Robust Negative Obstacle Detection in Off-Road Environments Using Multiple LiDARs

Autonomous vehicles must analyze and understand the surrounding terrain in real time when driving in off-road environments. Yet few UGVs can effectively detect pits and ditches in an off-road environment at high speeds autonomously. This paper presents an adaptive negative obstacle detection method for autonomous vehicles using multiple side-mounted LiDARs. The method begins by extracting range jump in radial direction from the raw sensor data to get potential negative obstacle feature point pairs, and then estimates the vector of the scanned ground surface around the potential negative obstacles each moment. Subsequently, the feature point pairs are filtered based on several geometrical features related to the range and the deviation between the feature point pairs and the ground vector. Then we fuse the feature point pairs from multiple LiDARs and multiple frames to improve robustness and the longest detection distance. A series of experiments have been conducted in several typical off-road scenarios. The experimental results show that the proposed method can accurately detect negative obstacles and could deal with complex terrain in off-road environments.