An Improved Path Planning Algorithm for Indoor Mobile Robots in Partially-Known Environments

Abstract

In this paper, an improved path planning algorithm with two stages is proposed for indoor mobile robots to help them navigate across dynamic crowded areas. First, an environmental model is built with a restricted tangent graph, and the shortest path is determined before the robot starts to move. Second, the obtained path is employed to create virtual corridor distances. This featured data along with local range-finder readings are used to train a deep Q-learning agent to efficiently plan the robot’s motion. Simulation results show above 70% success rate with significant improvements compared with a counterpart algorithm in terms of path length, learning cost, and generalization capability.