Research presentation: Spiridon Reveliotis, Georgia Institute of Technology
June 10 - April 10, 2019
9:00 AM - 10:00 AM
In many contemporary applications, automation is called to play an integrating role that can be effectively abstracted as a complex resource allocation function administering a finite set of reusable resources to a set of concurrently executing processes. Specific examples of such a function include the real-time management of the processing and the material handling equipment in flexibly automated manufacturing systems, the traffic management of guidepath-based zone-controlled transport systems through the pertinent allocation of the various guidepath-zones, the allocation of the supporting resources in internet-enabled workflow management systems, and the management of the semaphore allocation in multi-threaded software.
The autonomy, concurrency and flexibility that are sought by the aforementioned environments render the management of the corresponding resource allocation function a pretty challenging task. A complete solution to the abstracted control problem must ensure the efficiency of the resulting operation according to traditional performance measures like throughput maximization and congestion minimization, but it must also provide explicit guarantees for safety and other requirements of “behavioral correctness”. In addition, any pursued solutions must be efficiently implementable during the real-time operation of the considered systems, in view of the aforementioned operational and computational complexities.
This talk will overview a series of results from the presenter’s research program that have sought to provide a rigorous analytical solution to the aforementioned control problem, building upon the formal abstraction of “Sequential Resource Allocation System” and the control-theoretic framework of Discrete Event Systems. The considered results enable (i) a rigorous characterization of the underlying control problem, (ii) the pertinent decomposition of this problem into a number of subproblems that can be rigorously addressed by the current DES theory, (iii) the definition of pertinent notions of “optimality” for each of these subproblems and a characterization of their complexity, and (iv) the establishment of an effectively controllable trade-off between optimality and computational tractability, when the former is proven computationally (too) challenging.
Date posted
Apr 3, 2019
Date updated
Apr 3, 2019