Robotic Perception and Planning in Challenging Environments
Name
jia-yixuany-sm-aeroastro-2026-thesis.pdf
Size
20.81 MB
Format
Adobe PDF
Checksum (MD5)
cfc96e32f2b7f6f328a97c00b6f2b60f
Author(s)
Jia, Yixuan
Advisor(s)
How, Jonathan P.
Date Issued
February 2026
Publisher
Massachusetts Institute of Technology
Abstract
Autonomy in robotics has made significant progress in structured environments such as warehouses and road networks, yet extending these capabilities to unstructured, perceptually degraded, or adversarial settings remains a fundamental challenge. This thesis investigates several robotic tasks situated in these demanding environments, developing perception and planning algorithms that enable autonomous ground and aerial robots to operate under such conditions.
First, a real-time risk-aware planning framework is developed for UAVs operating near threat-emitting radar sites, combining hierarchical planning with imitation learning to generate orientation-aware, low-risk maneuvers efficiently.
Second, a self-supervision-enhanced imitation learning approach is introduced for nap-of-the-earth flight, improving low-altitude navigation using only onboard monocular imagery.
Third, a probabilistic data association framework is proposed to capture multiple solution modes in highly ambiguous global localization scenarios, enabling reliable point cloud and object map registration. Finally, an off-road navigation framework that leverages implicit neural representation to enable gradient-based trajectory optimization is proposed to jointly adapt path geometry and speed profile for off-road navigation. Together, these contributions advance the capabilities of autonomous robots in environments where conventional perception and planning methods struggle.
MIT Department
Massachusetts Institute of Technology. Department of Aeronautics and Astronautics
Terms of Use
In Copyright - Educational Use Permitted
Copyright retained by author(s)
Persistent DSpace Link
