Potential Game-Based Decision Making in Autonomous Driving (Abstract)

Abstract

Game-theoretic approaches characterize agents’ interactions from a self-interest optimization perspective, consistent with humans’ reasoning, and therefore, are believed to have the potential to solve the decision making for autonomous vehicles (AVs) when they interact with human-driven vehicles and/or pedestrians. However, despite high hopes, conventional game-theoretic approaches often suffer from scalability issues due to the complexity of multi-player games and from incomplete information challenges such as the lack of knowledge of other traffic agents’ cost functions that reflect the variability in human driving behaviors. In this talk, we will show how to address these challenges by developing a novel potential game (PG) based framework. Specifically, we will introduce a new PG framework that not only solves the multi-player game in real time but also guarantees the ego vehicle safety under appropriate conditions despite unexpected behaviors from the surrounding agents.