A validated failure behavior model for driver behavior models for generating skid-scenarios on motorways

Abstract

The automation of the driving task will gain importance in future mobility solutions for private transport. However, the sufficient validation of automated driving functions poses enormous challenges for academia and industry. This contribution proposes a failure behavior model for driver models for generating skid-scenarios on motorways. The model is based on results of the five-step-method provided by accident researchers. The failure behavior model is implemented using a neural network, which is trained utilizing a reinforcement learning algorithm. Hereby, the aim of the neuronal network is to maximize the vehicle’s side slip angle to initiate skidding of the vehicle. Concluding, the failure behavior model is validated by reconstructing a real accident in a traffic simulation using the failure behavior model.