Experiments in Optimal Sensing for Situated Agents

The design of systems, or agents, situated in dynamic environments is of considerable practical and theoretical importance. In this paper, we describe experiments examining the e cacy of static sensing policies when the time cost of processing sensor information is signi cant. Extending the Tileworld experiments performed earlier by Kinny and George , we demonstrate that several expected features of sensing cost do arise in empirical tests. In particular, we observe that for a given sensing cost and degree of world dynamism, an optimal sensing rate exists, and we show how this optimal rate is a ected by changes in these parameters. A mathematical model describing the data is presented and used to predict optimal sensing rates. Our results indicate that static sensing policies can be successful, provided that the rate of change in the environment and the cost of sensor processing do not vary too greatly.