Accurate and versatile multivariable arbitrary piecewise model regression of nonlinear fluidic muscle behavior

Wearable exoskeletons and soft robots require actuators with muscle-like compliance. These actuators can benefit from the robust and effective interaction that biological muscles' compliance enables them to have in the uncertainty of the real world. Fluidic muscles are compliant but difficult to control due to their nonlinear behavior. Precise control of these actuators needs accurate models that readily capture this behavior. Here we present the multivariable arbitrary piecewise model regression (MAPMORE) algorithm for automatically creating accurate data-driven, behavior-based models for fluidic muscles. MAPMORE integrates an arbitrary term dictionary based orthogonal forward regression algorithm with piecewise function fusion. We modeled the static and hysteresis force components of a McKibben pneumatic artificial muscle (PAM) and a Peano muscle with MAPMORE, Sárosi's empirical model, and a polynomial model. In all cases, MAPMORE's models had the best mean accuracy of below 15N. This shows it to be an easy to use, accurate, and versatile soft fluidic actuator modeling tool.