Degree

Bachelor of Science (B.S.)

Major(s)

Aerospace Engineering

Document Type

Temporary Embargo for Patent/Proprietary Reasons then Open Access

Abstract

Autonomous off-road navigation presents challenges due to the unpredictable and unstructured nature of real-world terrains, where traditional and learning-based navigation systems often fail to adequately address uncertainty and complex robot-terrain interactions. This thesis surveys recent advances in off-road navigation frameworks, focusing on self-supervised terrain traversability estimation, uncertainty quantification in deep learning, and the integration of physics-based robot-terrain interaction models. It compares geometric, semantic, and hybrid methods for traversability estimation, highlighting the limitations of each in handling deformable and novel terrains. This work emphasizes the necessity of explicit end-to-end uncertainty quantification to distinguish between aleatoric and epistemic uncertainty-to improve risk-aware planning and decision-making in off-road environments. Building on recent developments, the thesis proposes future work on a unified, self-supervised navigation framework that incorporates evidential deep learning into a differentiable, physics-informed architecture. This approach enables the estimation and propagation of uncertainty in terrain parameters, enhancing the reliability of downstream planning and control. By integrating geometric and semantic cues, self-supervision from proprioceptive and physics-based feedback, and advanced uncertainty modeling, the proposed framework lays a foundation for robust, adaptable off-road robotic navigation in unstructured environments.

DOI

https://doi.org/10.54718/WJCJ8348

Date Defended

4-29-2025

Funding Source

N/A

Thesis Director

Jingdao Chen

Second Committee Member

Jessie Cossitt

Third Committee Member

George Dunn

Rights Statement

"Uncertainty-Aware Navigation for Offroad Robotics", Copyright 2025 by Neil Sanipara. My thesis may be used for non-profit educational and research purposes. Note that in addition to my own works of authorship, this thesis may contain and provide citations to third party content. If your use goes beyond fair use, you would need to contact those rights holders for additional licensing/permissions.

Available for download on Sunday, May 14, 2028

Share

COinS