In this paper we present three Swarm Intelligence algorithms which we evaluate on the complex foraging task domain. Each of the algorithms draws inspiration from biologic bee foraging/nest-site selection behavior. The main focus will be on the third algorithm, namely STIGMERGIC LANDMARK FORAGING which is a novel hybrid approach. It combines the high performance of bee-inspired navigation with ant-inspired recruitment. More precisely, navigation is based on Path Integration which results in vectors indicating the distance and direction to a destination. Recruitment only occurs at key locations (i.e., landmarks) inside of the environment. Each landmark contains a collection of vectors with which visiting agents can find their way to a certain goal or to another landmark in an unknown environment. Each vector represents a local segment of a global route. In contrast to ant-inspired recruitment, no attracting or repelling pheromone is used to indicate where to go and how worthwhile a route is in comparison to other routes. Instead, each vector in a landmark has a certain strength indicating how worthwhile it is. In analogy to ant-inspired recruitment, vector strength can be reinforced by visiting agents. Moreover, vector strength decays over time. In the end, this results in optimal routes to destinations. STIGMERGIC LANDMARK FORAGING proves to be very efficient in terms of building and adapting solutions.
Implementing TVA as a Bayesian classifier in a foraging task
