Abstract
This paper presents a method for input selection for driver behavior model using neural networks. The selection of inputs can vary under the driver's situations. The driver behavior model is represented by a modular network with the input selection mechanism. The modular network consists of a gating network and some expert networks. The gating network learns from data autonomously the temporal segmentation of behavior which corresponds to driving decisions such as car-following or braking. The expert networks are self-organized corresponding to the behavior. The input selection mechanism consists of driving decision prediction and input selection for driver behavior model. In the experiments using a driving simulator, it was shown that the modular network with input selection mechanism learned the subject behavior from data and the input selection worked for each expert network.
