Applying Deep Q-Networks to Local Route Optimization

The article presents an approach that utilizes deep reinforcement learning, specifically the Deep Q-Network (DQN) algorithm, to enhance local route planning capabilities for autonomous mobile robots. The focus is on leveraging the Robot Operating System (ROS) framework. The research tackles the challenge of enabling robots to navigate dynamic environments while efficiently avoiding obstacles, thereby improving their autonomy and operational efficiency in various industries such as manufacturing, logistics, and service sectors. The DQN algorithm is first trained in a simulated environment, and then implemented on a KUKA youBot robot, both in simulation and in real-world scenarios. The robot’s learning process is guided by a reward system that encourages positive actions like approaching the goal and avoiding obstacles, while discouraging negative outcomes such as collisions. The algorithm optimizes local path planning by computing rewards based on the robot’s distance and orientation alignment with the goal. The results demonstrated the effectiveness of the DQN algorithm in training the KUKA youBot robot to navigate efficiently, adjusting its orientation and avoiding collisions. This research contributes to advancing autonomous mobile robot navigation capabilities, paving the way for more sophisticated and operationally efficient robotic systems across various industries.