Ontology-based Reasoning Approach for Long-term Behavior Prediction of Road Users

A human driver determines his/her driving action by predicting future behavior of other road users based on the consideration of relative relationships and reactions between all players. To get as close as possible to human driving behavior, it is important to predict the early and long-term behavior of other road users. For this sake, many motion-based methods has been proposed, although only for short-term prediction. On the other hand, reasoning-based methods for long-term prediction have been proposed recently. However, they can only predict the latest behavior to the observation and do not consider uncertainty, which is necessary to realize smooth autonomous driving behavior. Therefore, we propose a method that predicts long-term behavior of other road users with estimated probabilities based on semantic reasoning that considers interactions among various players. For this purpose, we define an ontology that provides a conceptual description of all road users with their interactions, and infer the likely behaviors with their associated probabilities using Markov Logic Network (MLN). This framework is used with a actual vehicle as it approaches intersections, to demonstrate its applicability. Results from the real-time implementation on the vehicle, operating under controlled conditions, indicate that the proposed approach predicts a sequence of behavioral possibilities with their associated probabilities as expected.