Project Details
Projekt
Integrated hybrid optimisation of autonomous self-adaptive systems (InHOSaS)
Subject Area
Computer Architecture, Embedded and Massively Parallel Systems
Term
since 2023
Project identifier
Deutsche Forschungsgemeinschaft (DFG) - Project number 516601628
Self-adaptive and self-organising (SASO) systems have been established in response to the ever increasing complexity in information technology. A SASO system comprises several subsystems that act with a local scope: Control is either issued globally by a centralised element or locally by each autonomous subsystem. InHOSaS claims that neither a global optimisation mechanism responsible for controlling all contained subsystems nor a fully local and autonomous decision mechanism is desirable. In contrast, an integrated hybrid approach has to combine global optimisation with the subsystems’ autonomy. Consider platooning of vehicles as example of the challenges: A platoon is a coordination of autonomous vehicles, where the participants experience benefits such as less energy consumption and cost from the coordination depending on the position in the platoon. Optimised global planning may result in the best trade-off between local goals, preferences, and environmental conditions but the autonomous participants may ignore the plan and render the process infeasible. On the other hand, a fully decentralised scheme will not identify globally optimal solutions due to communication and neighbourhood restrictions. To overcome these limitations, we propose a fundamentally different way of developing SASO systems based on the idea of integrating macro-level planning under longer time requirements with micro-level decisions that are subject to autonomous learning. We expect that such an integrated hybrid self-optimisation scheme leads to a new generation of SASO systems characterised by: a) increased robustness against faulty or even intentionally wrong behaviour through flexible plans, b) an improved utility of both, the subsystems and the overall system, and c) a fast adaptation to changes in the characteristics of the learning problem. This requires research from two perspectives: bottom-up (i.e., starting with fully local coordination of instances and increasingly integrating centralised planning) and top-down (i.e., starting with the assumption that all participants will follow the optimised solution and then relaxing these assumptions). To address the vision above, the project InHOSaS investigates and evaluates the fundamental concepts within the aforementioned platooning application. In order to generalise the insights, we transfer the developed mechanisms to existing use cases from the traffic management, industrial automation, and mobile code offloading domains.
DFG Programme
Research Grants
International Connection
France, Italy
Cooperation Partners
Professorin Ada Diaconescu, Ph.D.; Professor Dr. Michele Segata
