In autonomous driving systems, advanced sensing technologies (such as Light Detection and Ranging (LIDAR) devices and cameras) can capture high volume of data for real-time traversability analysis. Off-road autonomy is more challenging than other autonomous applications due to the highly unstructured environment with various types of vegetation. The understory with unknown density can create extremely challenging scenarios (such as negative obstacles masked by dense vegetation) by concealing potential obstacles in the terrain, leading to severe vehicle damage, significant financial loss, and even operator injury or death. This paper investigates the impact of understory vegetation density on obstacle detection in off-road traversability analysis. By leveraging a physics-based autonomous driving simulator, a machine learning–based framework is proposed for obstacle detection based on point cloud data captured by LIDAR. It is observed that the increase in the density of understory vegetation adversely affects the classification performance in correctly detecting solid obstacles. With the cumulative approach used in this paper, however, sensitivity results for different density levels converge as the vehicles incorporates more time frame data into the classification algorithm.

In this article, a compressive sensing-based reconstruction algorithm is applied to data acquired from a nodding multibeam Lidar system following a Lissajous-like trajectory. Multibeam Lidar systems provide 3D depth information of the environment, but the vertical resolution of these devices may be insufficient in many applications. To mitigate this issue, the Lidar can be nodded to obtain higher vertical resolution at the cost of increased scan time. Using Lissajous-like nodding trajectories allows for the trade-off between scan time and horizontal and vertical resolutions through the choice of scan parameters. These patterns also naturally subsample the imaged area. In this article, a compressive sensing-based reconstruction algorithm is applied to the data collected during a relatively fast and therefore low-resolution Lissajous-like scan. Experiments and simulations show the feasibility of this method and compare the reconstructions to those made using simple nearest-neighbor interpolation.