In order to develop accurate dynamic models and to implement new control algorithms, a lot of research has focused on identifying dynamic parameters of manipulator robots. The identification of the system, in general, is related to the construction of models that mathematically describe the physical process. These models are characterized by a structure and by dynamic parameters for which numerical values are unfortunately not entirely provided by the manufacturers.

Applied successfully on several robots and industrial prototypes, IDIM_LS is the method chosen to identify the robot parameters. It allows to estimate the parameters using the inverse dynamic model and the least square, knowing the torque and the articular positions. We have noticed that the vector identified using the least squares method has unrealistic values, which do not correspond to the physical meaning of the parameter. For this reason, a numerical optimization has been performed using the vector of parameters previously estimated as an initial vector i.e. IDIM_LS is a preliminary step. This optimization is based on estimating parameters by minimizing different criteria. The first criterion is the quadratic error between the measured and estimated torques. The implementation of numerical optimization is done with the "fminsearch" function of the MATLAB software. The inertia matrix of the robot must be strictly positive. Therefore, this second criterion has been added to the optimization program. In the literature, there are several methods for checking the second constraint. We have opted for the following steps: In the first step, we checked the positivity of the eigenvalues of the inertia matrix along the identification trajectory. In the second step, using the optimal estimated parameters derived from the first step, we checked the criteria throughout the workspace by modifying the path of the robot.

In our study, the experimental work was carried out on the ABB IRB 14000 collaborative robot named YuMi. YuMi is the first secure collaborative industrial robot with two arms with 7 degrees of freedom. This two-armed ABB robot is ideal for assembling small parts with a load capacity of 500g for each arm. The two arms of the robot are the same manipulator of 7 DOF with two different bases. For this reason, it is sufficient to identify the parameters of a single arm.

During the optimization iterations, the poorly identified parameters that did not affect the calculation of the estimated torques were eliminated. This allowed us to reduce the number of basic parameters from 71 to 63. These essential basic parameters simplify the dynamic model of the robot and minimize the errors between this model and the real dynamic model of the robot. Numerical values will be presented to help interested researchers to develop and improve the results of this study.