Game-based Decision Algorithm for Socially Compatible Automated Driving: a Case Study of Unsignalized Intersection Driving

Daofei Li,Ao Liu,Hao Pan,Wentao Chen
2021-01-01
Abstract:Smooth and harmonic interacting with other vehicles is one of the ultimate goals of driving automation. However, recent reports of demonstrative deployments of automated vehicles (AVs) indicate that AVs are still difficult to meet the expectation of other interacting drivers, which leads to several AV accidents involving human-driven vehicles (HVs). This is most likely due to the lack of understanding about the dynamic interaction process, especially about the human drivers. By investigating the causes of 4,300 video clips of traffic accidents, it is found that the limited dynamic visual field of drivers is one leading factor in inter-vehicle interaction accidents, especially for those involving trucks. Taking the interaction with a humandriven truck at an unsignalized intersection as an example scenario, a game-theoretic decision algorithm considering social compatibility is proposed. Starting from a probabilistic model for the visual field characteristics of truck drivers, social fitness and reciprocal altruism in decision are incorporated in the game payoff design. Results of human-in-the-loop experiments show that the proposed socially compatible algorithm can effectively improve both safety and time efficiency, and make AV decision more in line with the expectation of interacting human drivers. It can be viewed as a promising solution to handling the interactive issues between automated and human-driven vehicles.
What problem does this paper attempt to address?