Indoor 2.5D Navigation for Ground Robots

Navigation is a critical challenge for autonomous robots, which is usually restricted to flat surfaces for indoor ground robots. This paper introduces a 2.5D navigation approach for these robots when equipped with 3D LiDAR sensors, combining the advantages of a 3D environment representation with the efficiency of 2D path planning. Inspired by techniques used in aerial and outdoor robotics, the proposed approach dynamically projects 3D data onto a 2D plane while adapting to environment and changes to the robot body real-time. The approach ensures safer navigation by accounting for all detected obstacles regardless of their height, while avoiding excessive conservatism (it allows the robot to navigate through spaces that might seem unreachable with more cautious approaches). Contributions include a novel dynamic footprint updater, enhanced 3D-to-2D point cloud projections, and integration with the ROS navigation pipeline. Experimental validation in three different environments demonstrates superior reliability and adaptability compared to conventional 2D and fixed-height 3D projection approaches.