Path Optimization of Autonomous Mobile Robot using Deep Reinforcement Learning

Abstract

Automation systems for mobile robots have advanced significantly in artificial intelligence, especially in autonomous learning. Nevertheless, prior research has mostly concentrated on predetermined routes, ignoring obstacle avoidance and trajectory reconfiguration in real time. This article introduces a novel method that uses Deep Q-Network (DQN) which is based on reinforcement learning to enable an agent to perform activities, get information from a simulated world in the gazebo simulator, and optimize rewards. The algorithm's parameters were precisely set through carefully thought-out trials, and its functionality was thoroughly verified. Unlike traditional navigation systems, this method encourages environmental exploration and allows efficient navigational planning based on learned information. In comparison to conventional techniques, the DQN network has improved skills in computing complex functions utilizing randomized training circumstances within a simulated environment. This huge advancement highlights the potential of our approach to improve mobile robots' ability to learn on their own.