Abstract In unstructured environment, there exist negative obstacles, such as pits, ditches et al., which is below ground and affect the safety driving of autonomousland vehicle(ALV). Hence, the accurate and efficient negative obstacle detection is an important part in the area of ALV unstructured environment perception. Aiming at this problem, a negative obstacle detection algorithm based on multiple LiDARs and compositional features for unstructured environment is proposed. Firstly, for sensor installation, a mulit-LiDAR installation manner is proposed within complementary ability. Secondly, for obstacle detection, two methods are proposed: a 64-lines LiDAR negative obstacle feature point pair detection based on local convexity in amplitude direction and local density at up-side of a ditch, and a 32-lines LiDAR negative obstacle feature point pair detection based on range jump in radial direction and local density at up-side of a ditch. In order to fusing both the spatial and temporal view of the obstacle, a Bayes rule is used to fuse the feature point pairs of multi-sensors and multi-frames. Then the DBSCAN (Density-Based Spatial Clustering of Applications with Noise) is introduced to cluster and filter the feature point pairs. And finally, the data are gridded to extract the negative obstacle grid. The experimental results show that the proposed method obtains a good accuracy for detecting negative obstacles in unstructured environment.