AUTONOMOUS ROBOT NAVIGATION USING DEEP Q LEARNING

Abstract

Artificial Intelligence has been a topic of great interest in a wide range of fields, such as gaming, computer vision, and robotics. One of the impressive applications is the AI-based Alpha Go that has superhuman performance in the games of Go, with the help of Deep Reinforcement Learning based techniques. Majority of the research in robotics is concerned with mobile robots, especially mobile robot navigation using Deep Reinforcement Learning. Our interests focus on investigating whether a Deep Q Learning-based solution is beneficial to improve the robot behaviors when solving constrained autonomous navigation problems. This work, focus on implementing a robust, point to point navigation module (local path planner, decision making and control modules) using Deep Q Learning for autonomous navigation of indoor robot. And this path planner is able to navigate robot autonomously through cluttered, unstructured environments satisfying constrains like the robot should not collide with any static or dynamic obstacles, it should keep a safe distance from all objects while moving, the robot should traverse on the left-hand side of the corridor and should navigate through shortest path. The frameworks used in this work are ROS, OpenAI Gym and Gazebo Simulator. Train ing neural network architecture in Deep Q Learning requires a large data set which is obtained by stimulating robot on Gazebo environment. And this data set is used by Deep Q Learning algorithm (which is written as a node in ROS) for training