Application of game-theoretic learning to gray-box modeling of McKibben pneumatic artificial muscle systems

We consider a gray-box modeling of a McKibben pneumatic artificial muscle (PAM) actuated by a proportional directional control valve. This paper presents a hybrid nonlinear model of the PAM system and then proposes a systematic parameter identification procedure that uses a game-theoretic learning algorithm to obtain the appropriate parameter values for the modeling. With a practical example, finally, we verify the proposed method by illustrating a process of searching for the parameter values together with figures of after-and-before learning. As a result, we see that the resulting parameters are better than ones obtained by our previously-proposed heuristic and trial-and-error-based algorithm.