Simulation Testing Scenario Generation for Comfort Evaluation of Automated Vehicles

Simulation test is an essential step for the development in control strategy of automated vehicles. On the other hand, driving comfort of automated vehicles are getting more and more concerns. Therefore, generating simulation test scenarios can accelerate the development and iteration of control strategies on both safety and comfort. However, a lot of current studies about test scenario generation focus on the safety of automated vehicles and few studies concern on the driving comfort evaluation. In this paper, a testing scenario generation framework for comfort evaluation of automated vehicles is proposed. A raw scenario library is generated by applying the multi-parameter gibbs sampling method according to the fitting distributions of the influence values in the specific scenario. To identify the comfort level of the generated raw scenario library, simulation test and comfort evaluation is taken. A vehicle dynamic model and a basic automated driving controller with AEB and ACC functions are adopted in the simulation test, and a random-forest model is established to give comfort scores based on the dynamic parameters during simulation test. To extracted critical challenging scenarios from the raw scenario library, an objective searching function is designed with a comfort level factor and a occurring frequency factor. A cut-in case is analyzed to demonstrate the proposed framework and a naturalistic dataset with 30,875 data fragments is adopted as reference. The results show that critical scenarios generated by the proposed framework are more challenging on driving comfort than the scenarios in naturalistic dataset.