Performance Evaluation of Human Behavior Model Based on Recurrent Neural Network for Driving

- Comparison with the case using model predictive control -

Abstract

This article discusses the results of evaluating the performance of a recurrent neural network (RNN)-based human behavior model proposed for estimating human behavior that can be used for driver assistance. Parameters of the RNN-based model were identified so that the response of the closed-loop system can reproduce the actual output acquired through the positioning experiment with human steering wheel operation. The performance is evaluated in both the time and frequency domains, compared to the model based on model predictive control (MPC). As a result, it was confirmed that the RNN-based model reproduced human behavior with almost all the same accuracy as the MPC-based model.