Abstract:
Gathering information from the physical world is critical for applications such as scientific research, environmental monitoring, search and rescue, defense, and disaster response. Autonomous robots provide significant advantages for information gathering, particularly in situations where human access is constrained, hazardous, or impractical. By leveraging intelligent algorithms, these robots can efficiently collect data, enhancing decision-making and accelerating insights. Informative Path Planning (IPP) plays a key role in maximizing the effectiveness of robotic information gathering by generating paths that optimize data collection while respecting operational constraints.

First, this thesis addresses the challenge of solving IPP problems in high-dimensional spaces with non-trivial sensor constraints. Many existing methods rely on simplifying assumptions, such as modeling sensors as isotropic and using low-dimensional motion models. These simplifications can limit real-world applicability and degrade performance. This work introduces IA-TIGRIS, an adaptive and incremental sampling-based planner that efficiently computes long-horizon information-gathering paths while accounting for vehicle motion constraints and complex sensor footprints. The planner is validated in extensive simulations and real-world field tests on multiple UAV platforms.

Second, the impact of environmental disturbances on planning and execution is examined, with a focus on wind-aware UAV path planning. This thesis presents a methodology for estimating wind fields in real-time using onboard UAV measurements, which enables more informed decision-making. A time-optimal path planning approach for UAVs in wind is also developed, leveraging geometric reasoning to reduce computational costs. Additionally, a deep learning-based energy risk assessment framework is introduced to quantify UAV flight risks under uncertain environmental conditions.

Third, this work explores the role of world belief models in improving IPP performance. Effective data gathering relies on belief representations that guide decision-making. This thesis investigates new belief representations and objective functions tailored for informative planning. Additionally, prior information is leveraged to enhance exploration for more accurate information gain prediction. A novel reinforcement learning-based approach is developed to incorporate human-inspired exploration behaviors, demonstrating improved performance over conventional methods.

The methods presented are validated across simulations and extensive field deployments, contributing to the development of robust, real-world-ready autonomous systems. We demonstrate our approach across a range of planning scenarios and validate it on multiple UAV platforms, including fixed-wing and multirotor systems. While this work focuses on robotic information gathering, the underlying algorithms extend to broader domains such as robotic exploration, active perception, target tracking, and multi-agent coordination.

Thesis Committee Members:
Sebastian Scherer, Chair
David Wettergreen
Maxim Likhachev
Geoff Hollinger, Oregon State University

More Information