A Heterogeneous Model for Ant Colony Based on Learning Vibrating Potential Field Method

Wenwei Yu, Complex Systems Engineering Department, Hokkaido University, Sapporo, Japan
Email: yu@complex.eng.hokudai.ac.jp

H. Yokoi, Complex Systems Engineering Department, Hokkaido University, Sapporo, Japan
Email: yokoi@complex.eng.hokudai.ac.jp

Y.Kakazu, Complex Systems Engineering Department, Hokkaido University, Sapporo, Japan
Email: kakazu@complex.eng.hokudai.ac.jp

Lots of models inspired by the behavior of ant colonies were proposed and applied to a series of combinatorial problems, such as TSP and Telecommunication Networks etc. [Gambardella and Dorigo, 95], [Deneubourg, 98]. Most of the artificial ant models are homogeneous on two points. The first one lies in the fact that the agents under the models have only one same communicational channel, which may limits the model to simple communicational behavior. The second one is about the homogeneity of the ant agents. That is, the ant agents under the models can hardly have their own individuality, such as sensory ability and action decision mechanism, therefore all the agents have to share the identical action policy. In some cases, this will restrain the ability to model diversities of behaviors of ant colonies or even more complex animal society.

In this research, we propose an nature-inspired heterogeneous model for ant colonies. Owning some consistent elementary characteristics with existed models( e.g. [Colorni et al., 92] ), for example, the positive feed-back, the spatial pattern of transmission media etc., the model extends to a multi-channel communication and an independent action policy for each ant agent. The proposed model can also be understood as the trading-off between the ability to realize diversities of behavior and the simplicity of model. The new model is based on VPM, first proposed by [Yokoi and Kakazu, 94]. A learning version is proposed in [Yu et al., 96]. In this extended abstract, the model was first described in the next section. Then the application and experimental result will be briefly introduced. At last, it is discussed comparing with the some existed ant models.