LiDAR-Only Ground Vehicle Navigation System in Park Environment

Abstract

In this paper, a novel and complete navigation system is proposed for mobile ground vehicles in a park environment. LiDAR map representation and maintenance, dynamic objects detection and removal, hierarchal path planning and model-free local planning are developed in the system. The system is formulated in three layers. In the global layer, given the global point cloud map of the environment, the traverse area is detected and its skeleton graph is extracted to represent the global topology of the environment. Then, in the middle layer, the global map is divided into several submaps and each submap is represented by a modified multi-layer grid map. In the local layer, considering the dynamics of the environment, according to the real-time LiDAR observation, a probabilistic distribution-based representation and its updating mechanism are proposed. Based on the hierarchal environment map representation, the path planning and local planning are performed in a hierarchal way. Considering the complexity of the motion model estimation, a model free local planner is used. Extensive experiments are conducted in the real environment and the source code will be made open for the robotics community.