Rational swarms for distributed on-line Bayesian search

We present a novel scheme for distributed search in mobile sensors networks that is inspired by collective forms of intelligence present in many biological systems. Unlike the established paradigms of swarm intelligence, we posit a form of individual rationality governing each agent's decision. In the scheme proposed, a network of mobile sensors is tasked to find several targets over a search area. The sensing technology is imperfect so there are non-negligible probabilities for false positives and false negatives. Mobile sensors leave two data 'trails' across potential target locations that have been explored. One trail is associated with the frequency with which a given location has been probed while the other relates to the Bayes updated likelihood that a target is present. These trails are stored in a geographically distributed array of stationary motes. Each sensor processes the implicit information encapsulated in the two trails and chooses a decision that is aimed at maximizing the chance of detecting a target without unnecessary duplication in probing. By endowing mobile sensors with this simple optimization rule, we show that a form of 'rational swarm' intelligence emerges as sensors successfully coordinate indirectly (i.e. they achieve a one-to-one allocation of agents and targets) through active manipulation of the trails. This feature guarantees the proposed scheme is both reconfigurable and scalable.