Responding to changing situations: Learning automata for sensor placement

Security issues have received increasing attention in recent years. Due to the difficulty of predicting where a terror event will occur, it is a great challenge to develop methods of detection that preempt attack. Sensors are one such method. However, questions of effective sensor placement remain - it is hard to determine where to place sensors because of uncertainty over the location of an attack. In this paper we use the intruders' behavioral constraints in the face of environmental factors as input to a learning algorithm that optimizes sensor placement. We show through simulation results that this algorithm can dynamically optimize placement by letting sensors make local decisions about where to move in situ. The resulting configurations are more or less equivalent to those achieved by the global optimization of sensor placement. The technique is superior in the sense that re-optimization happens continuously, and can be done with distributed control. Also, in many situations the configurations achieved are better than spacing sensors equally: detection rates are far higher.