Mobile Robot Path Planning Based On Reinforcement Learning

With the rapid development of mobile robot technology, the technology gets more and more attention. Path planning in unknown environments is a key technology for mobile robots. Reinforcement learning is an important machine learning methods, which attempts to find an optimal strategy by trial-and-error interaction with the environment. According to the problem of mobile robot path planning in unknown environments, this paper addresses mobile robot path planning based on reinforcement learning.Firstly, in order to find a trade off efficiently between exploration and exploitation, an improved Q-learning algorithm based on action model approximation is designed. To reduce the computation, the true action model is simplified by observing the environment around the robot. Compared with the traditional action-selection policy, the simulated annealing based on action model approximation can be used to produce an optimal policy and improve the success rate.Secondly, a hierarchical reinforcement learning approach is applied for robot path planning in complex dynamic environments. The system of path planning is divided into three layers from top to bottom, comprising the root task collaboration layer, the subtasks select layer and the environment interaction layer. Decomposing the path planning task into avoiding static and dynamic obstacle and closing to target point can decrease the state space and learning difficulty.Lastly, the mobile robot platform is set up in the laboratory and the communication is handled by Ubuntu and ROS. The three-dimensional model of Pioneer3-AT mobile robot and simulation environments are designed and used to simulation experiment. A path planning based on reinforcement Learning is implemented with a P3-AT robot in a real world. The experimental results verify that the robot in an unknown environment can reach the target point successfully by independent study.