Examining DIDIM and IV Approaches through the Double Least Squares Method for Comparative Assessment

Typically, the process of determining dynamic parameters for robots entails the utilization of the Inverse Dynamic Model (IDM), which exhibits linearity concerning the parameters under identification. This is coupled with the application of Linear Least Squares (LLS) techniques. To execute this method, it is imperative to acquire highly accurate measurements of both torque and position at a high sampling frequency. Moreover, it is necessary to estimate velocities and accelerations by applying a band-pass filtering method to the positional data. Because of the presence of noise within the observation matrix and the iterative nature of the identification process, we have adapted the Instrumental Variable (IV) method to tackle the challenge of noisy observations. A recent successful application involved a novel identification approach called the Direct and Inverse Dynamic Identification Model (DIDIM), which relies solely on torque measurements as its input variables. The experimental findings demonstrate that the IV and DIDIM techniques yield comparable numerical results. In this paper, we undertake a comparative analysis of these two approaches using a double step least squares (2SLS) examination. To substantiate our findings, we carry out experimental validation using a 2-degree-of-freedom planar robot.