To Develop Human-like Automated Driving Strategy Based on Cognitive Construction: Appraisal and Perspective

Automated driving (AD) aims at human-level intelligence and good cooperation with humans. Therefore, it is necessary to consider the characteristics of human behavior and intelligence. Indeed, researchers have attempted to develop human-like AD strategies. However, there is no quantitative definition of human-like strategies. Meanwhile, the existing strategies cannot yet be applied in extensive scenarios well. In this paper, we look for inspiration from human intelligence and propose a general human-like AD solution. Specifically, we first summarize and classify driver behavior models, which are quantitative descriptions of the driver's natural characteristics. Among the existing methods, models based on cognitive process are considered to be advantageous in inspiring us to develop human-like AD strategies. Secondly, we summarize specific cognitive characteristics during driving and come up with three critical rules. The extensive evidence of these rules is collected from various disciplines, covering behavioral, cognitive, and neural layers. Finally, after a brief review of cognitive architecture, we propose a cognitive architecture dedicated to driving. We illustrate the functions of modules and the interaction among modules. This architecture can already provide insightful ideas for end-to-end AD designs. After further realization, it can help define the human-like characteristics and even work as human-like AD strategies.