Model-based control with quantitative feedback theory; empirical model analysis

An empirical approach to developing the linear time-invariant models necessary for quantitative feedback theory (QFT) robot controller design is presented. A Golubev curve fitting routine produces transfer functions based on the actual input/output empirical data from nonlinear robot system experiments. These linear time-invariant models are equivalent to the nonlinear model with respect to the restricted inputs used in their development. The empirical models are used to analyze several implementation issues for model-based control with quantitative feedback theory (MBQFT). A PUMA-560 is used for the case study. The resultant empirical models produce transfer functions with higher gains than transfer functions produced by the use of linearized Lagrange-Euler dynamic equations. While the empirical models are too complex to replace the analytical models currently used in MBQFT design, their higher gains explain previously observed QFT robot control phenomena. It is concluded that intuition gained from this analysis will aid the continuing development of better models for QFT robot controller design