In order to improve active safety technologies, some researchers have investigated adaptive assistance systems designed by modeling individual behavior of the assisted driver. However most of them could not build applicable model without large individual driving data sets, because they use only his/her database. We attempt to solve this problem with refering the data of another drivers whose driving behavior is similar to the analyzed driver’s. We call this approach “collaborative learning” and construct this method based on Dirichlet process mixtures [Escober et al. Journal of the American Statistical Association 1995].
