Models of the human driving behavior are essential for the rapid prototyping of error-compensation assistance systems. Various authors proposed control-theoretic and production-system driver models. Based on psychological studies various percepts and measures have been proposedfor driver models. These proposals are partly contradictory and depend on special experimental settings. A general computational vision theory of driving behavior is still pending. Resting on the assumption that there is considerable uncertainty about the relevant percepts for realization and classification of natural driving behaviors, we propose the use of machine-learning methods to obtain the relevant percepts in natural driving behaviors from the variety of proposed percepts. In this paper we present the learning of a hierarchical and modular probabilistic driver model, called a Bayesian Autonomous Driver Mixture-of-Behaviors model, representing the lateral and longitudinal human driving behavior on virtual highways. The model is able to autonomously drive on virtual highways, while performing lane-following behavior, car-following behavior, and lane-changes when necessary. Its relevant percepts are selected based on their statistical relevance using a discriminative scoring criterion based on the Bayesian Information Criterion.
Discriminative Learning of a Probabilistic Driver Model Representing the Human Driving Behavior on Virtual Highways
Fahrermodellierung in Wissenschaft und Wirtschaft, Berliner Fachtagung Fahrermodellierung, 4 ; 10-25
2013
16 Seiten, Bilder, Tabellen, Quellen
Conference paper
English
ESTIMATING DRIVER AWARENESS OF PEDESTRIANS FROM DRIVING BEHAVIOR BASED ON A PROBABILISTIC MODEL
British Library Conference Proceedings | 2013
|Driving the aftermath of collision-closed highways: road rage and advanced driver hints
British Library Conference Proceedings | 1997
|