Passability-Based Local Planner Using Growing Neural Gas for an Autonomous Mobile Robot

Abstract

3D spatial perception is one of the most important abilities for autonomous mobile robots. In environments with unknown objects, the ability to perform a local planner, which modifies the global path based on the perception results, is also required as an indispensable capability. In this paper, we propose a method based on Growing Neural Gas with Different Topologies (GNG-DT), which can be applied to unknown data, as a method for 3D spatial perception and local planner in unknown environments. First, we propose a method for extracting travelability perceptions from the features estimated by the topological structure of the GNG-DT. Next, we learn the topological structure of passability information based on the size of the robot from the extracted traversability percepts. Furthermore, we propose a local planner that uses the topological structure of traversability and passability learned from the point cloud currently perceived by the robot. In the experiments, we compared the cases where only traversability was used and where passability information was used in actual environments, and showed that the proposed method can plan a route that determines the area that the robot can actually pass through.