POMDP-Based Behavior Planning for Anticipating and Responding to Lane Excursions in Large Autonomous Vehicles
Large autonomous vehicles are rising in prevalence, and the lane excursion problem is a common occurrence in the navigation of large vehicles through urban roads. To the best of my ability, background research has not revealed any studies on solving this problem under the assumption that the excursion is unavoidable with a large vehicle body. This thesis introduces the lane excursion problem to the research space with a primary focus on dynamic vehicle interaction, and possible solutions are considered. The main solution provided is a POMDP (Partially Observable Markov Decision Process) method utilizing predictive planning given another vehicle with unknown intentions. A simulation environment is created for evaluation, and the POMDP method is evaluated using two measures introduced along with the problem. The POMDP method demonstrates good approximation ability of the hidden system variables and performs better than an uninformed non-predictive approach. With an additional section on the real-life implications of the problem, this study is expected to bring to light the importance of the lane excursion problem and prompt deeper discussion on the domain of large autonomous vehicles as a whole.
Read
- In Collections
-
Electronic Theses & Dissertations
- Copyright Status
- Attribution-NonCommercial 4.0 International
- Material Type
-
Theses
- Authors
-
Yücer, Batuhan
- Thesis Advisors
-
Siegel, Joshua
- Committee Members
-
Morris, Daniel
McKinley, Philip
- Date Published
-
2024
- Subjects
-
Computer science
- Program of Study
-
Computer Science - Master of Science
- Degree Level
-
Masters
- Language
-
English
- Pages
- 52 pages
- Permalink
- https://doi.org/doi:10.25335/eg9v-f774