Simon Ware

Incremental scheduling of discrete event systems

A common problem in the field of robotics is how to coordinate the motion of multiple robots. As this problem is PSPACE-complete it is generally solved by incomplete algorithms which are computationally tractable but cannot guarantee that a solution will be found even if a solution exists. In this work we use the coordination method of prioritized planning in a (Discrete Event Systems) DES context. We develop and implement an algorithm for using prioritized planning in DES. The algorithm is showned to be an improvement over the standard prioritized planning approach in terms of being capable of finding a valid solution to the multi-robot motion-planning problem in a greater percentage of instances.

Time Optimal Synthesis Based Upon Sequential Abstraction and Its Application to Cluster Tools

The Ramadge–Wonham supervisory control paradigm has been shown effective in dealing with logic control. Nevertheless, time-related performance is always one of the major concerns in industry. Current methods for synthesizing time optimal supervisors are incapable of dealing with large discrete-event systems (DESs) with massive state spaces. This paper proposes an approach for finding a time optimal accepting trace for large DESs based upon sequential language projection, and pruning. The algorithms are tested on a linear cluster tool to show their effectiveness.

Progressive time optimal control of reactive systems

In many practical applications, we need to compute a nonblocking supervisor that not only complies with pre-specified safety requirements but also achieves a certain time optimal performance such as maximum throughput. Previous time optimal synthesis methods have had the significant drawback of only being able to deal with acyclic behavior. In this paper we propose a time optimal synthesis method which can deal with potentially infinite behavior. This is accomplished by dividing a system into progressive behaviors with which the system must be capable of finishing it task within a finite amount of time, and nonprogressive behaviors which can lead to the system having more tasks to complete, and potentially infinite behavior. We also proposed an algorithm with pseudo polynomial time complexity for computing such a supervisor.

Reduction of distributions and its applications

In this work, a notion of reduction of distributions is proposed as a technical tool for improving the complexity of decomposability verification and supporting parallel verification of decomposability, by exploiting the rich structures of distributions. We provide some results that reduce the search space of candidate reductions, as a first step towards efficiently computing optimal reductions. It is then shown that a distribution has a reduction if and only if a particular candidate reduction is indeed a reduction. We then provide a sound substitution-based proof technique that can be used for (automatic) reduction verification. Techniques for refuting candidate reductions are also provided. We then explain an application of the decomposability verification problem in the lower bound proofs for the problem of supervisor decomposition and the problem of existence of decentralized supervisor.

Synthesis time optimal accepting traces using language projection and pruning

The Ramadge-Wonham supervisory control paradigm has been shown effective in dealing with logic control. Nevertheless, time-related performance is always one of the major concerns in industry. Current methods for synthesizing time optimal supervisors are incapable of dealing with large discrete event systems with massive state spaces. This paper proposes a method of finding a time optimal accepting trace for discrete event systems. It also proposes abstraction methods based upon language projection, and pruning capable of finding such traces for large discrete event systems. The algorithms are tested on a large discrete event system model in order to show their effectiveness.

Reduction of Distributions: Definitions, Properties, and Applications

In this work, a notion of reduction of distributions is proposed as a technical tool for improving the complexity of decomposability verification and supporting parallel verification of decomposability, by exploiting the rich structures of distributions. We provide some results that reduce the search space of candidate reductions, as a first step toward efficiently computing optimal reductions. It is then shown that a distribution has a reduction if and only if a particular candidate reduction is indeed a reduction. We then provide a sound substitution-based proof technique that can be used for (automatic) reduction verification. Techniques for refuting candidate reductions are also provided. We then explain an application of the decomposability verification problem in the lower bound proofs for the problem of supervisor decomposition and the problem of existence of a decentralized supervisor. Finally, some other applications of the notion of reduction of distributions are also shown.

Comparing MILP, CP, and A* for multiple stacker crane scheduling

This paper describes an optimisation model for the scheduling of a system consisting of three stacker cranes that are restricted to the same track. To improve the efficiency of the solution methods, a novel simplification of the model is presented, which has a low impact on the quality of the solution but greatly decreases its complexity. This model is then used to benchmark several popular solution methods, including both optimal and approximate methods. Some are based on monolithic models, whereas others solve the problem in phases by using sub-problem formulations. The result presented in this paper shows that evaluated solution methods have complementary strengths and weaknesses. Constraint Programming (CP) is very efficient on small scale problems, while Mixed Integer Linear Programming (MILP) scales much better when the number of movement orders increases. However, none of these methods are able to solve large instances of the problem to optimality. To handle the complexity of the problem, approximate solution methods are the only viable option. In this paper we show that promising results can be obtained even with simple methods using well known search algorithms such as A* and Tabu-search. However, preliminary results on more advanced search algorithms show that further improvements may be achieved, allowing the solution of very large problem instances.

Region-based dynamic forecast routing for autonomous vehicles

In this paper, we aim to implement an urban network-level traffic routing scheme for autonomous vehicles to mitigate congestion in urban areas. We first present an implementation of a region-based dynamic traffic model. Subsequently, we present an innovative predictive routing approach for autonomous vehicles, dynamic forecast routing, and apply it on a set of homogeneous regions, making use of Urban-scale Macroscopic Fundamental Diagram to define the state of congestion in each region. This approach is compared with periodically adjusted routing, as well as logit routing, which is used to model self-interested human driver behavior. We also integrate a public transit diversion mechanism on all approaches. Results show that our approach surpasses all other approaches on all performance metrics by at least 40%.