This research contributes to the development of intelligent and self-learning supervisory controllers that take into account external information sources to control their vehicle or machine without human intervention.
WHY THIS RESEARCH?
Production cells and machines must be increasingly able to operate autonomously. This allows to use them more efficiently and to create more robust, high-quality processes that consume but a minimum of energy. Furthermore, autonomous systems save time at the start-up and when managing their operations. Finally, safety is increased as human errors are excluded.
Self-driving vehicles are coming our way as well. These are not only able to drive autonomously, they also use cloud data to communicate with the world on road conditions, other vehicles and own performances. This way, self-driving vehicles will make traffic safer and more energy-efficient.
Awaiting full autonomy, maximum support to the operator and driver obviously remains a very important objective. Besides, all autonomous systems benefit from access to databases and real-time information.
CONCRETE RESEARCH OBJECTIVES
To be able to realise autonomous vehicles, machines and production cells, major breakthroughs are needed in the following research domains:
- Model-based optimal control – In a system that has disposal of a physical model, it is perfectly possible to design a supervisory controller that always ensures the optimal configuration of the system, whatever the operational conditions. Still, designing such controller is not at all easy. It requires complex, linked mechatronic systems and is faced with the limitations of existing models and the stochastic nature of many processes. In addition, such controller works best on the basis of efficient, non-complex algorithms. Further research into all these challenges is required.
- Self-learning control system – In complex processes, realistic modelling is impossible. In these situations, a self-learning controller is an attractive option. We already have an extensive theoretical basis for these controllers. But their integration in vehicles and production systems is still in its infancy. Further research is required, particularly for ensuring the necessary robustness and speed.
- Situation modelling – A supervisory controller will always be connected to external information sources: other vehicles, smart grids, traffic systems, etc. The information from these external sources is so fragmented and uncertain that information fusion in the controller is crucial for its optimal functioning. At the same time, research into the protection of such communication links is absolutely necessary.
- Shared human/machine control – Fully autonomous systems will but be developed gradually. Meanwhile, an optimal and safe cooperation with human operators and drivers is a top priority. Controllers must learn preferences, know how to deal with changing behaviour, etc. On the other hand, operators can learn from the machine as well.
WHO BENEFITS FROM THIS RESEARCH?
Several Flemish companies benefit from developments in the field of autonomous systems:
- Manufacturing companies that want to transform their own production cells into autonomous environments.
- Machine builders that want to deliver autonomous machines.
- Developers of vehicle or machine parts that wish to integrate their products in autonomous systems.
- Developers of devices that want to increase the interaction with users.
- Information providers that want to build on the connection of their systems with vehicles and industrial production cells.
INVOLVED RESEARCH DEPARTMENTS OF FLEMISH UNIVERSITIES
- Model-based controllers: KULeuven-PMA, UGent-SYSTeMS
- Learning machines: KULeuven-PMA, KUleuven-Mebios, KULeuven-CW, VUB-COMO
- Optimisation: KULeuven-PMA, UGent energy efficiency lab
- Information fusion: VUB-ETEC/MOBI, UA-CoSys-Lab
- Interaction with users: KULeuven-PMA, KULeuven-CIB, VUB-R&MM