Robust Inverse Vehicle Map Regression Based on Laplace Distribution

This paper deals with the identification of the relationship between vehicle acceleration and driver available actuators. The vehicle is modeled based on how a driving task is performed. The model is constructed using neural networks whose weights are identified using data collected through non tailored driving sessions. To take into account disturbances, the model follows a Laplace distribution. This leads to a more robust estimate of the vehicle knowledge and the confidence we have in it. The approach is illustrated on a prototype vehicle equipped with a petrol engine, plus a device to actuate the pedals.