On the robustness of force estimation methods for robot manipulators: An experimental study

Abstract

Force control of robot manipulators has been extensively studied for the last decades. Whether a direct or an indirect approach is pursued, it is always desirable to know the contact forces/torques, which are usually attained through force sensors. However, there are a variety of reasons for which it may not be desirable to employ them. For instance, they are expensive and not free of noise, and for some applications, they might just be too big to be a practical solution. Following these reasons, there have been many efforts to develop estimation techniques. Some of the main challenges are the need of exact robot models as well as position, velocity, and even acceleration measurements. In this work, a compilation of some of the most employed approaches developed in the last two decades is presented, along with the design of a novel force estimation algorithm. The proposed method is a modification of the well-known Generalized Momentum approach, with the advantages of finite-time convergence and improved performance. A comparative analysis is carried out to elucidate which of the methods delivers a better performance based on two key assumptions presented at the experimental implementation: 1) the robot model is accurate, and 2) the estimated forces/torques should be employed in a control scheme without a previous closed-loop stability analysis. Both assumptions are meant to test the robustness of the different approaches since the former is not realistic, while the latter is not advisable because no separation principle is valid for nonlinear systems. However, from a practical point of view, it is interesting to substitute force measurements with no further analysis. The experimental outcomes show closed-loop stability in all cases.