Robust data-driven dynamic model discovery of industrial robots with spatial manipulation capability using simple trajectory

Industrial standard model-based controllers need an accurate dynamic model to perform reliably. The classical technique for dynamics modeling of industrial robots relies on deriving the closed-form dynamic equations and then identifying the inertial parameters on the assumption that all the kinematic and geometric details of the robot are known. This study extends the recently developed data-driven SINDy method to obtain the dynamic model of industrial robots with spatial manipulation capabilities. The proposed method is applied to the 6-DOF UR10 robot in simulated and experimental environments. Furthermore, payload and friction are considered in both situations. Contrary to other techniques that rely on optimizing the trajectory to improve the regressor matrix conditionality, the proposed method uses a simple trajectory constructed using a combination of quintic polynomial and sinusoidal trajectory. Three distinct trajectories are used to assess the accuracy of the proposed method in modeling the robot’s dynamics. Although simple trajectories result in an ill-conditioned regressor matrix, the proposed method accurately identified the dynamic model. The identified dynamic model has accurately replicated the performance of the robot in both simulation and experimental settings. The model parameters obtained by the proposed method in the simulation settings are compared with the parameters of the Euler–Lagrange closed-form model, and good conformance has been found. To assess the robustness of the proposed method, different levels of noise are added to the simulation data. For noise levels in the range of σ =0.001 to σ =0.08 , the method was able to converge to a model that replicates the generalized torque curve obtained by the simulation.