Vehicle Control and Machine Learning Research Group

The Vehicle Control and Machine Learning Research Group deals with two research areas in the field of highly automated vehicles. The first area is vehicle dynamics modeling and control using classical and robust control techniques and machine learning. Research is carried out at several levels of motion planning, from strategic decision-making through various tasks of trajectory planning to low-level control. Our narrower area of study within machine learning is research on reinforced learning methods, where, unlike supervised learning, no training data is available. In this case, the control planning or decision-making agent can learn the expected behavior in a simulated environment modeled as a Markov decision process based on self-generated experience. We carry out basic research in the field with the further development of various agents, for example, in planning agents or multi-agent systems, and use them in the automotive industry. Thus, we have achieved new results in traffic management, motion planning in a static and dynamic environment, and trajectory tracking tasks. The second area is environmental perception and sensor fusion in the automotive industry, based on supervised learning and Bayesian inference. The goal here is to process the measurements provided by the radar, camera, and lidar sensors, loaded with various errors and noises, and create the most stable and accurate environmental representation possible. Our research aims to generate linearly scalable models that also have a real-time implementation capability. These supply essential information for vehicle control planning. The generated data sets can be used to estimate the probability of vehicle movement on a model basis or using supervised learning.