Practical Implementation of Q-Learning and Object Detection for Mobile Robot Path Planning

Expanding on an earlier study that assessed the performance of a Q-learning approach for solving the path planning problem for mobile robots, this research implemented the RL approach in a real-world setting employing the RoboMas-ter EP Core. The robotic system also included object detection and recognition through the robot's sensors and a pre-trained YOLOv9 model. The robot navigated to predefined target points while avoiding stationary obstacles. The Q-learning algorithm was trained using the Google Colaboratory platform. Experi-ments conducted at various speeds identified an optimal speed and high success rates in obstacle avoidance and target region accuracy were achieved. Additionally, the object detection sys-tem demonstrated strong performance in real-time applications. Despite these successes, challenges such as high friction and multitasking inefficiencies were identified. Future research should address these limitations by enhancing control systems and the robot's multitasking capabilities, as well as using a computer with better processing power to improve overall system performance further.