The Budapest University of Technology (BME) Department of Automotive Technology (GJT), BME AutomatedDrive, Vehicle Dynamics and Control research group, led by Dr. Ádám Bárdos and SZTAKI's Research Laboratory on Engineering & Management Intelligence, Intelligent Processes research group, led by Dr. Zsolt János Viharos, have been collaborating for several years to investigate how to adaptively control the motion control and drifting of self-driving vehicles at the limit of adhesion using artificial intelligence and learning algorithms, adapting to the current environmental conditions. SZTAKI and BME are partners in the National Laboratory for Autonomous Systems coordinated by SZTAKI. ZalaZONE is an industrial partner of ARNL.
The primary objective is to increase the safety of self-driving vehicles, to ensure the controllability of vehicles moving at the limits of stability (e.g. in rain, snow or when cornering at excessive speed), even by drifting. The researchers aim to solve the problem of drift control using reinforcement learning algorithms, which is the current main goal of the research. Previously, they have had several joint results to solve this problem in a vehicle simulation environment, first starting from a drifting state to maintain it, then, after successfully achieving this, from a simple turning and finally from a straight driving state, they managed to control the vehicle to a drifting state (see publications). With these results, BME research student Hunor Szilárd Tóth Szilárd won first place in a TDK last year.
For the testing day, the joint team travelled to Zalaegerszeg to the ZalaZONE test track. As a first step, they presented the current status of the research to Dr. Zsolt Szalay, head of the BME GJT department and ZalaZONE Research and Innovation team, and discussed future plans. Presentations were also given by other research students. Jeffrey Boahene Kwaku and András Burján demonstrated graphical interface and self-driving real-time software functions developed on NXP i.MX8 hardware and Blackberry QNX operating system platform.
After the presentation of the students' results, the team went out to the test track, the so-called Dynamic Plaform. During the tests, the vehicle calculated the manoeuvre based on differential equations and sensor measurements (developed by Adam Domina) and performed the required drifting. At the end of the day, the researchers "took back the controls" from the self-driving vehicle and tried to perform the drifting operation themselves, in the same way as the learning agent had already demonstrated it to them earlier. Surprisingly, the researchers were unsuccessful: no one could achieve, or even approach, the drift manoeuvre that the computer and associated actuators (steering control and computer-controlled accelerator) had easily achieved.