How do drivers respond to driving risk during car-following? Risk-response driver model and its application in human-like longitudinal control

The blooming of intelligent connected vehicle (ICV) has been continuously shaping a hybrid traffic environment in which the road is shared among ICVs and vehicles driven by human drivers. However, due to the insufficient understanding of the human driving strategy and style, the conflicts between ICVs and human drivers have arisen public attention, threatening the road safety and bottlenecking the development of ICV. In order to embed the human driving strategy in the intelligent driving system, researchers have been rolling out efforts on driver modeling. Most driver models, however, still suffer from the limited application scope or poor transparency. Within our finite horizons, a unified and readable driver model for various driving scenarios is generally unobtainable. In this work, we tried to model the human driving strategy from an aspect of human nature, that is, the way human drivers respond to the driving risk. We employed the risk field theory (also known as the safety field theory) to model the environmental risk in a comprehensive manner. By studying the risk-response strategy from the driving data of 24 human drivers, we proposed a unified structure, which we call the risk-response driver model (RRDM), to model the human driving strategy. This model provides access to learning not only the average driving strategy of a group of human drivers but also the specific driving style of a single driver. The explicit and readable driving strategy produced by RRDM can be directly employed to reproduce human-like longitudinal driving control. We verified the performance of our model in car-following tasks and found that its human-like driving performance is recoverable among the human drivers who participated in the tests.