning. This article will present the current achievements in the Improvement of Q-learning (IQL) and Deep Q-network (DQN). In the future, Reinforcement Learning and Deep Reinforcement Learning will generate more and better algorithms to solve problems with higher complexity and need shorter response times.