Date of Award

2016

Degree Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

Graduate Group

Computer and Information Science

First Advisor

Rajeev Alur

Second Advisor

Ufuk Topcu

Abstract

With growing complexity of systems and guarantees they are required to provide, the need for automated and formal design approaches that can guarantee safety and correctness of the designed system is becoming more evident. To this end, an ambitious goal in system design and control is to automatically synthesize the system from a high-level specification given in a formal language such as linear temporal logic. The goal of this dissertation is to investigate and develop the necessary tools and methods for automated synthesis of controllers from high-level specifications for multi-agent systems. We consider systems where a set of controlled agents react to their environment that includes other uncontrolled, dynamic and potentially adversarial agents. We are particularly interested in studying how the existing structure in systems can be exploited to achieve more efficient synthesis algorithms through compositional reasoning.

We explore three different frameworks for compositional synthesis of controllers for multi-agent systems. In the first framework, we decompose the global specification into local ones, we then refine the local specifications until they become realizable, and we show that under certain conditions, the strategies synthesized for the local specifications guarantee the satisfaction of the global specification. In the second framework, we show how parametric and reactive controllers can be specified and synthesized, and how they can be automatically composed to enforce a high-level objective. Finally, in the third framework, we focus on a special but practically useful class of multi-agent systems, and show how by taking advantage of the structure in the system and its objective we can achieve significantly better scalability and can solve problems where the centralized synthesis algorithm is infeasible.

Share

COinS