Policy graphs and theory of mind for explainable autonomous driving

Autonomous driving has made remarkable strides over the past two decades, propelled by advancements in artificial intelligence (AI). However, the opacity of decision-making processes in autonomous vehicles (AVs) has created significant barriers to societal trust and regulatory acceptance, primarily...

Descripción completa

Detalles Bibliográficos
Autor: Montese, Sara
Tipo de recurso: tesis de maestría
Fecha de publicación:2024
País:España
Institución:Universitat Politècnica de Catalunya (UPC)
Repositorio:UPCommons. Portal del coneixement obert de la UPC
Idioma:inglés
OAI Identifier:oai:upcommons.upc.edu:2117/419094
Acceso en línea:https://hdl.handle.net/2117/419094
Access Level:acceso abierto
Palabra clave:Automated vehicles
Explainable AI
Policy Graphs
Autonomous Driving
Vehicles autònoms
Àrees temàtiques de la UPC::Informàtica::Intel·ligència artificial::Llenguatge natural
Descripción
Sumario:Autonomous driving has made remarkable strides over the past two decades, propelled by advancements in artificial intelligence (AI). However, the opacity of decision-making processes in autonomous vehicles (AVs) has created significant barriers to societal trust and regulatory acceptance, primarily due to concerns surrounding trustworthiness, safety, and accountability. This thesis explores the application of Policy Graphs (PGs), an innovative explainable AI (XAI) technique, in autonomous driving. PGs represent an agent's policy as a directed graph with natural language descriptors, offering a human-readable explanation of the agent's behaviour. This framework is further enhanced by incorporating Theory of Mind concepts, enabling a deeper understanding of these systems as if they possessed beliefs, desires, and intentions. This approach allows the graph to capture what the agent does and what it desires and intends to do. Our research aims to make three primary contributions: 1. A comprehensive review of current XAI techniques in the context of autonomous driving provides a solid foundation for our work. 2. Development of an advanced framework that integrates the agent's desires and intentions into Policy Graphs, thus facilitating the extraction of motivations behind specific driving decisions and identifying abnormal or undesirable behaviours. 3. An exploration of how external factors, such as weather and lighting conditions, influence AV decision-making, uncovering potential harmful biases and patterns under various driving scenarios. The findings of this study show that combining Policy Graphs and Theory of Mind concepts offers an effective approach to explaining and interpreting vehicle behaviour. This innovative methodology significantly enhances our understanding of autonomous driving systems. More importantly, it has the potential to increase public trust and contribute to more robust regulatory frameworks in the field of autonomous vehicles, thereby contributing to the safe and widespread adoption of AV technology.