Identification of Shepherd Guidance Model Based on Real Shepherding Data -A GMDH Approach-

Shepherding is a typical collective behavior arising from heterogeneous interactions between swarms and individuals of other species. Understanding the herding model by which a few shepherds command a sheep flock can lead to guide robot groups and enable control of artificial objects that indirectly guide swarms. Most previous studies have proposed empirical guidance models based on human observation and analyzed the navigation performance for group models in simulations. Thus few research attempts have been made to reproduce the models for real shepherds. In this study, we aim to estimate a minimal navigation model of the shepherd resembling the behaviors of shepherds by using system identification based on positioning data acquired during actual herding. For this purpose, we adopted the nonlinear system identification method GMDH (group method of data handling). Using thsi method, we constructed estimation functions by natural selection from as many input variables as possible to determine the factors influencing the estimation target.