Comparison study of model-based and non-model-based robot controllers

In this paper, model-based and non-model-based robot controllers are examined through theoretical analysis and experimental studies. Three controllers are studied in this paper: a classic PD controller which does not use a robot dynamic model; the computed torque controller which requires an exact robot dynamic model; and a new adaptive controller which works with unknown or partially known robot dynamics. Those three controllers are analyzed theoretically and then tested through experimental studies performed on a PUMA 560 robot. The paper concludes that a model-based controller gives better tracking performance than a non-model-based one, but the quality of a model-based controller, such as the computed torque controller, degrades dramatically if the exact dynamic model is not available. The model-based adaptive controller offers both the high performance of a model-based controller and the robustness of a non-model-based controller. Computational aspects of the controllers for real-time implementation are also discussed