Deep Deterministic Policy Gradient for Navigation of Mobile Robots in Simulated Environments

摘要

This paper presents a study of a deep reinforcement learning technique that uses a Deep Deterministic Policy Gradient network for application in navigation of mobile robots. In order for the robot to arrive to a target on a map, the network has 10 laser range findings, the previous linear and angular velocity, and relative position and angle of the mobile robot to the target as inputs. As outputs, the network has the linear and angular velocity. From the results analysis, it is possible to conclude that the deep reinforcement learning's algorithms, with continuous actions, are effective for decision-make of a robotic vehicle. However, it is necessary to create a good reward system for the intelligent agent to accomplish your objectives. This research uses different virtual simulation environments provided by ROBOTIS in the robot simulation software Gazebo in order to test the performance of the algorithm. A supplementary video can be accessed at the following link: https://youtu.be/NhGxEC3g7sU. That shows the performance of the proposed system.