LiDAR-Based Online Cost Map Construction in Complex Off-Road Environments

Cost map construction is of vital importance for unmanned ground vehicle (UGV) planning in complex off-road environments. However, stability of cost map can be badly affected in two aspects. One is the higher false obstacle detection rate caused by complexity of environments, such as irregularly shaped and distributed obstacles. The other is the drastically changing view field of onboard sensors due to large-angle maneuver of UGV. Besides, to meet requirements of safe and smooth trajectory planning in off-road environments, costs of various terrains should be calculated accurately based on terrain features and ability of UGV. To solve these problems, this paper proposes a LiDAR-based online cost map construction framework. To improve stability of the map, geometry-based obstacle detection algorithm is applied to single frame point cloud and binary Bayesian filter is utilized to filter out detection noises. Blind spot of current laser scan is also eliminated by fusing historical observations. To calculate costs of various terrains accurately, Kalman filter is used to mitigate effect of localization uncertainty and maintain a stable elevation map. Then terrain features in traversable area are extracted by plane fitting and costs are calculated. Real-world experiments illustrate that the proposed framework could provide stabilized local cost map with detailed traversability description in the case of various obstacles and large-angle maneuver of UGV.