Nonlinear Dynamical System Identification Under External Disturbances by Maximum a Posteriori (MAP) Estimation for Robotics

Daisuke Yanabe, Suguru Kanoga, Vincent Frémont, Toshiyuki Murakami

Abstract: In nonlinear dynamical system identification, a flexible statistical model is commonly used for identification to compensate for non-linear elements. The flexibility in the statistical model induces a high sensitivity to the existence of external disturbances that could deteriorate the identification accuracy substantially. Most of the identification techniques assume a processing noise as a statistical variable and apply a maximum likelihood-based method to identify the dynamical system. Most of these methods do not take into account the effect of external disturbances. In this work, we also considered external disturbances as one of the statistical parameters to estimate. By estimating values for external disturbances and applying maximum a posteriori estimation, identification accuracy could be improved compared to models without considering external disturbances.