Date of Award
2026
Document Type
Open Access Dissertation
Degree Name
Doctor of Philosophy in Electrical Engineering (PhD)
Administrative Home Department
Department of Electrical and Computer Engineering
Advisor 1
Jeremy P. Bos
Advisor 2
Darrell L. Robinette
Committee Member 1
Anthony J. Pinar
Committee Member 2
Tan Chen
Abstract
Autonomous driving in winter weather faces critical challenges across perception, planning, and control. Perception systems are frequently compromised by sensor obscuration from snow, invisible hazards such as black ice, and dynamic noise from turbulent snowfall, all of which reduce data reliability. These limitations propagate through the autonomy stack to planning, where obscured road features and degraded localization create an incomplete environmental understanding. Consequently, control systems must handle unpredictable low-friction conditions and chaotic vehicle dynamics that often defy conventional models.
This research advances the state of winter autonomy through three interconnected contributions. First, an integrated control scheme and a scaled testing platform were developed using sampling-based planning and control strategies to validate autonomous performance on snow and ice. These experiments provide critical insights into how uneven, low-friction terrain influences vehicle stability, establishing the fundamental requirements for real-time traversability estimation on featureless surfaces. Addressing these requirements, the second contribution introduces a robust traversability estimation and wheel-track detection method utilizing RGB-thermal fusion. Specifically architected to provide reliable navigation cues in texture-sparse, snow-covered environments, this approach utilizes an ablation study to demonstrate that thermal imagery significantly improves wheel-track detection accuracy.
Finally, this work presents a systematic quantitative study of LiDAR-based 3D object detection under adverse weather conditions. Contrary to the common assumption that detection performance degrades as snowfall intensity increases, the analysis reveals a counterintuitive trend where higher intensities correlate with improved detection metrics. By investigating this phenomenon through meta-information analysis and ablation studies, the research identifies key confounding variables, such as object point density and distance, that influence perception reliability, thereby uncovering the underlying factors driving this trend.
By addressing the pipeline from perception to integrated control, this work provides crucial qualitative and quantitative evaluations of autonomous system reliability in adverse weather. Furthermore, this dissertation identifies vital future research trajectories, including the adaptation of large-scale Foundation Models for extreme weather and the development of high-fidelity physics simulations for snow-surface interactions. Collectively, these findings help bridge the gap between theoretical robustness and the practical deployment of autonomous vehicles in challenging winter climates.
Creative Commons License
This work is licensed under a Creative Commons Attribution-Noncommercial-No Derivative Works 4.0 License.
Recommended Citation
Yang, Yiming, "Robust Autonomous Driving in Winter Weather: Perception and Integrated Control Under Adverse Conditions", Open Access Dissertation, Michigan Technological University, 2026.