Online Inverse Linear-Quadratic Differential Games Applied to Human Behavior Identification in Shared Control
Jairo Inga, Andreas Creutz, Sören Hohmann
Abstract
In this paper, we propose an inverse differential game method for the online identication of cost function parameters in a shared control situation, where several players simultaneously control a dynamic system. The method is based on the sequential processing of state and control pairs to first estimate control law matrices of an infinite-horizon linear-quadratic differential game and the subsequent calculation of corresponding cost function parameters based on the Algebraic Riccati Equations. A second inverse optimal control method from literature was extended in order to benchmark our approach. Their performance is evaluated and compared via simulations. Afterwards, we use data from a steering experiment where a pair of subjects cooperatively track a trajectory with coupled force-feedback steering wheels. Our approach shows good results concerning parameter estimation and robustness to noise as well as good state trajectory reconstruction from estimated parameters for the experimental results.