In this paper an approach for the identification of the dynamic parameters, i.e. base parameters, of rigid robots is presented. By using the polynomial approximation operator, an equation is obtained for the identification of the parameters which solely depends on measurable signals and thereby contains no equation error. The resulting expressions can be evaluated online or offline by filtering the measurable signals with FIR filters. In order to identify the parameters on the basis of measurements, an algorithm is presented to calculate the parameters numerically stable, even if the data is obtained sequentially, without a singular value decomposition. The parameters can be determined meaningfully by considering box constraints in order to ensure physical feasibility. The presented methods are finally used to identify the dynamic parameters of a delta robot and compared to the standard approach.