Leveraging multiple sources of information to search continuous spaces
Share
- Publication Date
- Abstract
Path planning algorithms can solve the problem of finding paths through continuous spaces. This problem appears in a wide range of applications, from navigating autonomous robots to automating assessments of surgical tolerances. The performance requirements on these algorithms tend to become more demanding as the problems they are applied to become more sophisticated. This simultaneous increase in performance requirements and application complexity calls for new approaches to the path planning problem and makes it an active area of research in robotics and beyond.
This thesis demonstrates how different types of information can be leveraged to solve the path planning problem more effectively. Optimization-specific information can guide the search towards high-quality solutions, environment-specific information can exploit incremental information about the surroundings, and intent-specific information can directly align the search of a problem with its priorities.
These three types of information are leveraged in this thesis by integrating advanced graph-search techniques in sampling-based path planning algorithms. The resulting planners, Advanced BIT* (ABIT*), Adaptively Informed Trees (AIT*), and Effort Informed Trees (EIT*), are theoretically shown to be almost-surely asymptotically optimal and experimentally demonstrated to outperform existing planners on diverse problems in abstract, robotic, and biomedical domains.
- Publication Details
- Type
- D.Phil. Thesis
- Institution
- University of Oxford
- Manuscript
- Open-Access PDF
- https://robotic-esp.com/papers/strub_dphil21.pdf
- Google Scholar
- Google Scholar
- BibTeX Entry
@phdthesis{strub_dphil21,
author = {Marlin P Strub},
title = {Leveraging multiple sources of information to search continuous spaces},
school = {University of Oxford},
year = {2021},
}