Learning haptic feedback for guiding driver behavior

Information about the driving state can be conveyed to automobile drivers through force feedback signals sent via the pedals and steering wheel. Because the set of possible haptic signals and driver responses is huge, it is desirable to automatically learn which signals are most useful to drivers. Thus, it is instructive to explore how machine learning techniques can be used as a step in the design of a haptic interface system. In this paper, we present a learning algorithm that learns useful haptic feedback and apply the algorithm to learning feedback for automobile drivers. We present evidence to show that the algorithm is sensitive enough to learn useful feedback under some circumstances, but that its scope may be limited by people's ability to act as admittance controllers.