A Novel Integrated Decision-Making Evaluation Method Towards ICV Testing Involving Multiple Driving Experience Factors

Intelligent Connected Vehicle (ICV) decision-making is one of the critical techniques towards a more intelligent transportation system. How to assess the actual performance decision algorithms is an important problem for development and further application. Various evaluation metrics have been established to achieve this, such as the distance to real human trajectories, collision rates, travelling time, etc. However, since human also make mistakes, fitting human trajectories is not accurate, while single-dimensional evaluation indicators also overlook other aspects of driving performance. Therefore, an integrated and quantitative assessment method is needed for more comprehensive evaluation towards driving performance. This paper proposes a novel integrated driving decision-making evaluation model towards ICV testing considering multiple factors including safety, time efficiency, comfort, and energy consumption. After establishing the four single item models, we combine them into a linear form and obtain the relative weights by fitting expert ratings on driving events. Drivers with different styles are selected to conduct driving under 12 various designed scenarios on cabin simulators, and volunteers are recruited to give percentage scores to each record for fitting weights. Validation results show that our method achieves far more precise evaluation compared with current single-dimension models, reducing the mean absolute error from 55.82%, 45.74%, 51.52%, and 42.81%.