Scenario interpretation for automated driving at urban intersections

Abstract

The research interest in the field of automated driving has been growing rapidly from the first milestones in the 1980s decade to our days. Actually, this interest keeps growing exponentially. The increasing research effort implies a large number of novel approaches, techniques and solutions. Nevertheless, further research is needed to achieve automated driving systems in all different kind of scenarios. Particularly, automated driving at urban scenarios still represents a very complex challenge for the research community.

This thesis deals with the development of a scenario interpretation module for automated driving at urban intersection. The main objective is to interpret the information from the perception module and guide the ego vehicle along the desired path to complete the maneuver.

The first contribution of this work is focused on the analysis of the problem, which yields to a classification of all scenarios considering the potential conflicts with other road users. Based on this statement, the need of a proper pass permission interpretation becomes crucial. The proposed solution in this thesis consists on calculating the probability that every pass permission state is valid for the ego vehicle. This is done by representing all possible traffic lights and traffic signs in two different probability mass functions, which are consequently combined. The developed approach deals not only with the uncertainty of the received inputs, but also with the fluctuations over time.

The concept of primary situations is presented as the core approach for further contributions. The key idea is to interpret the scenario by breaking down the whole desired maneuver into a sequence of primary situations. Using this concept, the object prediction module is significantly simplified. The movement of other involved road users is estimated, and this estimation is used to calculate the probability that every primary situation is occupied. In this thesis, the concept is presented for vehicles and pedestrians.

Another relevant contribution of this thesis is the use of the generated primary situations to perform the subsequent tactical decision making based on a state machine algorithm. The proposed solution enables to handle occlusions in a simpler manner imitating the human reaction.

The implementation of the developed approaches in a demonstrator vehicle and the consequent evaluation confirms that the presented solution is suitable for interpreting the scenario at urban intersections.