Machine learning for car navigation

Abstract

In this paper we present the usage of neural networks and hidden markov models for learning driving patterns. We used neural networks for short-term prediction of lateral and longitudinal vehicle acceleration. For longtime prediction, hidden markov models provide recognition of individual driving events. The experiments performed showed that both techniques are very reliable. Recognition rate for driving events is above 98% and prediction error for events in the near future is very low. Predicted events will be used to support drivers in solving guidance navigation tasks.