Comparison of linearized dynamic robot manipulator models for model predictive control

Abstract

When using model predictive control (MPC) to perform low-level control of humanoid robot manipulators, computational tractability can be a limiting factor. This is because using complex models can have a negative impact on control performance, especially as the number of degrees of freedom increases. In an effort to address this issue, we compare three different methods for linearizing the dynamics of a robot arm for MPC. The methods we compare are a Taylor Series approximation method (TS), a Fixed-State approximation method (FS), and a Coupling-Torque approximation method (CT). In simulation we compare the relative control performance when these models are used with MPC. Through these comparisons we show that the CT approximation method is best for reducing model complexity without reducing the performance of MPC. We also demonstrate the CT approximation method on two real robots, a robot with series elastic actuators and a soft, inflatable robot.