Lead researcher:

     Péter Gáspár, DSc, Professor, Head of Research Laboratory, SZTAKI
     Zsolt Szalay, PhD, Associate Professor, Head of Department, BME
     Ferenc Szauter, PhD, Associate Professor, Office Manager, SZE

Background

  • The most important international manufacturers – Audi, Mercedes, Opel, Toyota to name a few – and suppliers of the automotive industry, such as Bosch, Knorr-Bremse, and Thyssen-Krupp, have built up significant R&D centres in Hungary.
  • It is essential for Hungary in many respects to be an important actor in the vehicle industry in the European and global arena, and to reap success in projects in the field, thereby increasing the national productivity.

Basic research directions and activities applicable to the field

  • The fundamental research connected to the NLAS’s R&D directions and activities is on one hand indispensable for the solution of the various complex application tasks occurring in the field. On the other hand ensures visibility and reputation of the NLAS’s researchers within the international scientific community with regards to their existing and expected prime research results fitting the international research trends and satisfying the expectations of the research community.
  • System modelling, model reduction, model identification, furthermore control for adaptive, robust, as well as for distributed and networked systems. The investigated control methodologies include: the linear parameter varying (LPV), the model predictive (MPC) methodologies, the application of reinforcement learning for the purpose of learning optimal control from data, development of real-time capable methods in particle for trajectory planning, and for sensor fusion in the context of environment detection around the ego-vehicle.
  • Machine learning and the modern control methods relying on efficient computing and optimisation methods constitute a new research field within the system and control theory. In such methods, their developers aim to augment classical control design methodologies with computing structures that are capable of learning, so that certain traditionally timeconsuming complex control problems become manageable, i.e., the respective control methods become real-time capable. Much as covetable such combined methods and control architecture are, the stability and performance guarantees are very challenging to provide for them.

Research and development directions and activities in the field

  • Researching and developing control methods applicable to the secure and efficient operation and control of road vehicle components, where these components are required within the vehicular system to reach given levels of vehicular autonomy.
  • Researching and developing control design methods applicable for complex systems formed of and intricate missions accomplished by interconnected autonomous road vehicles, aircraft and mobile robots.
  • Designing, analysing and validating effectual, secure and fuel-efficient optimal control methods applicable for networked and automated road vehicles. In particular, for the purpose of vehicle, pedestrian and obstacle avoidance, overtaking another vehicle, fast and safe crossing of road junctions, adapting vehicle speed to cycles of traffic signs, as well for transport networks.
  • Analysis of cooperative traffic control methods with special regards to the requirements and aspects associated with the efficient, safe and secure transport of humans and goods.
  • Analysis of the effects of various coordinated and cooperative tasks and missions accomplished by private, shared and public vehicles on transport networks, and of the role of the transport management systems within such networks.
  • Advantageous amalgamation of autonomous vehicle control and transport control presume the proper identification of the driver intentions and of the aims set and observed by the vehicle operator. The expected driver behaviours and attitudes, as well as the inferred operator priorities can be incorporated in the control design and in the planning of the vehicle motion, e.g., of the speed profile.
  • It should be noted that the task outlined above is far from being trivial, as in the vicinity of the autonomous ego-vehicle there may be a mixture of autonomous and human-driven vehicles. Therefore, the vehicle control system onboard the vehicle must consider not only the comfortability aims and expectations of the humans within the ego-vehicle – i.e., those of the driver and of the passengers – but also need to harmonize the vehicle’s motion to the aims and expectations of the other nearby human participant of the traffic, while observing the customary driving practice and the commonly accepted rules of transport, which might be difficult to formulate in technical and/or algorithmic terms.
  • The proper interpretation of the ego-vehicle’s environment, as well as the prediction of the behaviour characterising the dynamic components of the scene is an important input to the control design for autonomous systems. Such control design problems cannot be solved in general, application-specific approaches must be used in order to assure proper operation of the target systems. The above behavioural predictions are typically based on Bayesian inference techniques.
  • In the development of the methodology implementing the control strategy, it is necessary to harmonize the aims associated with the nearby vehicles, and those associated with the human participants of the traffic, furthermore, also the aims of the transport in the wider area must be considered. Beyond the complexity of the problem, the ponderosity of the research lies in the fact that only limited information on the humans present in the scene is available for the ego-vehicle. This information, however, could be enhanced by the intelligent control system of the autonomous vehicle by the application of algorithms that recollect past traffic and spatiotemporal patterns and situations to facilitate efficient, safe and secure operation.
  • Great quantity of data – either concerning vehicle dynamics, or describing the surroundings of the ego-vehicle – is produced, transferred, and therefore is accessible within/over the communication and sensor networks of autonomous vehicles. Mining and analysing the mentioned data can contribute to a more precise estimation of the controllability regions of the vehicular and transport systems, and thereby to a better tuning of the control systems involved, as well as to the cooperative operation of the vehicles. A deep theoretical research is set out to incorporate the mentioned data mining results and conclusions into design methods, e.g., in conjunction with the MPC and the LPV control design methodologies.
  • The non-deterministic realization of certain design and operational functions is a novel challenge and also a promising prospect in the design of autonomous systems. Typically, such realizations involve various aspects and combination of heuristics-based optimisation, probabilistic inference, or machine learning. Such realizations tend to provide efficient solutions e.g., in the context of autonomous systems, but provide no robustness guarantees. The development of methods that turn non-robust solutions into robust ones pose a challenge to the researchers of the field, as do also the development of design methods for high-level (supervisory) control.