Development of Explainable Artificial Intelligence Approaches for Autonomous Vehicles

• Author / Creator

  Atakishiyev, Shahin

• Autonomous driving, as a rapidly growing field, has received increasing attention from the general society and the automotive industry over the last two decades. However, road accidents involving autonomous vehicles have hindered societal acceptance and deployment of this technology on roads. As self-driving decisions are powered by artificial intelligence approaches, intelligent driving systems must justify their actions,
  particularly in critical traffic scenarios. Consequently, explainability of autonomous driving has emerged as a vital research direction in the field.
  This dissertation aims to develop explainable artificial intelligence techniques for autonomous vehicles by approaching the existing issues from three essential aspects: interactivity,
  robustness analysis, and time granularity of explanations. In this sense, I first present a comprehensive overview of explainable artificial intelligence approaches for autonomous vehicles and describe the research gaps in this direction. Second, I introduce a visual question answering approach to explain autonomous driving actions in an interactive manner. Third, I propose a situation awareness framework for autonomous vehicles backed by explanations and human-machine interfaces. Finally,
  I thoroughly investigate safety implications of explainable artificial intelligence in end-to-end autonomous driving via critical case studies and an empirical analysis.
  Overall, in pursuit of developing explainable artificial intelligence approaches for autonomous vehicles, this dissertation highlights (1) how to build intelligible and interactive explanations, (2) critical challenges in building trustworthy interactive explanations, and (3) how to leverage explanations in enhancing self-driving safety.

• Subjects / Keywords

  • Explainable Artificial Intelligence (XAI)
  • Autonomous Driving
  • Regulatory Compliance
  • Interactive Explanations
  • Deep Learning

• Graduation date

  Fall 2024

• Type of Item

  Thesis

• Degree

  Doctor of Philosophy

• DOI

  https://doi.org/10.7939/r3-h9y1-ww82

• License

  This thesis is made available by the University of Alberta Library with permission of the copyright owner solely for non-commercial purposes. This thesis, or any portion thereof, may not otherwise be copied or reproduced without the written consent of the copyright owner, except to the extent permitted by Canadian copyright law.