Date of Award

2022

Document Type

Campus Access Dissertation

Degree Name

Doctor of Philosophy in Computer Engineering (PhD)

Administrative Home Department

Department of Electrical and Computer Engineering

Advisor 1

Jeremy P. Bos

Committee Member 1

Timothy Havens

Committee Member 2

Darrell L. Robinette

Committee Member 3

Anthony J. Pinar

Abstract

Unstructured environments present several challenges to autonomous agents such as robots and autonomous vehicles. Off-road navigation demands traversal over complex and often changing terrain, understanding which can improve path planning strategies by reducing travel time and energy consumption. A terrain classification and assessment framework has been introduced that relies on both exteroceptive and proprioceptive sensor modalities. Images of the terrain are used to train a support vector machine in an offline training phase and classify the terrain in the operating phase. Acceleration data is used to calculate statistical features that capture the roughness of the terrain and angular velocities are used to calculate roll and pitch angles. These features are used to train a k-means clustering classifier, where k is the number of anticipated terrain types. In the operating phase, cluster centers predict the vibration features associated with the terrain type. Vibration features are measured and the clusters are updated upon traversal, thus adapting to changes in terrain over time. For autonomous vehicles to viably replace human drivers, they must be able to operate in all weather conditions. There is, however, a distinct lack of datasets focused on inclement weather leading to a gap in the development of autonomous systems in such conditions. Falling rain and snow introduce noise in LiDAR returns resulting in both false positive and false negative object detections. We introduce the Winter Adverse Driving dataSet (WADS), a novel dataset collected in the snow belt region of Michigan's Upper Peninsula. WADS is the first multi-modal dataset featuring dense point-wise labeled sequential LiDAR scans collected in severe winter weather; weather that would cause an experienced driver to alter their driving behavior. Our dataset features exclusively events with heavy wet snow and occasional white-out conditions. Over 26 TB of adverse winter data have been collected over three years of which over 7 GB of LiDAR points have been labeled. I also present the Dynamic Statistical Outlier Removal (DSOR) filter, a statistical PCL-based filter capable of removing snow with a higher recall than the state-of-the-art snow de-noising filter while being 28% faster. The DSOR filter is shown to have a lower time complexity, resulting in improved scalability.

Available for download on Sunday, April 16, 2023

Share

COinS