IRIS-02-411

Global Sequential Control in a Minimalist Distributed System

Chris Jones and Maja J. Mataric

Distributed systems have been shown to be capable of accomplishing globally complex tasks using only local sensing and control. A distributed solution is likely to be more robust, flexible, and scalable than a centralized solution to a given task. The advantages of distributed systems said, a current important disadvantage of minimalist distributed systems is their inability to perform globally coherent sequential tasks using only local sensing and control. Each simple agent in the distributed system does not have access to complete world knowledge or task progress; therefore, sequential control must be achieved solely through the use of local sensing and control. This project investigates extending the capabilities of minimalist distributed systems to accomplish globally coherent sequential tasks. Specifically, the distributed system will consist of simple, behavior-based agents with no inter-agent communication or deliberative capabilities that are, nonetheless, capable of carrying out a global sequentially task. The addition of probabilistic activation conditions in a behavior-based system is experimentally shown to result in increased sequential control in a sequential foraging task.