In connection with vehicle control tasks, the research topic would like to examine in what context and how it is possible to achieve a control goal directly from the measured data (design) and to examine the quality of control already planned and implemented (performance analysis and validation). A new area of research is the development of state-of-the-art controls based on machine learning and efficient computational and optimization methods. The aim is to combine classical control design procedures with learning structures for real-time implementation.
Our goal is to develop modeling procedures that are able to combine the statistical reliability of current identification procedures with the efficiency and flexibility of machine learning-based methods. A key goal is to automate the entire modeling process (toolchain) from data collection to verification and modeling, both to preserve physical interpretability and to map unknown, physically difficult-to-describe, correlations and dynamic features.
Developing optimal control through machine learning.
Supporting complex, computationally intensive nonlinear optimization tasks with machine learning for simpler real-time implementation.
Development of self-learning, physically interpretable and adaptive modeling algorithms and related experiment design and verification procedures, their implementation on real systems, demonstration of industrial applicability.
Initiation of methodological research on reinforced learning. Design and teaching of RL structures using a continuous intervention signal. Implementing trajectory tracking with the help of a design agent and using forward-looking sensor information. Managing highway traffic using an RL learning agent.
Research on optimal trajectory design and execution algorithms, combining machine learning and modern control theory. Development of a designer using reinforced learning and an executive using model predictive control. Guarantee robustness and safety by using linear variable parameter control.
Coordination of charging, availability and scheduling of AGVs:
In the case of autonomous transport vehicles (AGVs) performing internal logistics tasks, the basic requirement is to ensure that the vehicles are properly charged and available, even with changing needs. With the help of calculations and simulations performed on the digital twin model of the transmission system, the goal is to define a scheduling policy that implements the coordination of the system operations according to the above, opposite criteria.
In addition to the simulation studies performed on the digital twin model, real experience can be gained by implementing the charging with docking of autonomous vehicles operating in the SZTAKI Győr Industry 4.0 sample system. The goal here is to implement a docking station that also allows calibration of the positioning system. This allows the digital twin model of the entire system to be updated from time to time with information from physical measurements.