Half time seminar: PhD student Lars Svensson
Tid: Ti 2019-06-18 kl 14.00
Plats: Brinellvägen 83, groundfloor, room B242
Respondent: Lars Svensson
Opponent: Mohammad Ali, Zenuity
Automated driving technology has the potential to drastically reduce the numerous road accidents caused by human error. Increasing deployment of driving automation functionality have already started to make an impact in this regard. However, despite intense combined efforts of industry and academia in the last decade, fully automated vehicles (SAE level 4) are not yet deployed at scale and several major challenges remain to be addressed. One such challenge is safe handling of rare critical events. Examples of such events are internal vehicle faults in software or hardware, or unforeseen hazardous events in the traffic scene, such as a pedestrian suddenly appearing in the vehicle path. This research project deals with motion planning and control of automated road vehicles in critical situations. The aim is to identify the common properties of critical situations, and to propose a motion planning and control framework that fully utilizes the dynamic capacity of the vehicle to avoid accidents in such situations. Furthermore, we recognize and consider that the dynamic capacity varies with the present operational conditions, e.g. locally varying traction.
First, we define a motion planning problem for critical scenarios, based on a safety supervisor architecture, and highlight how it differs from the nominal motion planning problem. We propose a solution method and integrate the planner with a supervisory controller that oversees switching between motion planners under formal guarantees. The concept is verified through simulations and full scale vehicle experiments.
Second, we propose a traction adaptive algorithm capable of planning and executing critical maneuvers at the limits of the vehicle's physical capability, in the presence of locally varying traction. In order to realize this, we have developed a planning method called sampling augmented adaptive sequential quadratic programming, that employs time varying polytopic constraints to model the boundary of tire forces. Through extensive simulation studies, we show that the approach maximizes the vehicle's capacity to avoid collision with a suddenly appearing obstacle, in the presence of local traction variations.
Future work towards the PhD is intended to investigate how to utilize the time varying constraints concept to extend the framework to classes of internal faults that impact the dynamic capacity of the vehicle. Such faults include (partial) loss of brake or steering due to tire failure or minor collision.