A Study on switching AR-HMM driving behavior model depending on driver's states

To offer appropriate information, predicting driver's behavior is effective because we can show information only when they are useful, such as missing stops or cross to turn. But driver's behavior depends on his or her mental states, so it is better to predict driver's behavior considering his or her states. In this paper we focused on driver's "hasty state", and induced it with quota to run using driving simulator. And we found that driver's "hasty state" has effect on AR-HMM model parameters such as gas pedal stroke and brake pedal stroke, and also on autonomic nervous activity. As a result, we could show possibility to improve prediction accuracy by switching driving behavior model depending on driver's states.