A Role for HTN Planning in Increasing Trust in Autonomous Driving

Abstract

The adoption of autonomous vehicles mainly depends on the driver's trust in the vehicle's capabilities. Influencing trust requires giving it a central role when designing the vehicle's functionalities, including the one for driving from one location to another. Addressing this driving task requires not only considering environmental and vehicle's conditions (e.g., road obstacles, fuel level, but also factors that influence trust, such as variability of trust, use of understandable and structured knowledge, and operation transparency. One way to address such a driving task is to solve it as a planning problem. Among AI planning techniques, Hierarchical Task Network (HTN) planning provides a powerful approach to model rich domain knowledge using hierarchical constructs, simulating the way in which one conceptualises knowledge and performs decision making. Here, we analyse the suitability of HTN planning for the trust-based driving task and define the respective planning problem. Based on this, we model an HTN domain for the driving task and use it to solve the driving task in two case studies. The results indicate that trust-based HTN planning provides a feasible approach for efficiently computing plans that maximise trust.