International audience—Usually, the joint transmission friction model for robots is composed of a viscous friction force and of a constant dry sliding friction force. However, according to the Coulomb law, the dry friction force depends linearly on the load driven by the transmission. It follows that this effect must be taken into account for robots working with large variation of the payload or inertial and gravity forces, and actuated with transmissions as speed reducer, screw-nut or worm gear. This paper proposes a new inverse dynamic identification model for n degrees of freedom (dof) serial robot, where the dry sliding friction force is a linear function of both the dynamic and the external forces, with a velocity-dependent coefficient. A new identification procedure groups all the joint data collected while the robot is tracking planned trajectories with different payloads to get a global least squares estimation of inertial and new friction parameters. An experimental validation is carried out with a joint of an industrial robot

To cite this version:P. Hamon, M. Gautier, Philippe Garrec. New dry friction model with load-and velocity-dependence and dynamic identification of multi-dof robots.Abstract-Usually, the joint transmission friction model for robots is composed of a viscous friction force and of a constant dry sliding friction force. However, according to the Coulomb law, the dry friction force depends linearly on the load driven by the transmission, which has to be taken into account for robots working with large variation of the payload or inertial and gravity forces. Moreover, for robots actuating at low velocity, the Stribeck effect must be taken into account. This paper proposes a new inverse dynamic identification model for n degrees of freedom (dof) serial robot, where the dry sliding friction force is a linear function of both the dynamic and the external forces, with a velocity-dependent coefficient. A new sequential identification procedure is carried out. At a first step, the friction model parameters are identified for each joint (1 dof), moving one joint at a time (this step has been validated in [23]). At a second step, these values are fixed in the n dof dynamic model for the identification of all robot inertial and gravity parameters. For the two steps, the identification concatenates all the joint data collected while the robot is tracking planned trajectories with different payloads to get a global least squares estimation of inertial and new friction parameters. An experimental validation is carried out with an industrial 3 dof robot.