A New Approach for Path Planning Algorithm Utilizing Modified Deep Q-Network Combined with Artificial Potential Field

Efficient path planning for mobile robots is critical across diverse domains.Therefore, the agents should be provided with a path that is the least expensive, safe, effective, and free of collision.In this research, a hybrid approach is used for path planning, which combines the Artificial Potential Field method (APF) and the Reinforcement Learning (RL) method, especially the Deep Q-Network method (DQN).A modified version of traditional DQN-APF, namely a modified DQN-APF method (MDQN-APF), uses 25 connected neighbours for both APF and DQN values with constructing the dynamic reward model to enhance the agent direction.The evaluation experiments of the proposed approach have empirically proven its effectiveness and can generate safe and efficient paths even in structured, unstructured, and complex environments.The proposed MDQN-APF approach provided improvement in minimizing iterations in the first learning episode to find the target point and the number of episodes to decide the best path by 42.867% and 45.292%, respectively compared to traditional DQN-APF.