On decentralizing intelligence in cyber-physical systems

No Thumbnail Available
Journal Title
Journal ISSN
Volume Title
Advancements in the low-cost computing and communication technologies have led to the mass proliferation of devices connected over a network. The networked devices have engendered a new era wherein they sense, compute and share information thereby forming loosely coupled Cyber-Physical Systems (CPS). Managing data and making intelligent decisions form the major domain of research in a CPS. Cloudbased centralized computation has always been the mainstream architecture due to its ease of implementation and enhanced control. However, data explosion, scalability and privacy issues, are certainly pointing toward the limits of such centralized systems. Decentralizing control and distributing the computing among the devices could be a better alternative for sharing intelligence. Investigating new decentralization mechanisms, thus, forms the major crux of this thesis. Realizing such decentralized Cyber-Physical Systems (dCPS) is fraught with challenges such as choosing the appropriate communication method, incorporating the right learning and knowledge sharing schemes, ensuring robustness and adaptivity, and the need for a proper middleware to cater to its functioning.This thesis takes a bottom-up approach and presents its first contribution on extending the functionalities of the Tartarus, a multi-agent platform, in order to realize complex dCPSs. This section begins with the motivation behind Tartarus and discusses features which makes it a disparate environment for developing and deploying dCPS. A real-world CPS application comprising robots, a Raspberry Pi with a camera and a human administrator in-the-loop, described herein validates the feasibility of Tartarus for developing mechanisms to embed decentralized intelligence in CPSs.
Supervisor: Shivashankar B. Nair