摘要：AbstractIn this work, we propose control improvisation to synthesize voluntary lane-change policy that meets human preferences under given traffic environments. We first train Markov models to describe the lane-change environment for an automated vehicle under assumed traffic patterns. Parameters in the environment model are trained using traffic data and calibrated using control improvisation. Then, based on human lane-change behavior, we train a voluntary lane-change policy using explicit-duration Markov decision process. Parameters in the lane-change policy are calibrated through control improvisation to allow an automated car to voluntarily change lanes while avoiding overly frequent lane-change maneuvers under various traffic environments.
