Robotic path planning with sparse environment representations
Abstract/Contents
- Abstract
- This thesis explores different aspects of robot resiliency in path planning problems. Our goal is to endow autonomous robots with the ability to be robust to uncertainty and adaptable so that they can be entrusted in challenging real-world applications. To do so, we focus on four application areas, and in each we carefully co-design the robot's environment representation and path planning algorithms so that they jointly capture the key properties of each problem. We first focus on unknown environment exploration, which is a problem where model uncertainty is so high that the planners proposed in the literature are myopic and greedy. We propose an environment representation and coverage planning algorithms that strategically balance greedy information gain and long-horizon planning so that the robot is able make strategic planning decisions over its full time budget. By doing so, we outperform methods that either ignore model uncertainty, or compensate with a myopic planning strategy, while remaining robust to perceptual challenges. We used the method in practice to explore a variety of urban, tunnel, and mine-like environments with autonomous Spot and Husky robotic platforms, and we integrated the solution in Team COSTAR's global planning framework for participating in the DARPA SubTerranean Challenge in 2021. The second application area in this thesis is the problem of searching for radio signal sources in an unknown environment. We find a similar gap in the literature where to overcome the high model uncertainty, solutions proposed unfortunately limit the robot's action space to the extent of its explored regions. We propose a formulation where the robot looks ahead into the unknown space to search for radio emitters that may be located far away from its visited locations. In the third application area, we focus on the robot's ability to adapt to human indications, which is a key aspect of making a robot resilient and trustworthy in the real world. We propose an encoding of a human's intention that enables the robot to infer a human's path preference from indications and adapt to these preferences online. Finally, we explore the Multi-Agent Path Finding problem, and propose a stochastic travel time model for robots to compute solutions optimizing their expected performance, making them robust to unexpected delays in travel time.
Description
| Type of resource | text |
|---|---|
| Form | electronic resource; remote; computer; online resource |
| Extent | 1 online resource. |
| Place | California |
| Place | [Stanford, California] |
| Publisher | [Stanford University] |
| Copyright date | 2023; ©2023 |
| Publication date | 2023; 2023 |
| Issuance | monographic |
| Language | English |
Creators/Contributors
| Author | Peltzer, Oriana Claudia |
|---|---|
| Degree supervisor | Cutkosky, Mark |
| Degree supervisor | Kochenderfer, Mykel |
| Thesis advisor | Cutkosky, Mark |
| Thesis advisor | Kochenderfer, Mykel |
| Thesis advisor | Schwager, Mac |
| Degree committee member | Schwager, Mac |
| Associated with | Stanford University, School of Engineering |
| Associated with | Stanford University, Department of Mechanical Engineering |
Subjects
| Genre | Theses |
|---|---|
| Genre | Text |
Bibliographic information
| Statement of responsibility | Oriana Peltzer. |
|---|---|
| Note | Submitted to the Department of Mechanical Engineering. |
| Thesis | Thesis Ph.D. Stanford University 2023. |
| Location | https://purl.stanford.edu/hk664kf9285 |
Access conditions
- Copyright
- © 2023 by Oriana Claudia Peltzer
- License
- This work is licensed under a Creative Commons Attribution Non Commercial 3.0 Unported license (CC BY-NC).
Also listed in
Loading usage metrics...