Modelling Interactions Between Traffic and Crosswalk Agents at Unsignalized Crossings

Abstract

Agent behaviour modelling is currently one of the most challenging topics in the autonomous driving industry meaning that more research is needed in order to understand how autonomous cars should be interacting and behaving around pedestrians and cyclists. The research includes modelling vehicle-bicycle interactions based on hierarchical modelling framework [1] generating different features mentioned in literature and some proposed by author, selecting relevant ones using F value and finally creating a simple logistic regression model for estimating the drivers yielding probability. As a result, a model with following features is suggested: bicycle distance from conflict zone when vehicle enters the interaction zone, vehicle mean speed inside interaction zone, minimum speed distance and minimum speed inside interaction zone. The research also indicates that driver yielding probability is reduces with higher vehicle speed and higher bicycle distance. Furthermore, the data indicates that drivers who perceive a conflict but do not yield experience the highest mean speed referring that some vehicles do accelerate in order to avoid giving way. Additionally, the low variation in bicycles’ speed indicates cyclists’ assertive behaviour and perhaps expectations for the vehicle to yield.