No abstract

We describe the design and implementation of PEG, a networked system of distributed sensor nodes that detects an uncooperative agent called the evader and assists an autonomous robot called the pursuer in capturing the evader. PEG requires services such as leader election, routing, network aggregation, and closed loop control. Instead of using general purpose distributed system solutions for these services, we employ whole-system analysis and rely on spatial and physical properties to create simple and efficient mechanisms. We believe this approach advances sensor network design, yielding pragmatic solutions that leverage physical properties to simplify design of embedded distributed systems.We deployed PEG on a 400 square meter field using 100 sensor nodes, and successfully intercepted the evader in all runs. While implementing PEG, we confronted practical issues such as node breakage, packaging decisions, in situ debugging, network reprogramming, and system reconfiguration. We discuss the approaches we took to cope with these issues and share our experiences in deploying a large sensor network system.

Abstract. An algorithm for computing the maximal controlled invariant set and the least restrictive c o n troller for discrete time systems is proposed. We show h o w the algorithm can be encoded using quantier elimination, which leads to a semi-decidability result for de nable systems. For discrete time linear systems with all sets speci ed by l i near inequalities, a more e cient implementation is proposed using linear programming and Fourier elimination. If in addition the system is in controllable canonical form, the input is scalar and unbounded, the disturbance is scalar and bounded and the initial set is a rectangle, then the problem is decidable.

No abstract

No abstract