MENTOR — Model-based Explainable Coordination of Complex Reconfigurations
Complex systems with software and hardware components are often expected to have the ability to self-adapt. They should be able to decide autonomously at runtime in a dynamic environment how best to achieve their goals and implement these decisions by changing their own architecture or adapting their components through reconfigurations.
It is particularly challenging to ensure at design time that a system uses its reconfigurations as ideally as possible. This is especially true if the reconfigurations are not singular, isolated reconfigurations, but coordinated reconfigurations that interact with each other, possibly leading to unexpected or undesired system behavior. Such interactions often impede the ability to analyze and explain decisions of the system.
MENTOR (Model-based Explainable Coordination of Complex Reconfigurations) is intended to address these challenges. The project is conducted in cooperation with the Department of Software Quality and Architecture of the Institute of Software Engineering at the University of Stuttgart, and is designed to run for six years. MENTOR is initially funded by the German Research Foundation (DFG) for a period of three years.
For MENTOR, we are primarily concerned with the following research questions:
- How can effects of reconfigurations on the achievement of objectives be specified and tested?
- How can the coordination of reconfigurations with respect to complex objectives be modeled and automatically planned?
- How can effects of coordinated, reactive or proactive reconfigurations be predicted at design time in a dynamic environment subject to uncertainty?
- How can a high degree of explainability be achieved for the system's decisions about executed (or omitted) reconfigurations?
The results of the project will be evaluated using two demonstrators. First, we consider cloud computing systems where reconfigurations consist of, e.g., provisioning resources such as processors or virtual machine instances in different load situations. As a second demonstrator, we use a swarm of quadrocopter drones as an example of a mechatronic cyber-physical system. This swarm is to perform tasks such as delivering packages, which requires coordination of the quadrocopters.