A Reinforcement Learning-Driven Algorithm for Rapid Path Replanning of Robot Navigation in Indoor Uncertain Discrete Environments

In the context of indoor dense discrete environments, the rapid replanning of paths upon encountering novel obstacles has remained a formidable challenge. Traditional rule-based global path planning algorithms often struggle to achieve a balance between optimal global path determination and swift decision-making. Consequently, this paper introduces a reinforcement learningbased path replanning algorithm designed for indoor discrete environments. The algorithm employs a Q-table to store the robot's movement strategies at various locations and utilizes the persistent interactive capabilities of reinforcement learning with the environment, thereby enabling rapid decision-making while ensuring that the re-planned paths trend towards optimality. We conducted simulation experiments within three distinct sizes of indoor dense discrete environments to validate our approach. The results demonstrated that, under conditions where the re-planned path quality approaches optimality, the decision-making speed of our proposed reinforcement learning-based path replanning algorithm significantly surpassed that of the $\text{A}^{\ast}$ algorithm. Furthermore, we have proposed an auxiliary path replanning algorithm grounded in deep reinforcement learning. This supplementary algorithm is adept at integrating with the main reinforcement learning algorithm, thereby effectively accommodating scenarios where the robot's target destination changes during the navigation process, facilitating robust path replanning in response to such alterations.