Abstract
dc:description.abstractDefensive driving aims to increase the safety of drivers on the road and decrease their likelihood of being involved in an accident. However, autonomous defensive driving remains an under-researched area today, especially in vision-based E2E driving models. This thesis aims to determine if making these models follow defensive driving practices will increase their safety and lower the chances of getting into an accident. This work aims to fill this gap by firstly establishing metrics and thresholds to define a model as a defensive driver. Secondly, I will collect a defensive driving dataset based on the thresholds defined previously. Lastly, I will train and test these models for increased safety and decreased likelihood of getting into an accident based on these metrics. In addition, I will test them on the Bench2drive benchmark to ensure a comprehensive evaluation of the experiments. This work makes extensive use of the CARLA simulator for all data, experiments, benchmarking and tests, making it a closed-loop scenario. Results from the experiments show that there is a considerable increase in the safety of the model “DriveTransformer”. This shows that defensive driving can contribute to the increased safety of autonomous drivers and further highlights the need for these models to prioritize safety.
Degree
thesis:*- Name thesis:degree_name
- Master of Science
- Level thesis:degree_level
- masters
- Discipline thesis:degree_discipline
- Computer Science
- Grantor
- Boston University
- Year dc:date.issued
- 2026
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Kingsly, Sujayanand
- Advisor dc:contributor.advisor
-
- Xiao, Chang
- ORCID dc:identifier.orcid
Subjects
dc:subject × 6Rights
dc:rights- Statement dc:rights
-
- Attribution-NonCommercial-ShareAlike 4.0 International
- Licence dc:rights.uri
- Language dc:language.iso
- en_US
Identifiers
dc:identifier.*- Handle dc:identifier.uri
- https://hdl.handle.net/2144/53015