Among the many fascinating and useful aspects of AI, reinforcement learning performs well. Using the principles of action and reward, reinforcement learning makes it easier to learn new tasks. The issue of robot navigation is tackled by motion planning. The ability to automatically react in real-time to changes in the environment is currently missing from motion planning methods. An intricate setting full of impediments exacerbates the situation. As a result of the capabilities of the reward system and feedback to the environment, robotic systems can be enhanced through reinforcement learning. Managing a complicated setting may get easier by using this. Current path planning algorithms converge to a solution late because they are computationally expensive, less responsive to the environment, and slow. Additionally, because of the need for post- processing, they are not as effective for task learning. The problem-solving capabilities of reinforcement learning lie in its action feedback and reward policies. This study introduces a new reinforcement algorithm that combines deep learning with Q-learning. The suggested method is tested in a space with limited space and a lot of obstacles. Additionally, we handle ways to improve the merging of collision avoidance and motion planning based on reinforcement learning. At the 640th and 690th episodes in a crowded and a small route environment, the agent of the suggested method converged. Based on the amount of turns and the planner's ability to converge the path, a state-of-the-art comparison reveals that the suggested strategy beat existing alternatives.
Babu et al. (Tue,) studied this question.