Predicting Spatial Self-Organization with Statistical Moments

We have developed a self-organizing shape formation system based on locally interacting agents whose behaviors are inspired by living cells. Given a predefined macroscopic shape, genetic programming is used to find a finite field function that defines the agents’ interactions. By following the gradient of the cumulative field the agents form into a desired shape. It has been seen that the self-organization process may form two or more stable final configurations. In order to control the outcome of the shape formation process, it is first necessary to accurately predict the outcome of the dynamic simulation. This paper describes an approach to predicting the final configurations produced by our spatial self-organization system at an early stage in the process. The approach calculates statistical moments of the coordinates of the agents, and employs Support Vector Machines to predict the final shape of the agent swarm based on the moments and their time derivatives.