Path planning of mobile robot based on reinforcement learning to reach faster training
Niloofar Takzare, Naeim Yousefi Lademakhi, Moharam Habibnejad Korayem · 2024
Recent advances in reinforcement learning have to its adoption for solving complex and challenging problems in robotics, which has been well received by researcher. One of the advantages of deep reinforcement learning (DRL) is its ability to be applied without prior knowledge of the system. Path planning of mobile robots and avoid obstacle are among the biggest challenges in mobile robotics. In this article, a reinforcement learning algorithm is proposed, which automatically learns an optimal state for controlling the robot using several action parameters and a proposed artificial potential new reward function. Therefore, the robot path planning finds its way to the goal with fast training and less time. The simulation has been performed for two command and proposed algorithms in the environment with and without obstacle. The results of the simulations showed that the robot is able to reach all the set goals faster with the proposed new reward function. Also, the path planning has been validation with the feedback linear control. The reason is that the feedback linearization method takes a smoother path than DQN, but the DQN algorithm with the proposed reward function learns much faster and reaches all the set target points.