The authors consider the problem of improving the parameter identifiability properties of a robot model and derive a sequential estimation algorithm which substantially simplifies the estimating procedure. The estimation of the invariants (masses, inertias, etc.), which usually requires at most 11n parameters for a robot manipulator with n degrees of freedom, can be performed link by link in a sequential manner by n algorithms of size n/sub i/, where Sigma n/sub i/ is smaller than 11n. Optimization of the robot trajectories seeking to improve parameter identifiability can be simplified. This method enhances the numerical algorithm conditioning and facilitates the selection of a high excited identification sequence, improving the parameter identifiability. The convergence of the estimates to their true values can be obtained provided that the information vector associated with each link is persistently exciting.>

The problem of modeling and compensation of friction velocities close to zero is analyzed. A model, linear in parameters, which captures the downward bends at low velocity is used to adaptively compensate for friction. The need for this type of model is mainly motivated by instability phenomena that can be caused by overcompensation when simple models (such as Coulomb friction models) are used as a basis for the friction compensation. This model, in combination with an adaptive computed torque method, is tested experimentally in the last link of a robot manipulator. The prediction error e/sub k/, which directly gives the accuracy of the predicted friction, remains within the preestablished threshold of +or-0.8 Nm. The experiments were realized using an IBM PC-compatible computer, leading to a sampling time of 13 ms.>