Integration of a physical system, machine learning, simulation, validation and control systems towards symbiotic model engineering

Abstract

System simulation without detailed prior knowledge or data of the system is a complex challenge. In this paper we present an approach to automatically generate a model on the fly in a symbiotic way. Basically the data based model generation system introduced is an agent based evolutionary optimization system creating continuous differential equations from simple predefined operators. The well known paradigm of symbiotic simulation is then enhanced with this agent based machine learning system. Here we focus on the emergent behavior of the model generation system resulting from the interaction of multiple agents optimizing a common model and the effects arising from the direct coupling and steering of the connected physical system. Different emergent mechanisms and effects can be observed speeding up the model generation process. To measure and evaluate this effects multiple experiments with a robotic system are discussed.