Stepwise dynamic parameter identification of serial robots with a nonlinear friction model

Dynamic parameter identification is essential for achieving high-performance robot control. Conventional identification methods are mostly based on linear friction models and therefore cannot accurately capture joint friction characteristics. To overcome this limitation, this paper proposes a stepwise dynamic parameter identification method. To enhance the accuracy of friction modeling, the sign function is replaced with an arctangent function, an exponential velocity term is introduced to describe viscous friction, and two additional coefficients are incorporated to characterize the Stribeck effect of Coulomb and static friction. To reduce identification errors, a reciprocating S-curve trajectory is employed for the pre-identification of nonlinear friction parameters. An iterative weighted estimator that can accommodate the nonlinear friction model is then developed, and the pre-identified results are used as its initial inputs. Experiments on a UR16e robot demonstrate that the proposed method improves joint torque prediction accuracy by nearly 22.35% compared with conventional methods.