A New Deep Reinforcement Learning Based Robot Path Planning Algorithm without Target Network

Intelligent agent navigation has broad application scenarios, one of the hot research directions in this field is agent dynamic path planning. For the target network in deep reinforcement learning to gradually deviate from online reinforcement learning, a new softmax operator is applied to replace the max operator in the original network after deleting the target network. New prioritized experience replay is applied to enhance the experience utilization of the agent and dueling network is employed to improve the perceptions of the environment in the path planning process. A modified dynamic ϵ-greedy algorithm is then developed to select actions. The experimental results in the simulation environment show that even after deleting the target network, the algorithm in this paper can still converge to a higher value within limited episodes, which proves its effectiveness.