Quality of results computed during parameter identification problems relies on the selection of system’s states while performing measurements. This choice usually does not take into account the uncertainty of states and of measures. For identifiability, classical methods focus only on the contribution of model errors on the uncertainty of parameters. We present an alternative approach that tackles this drawback: taking into account influence of all uncertainty sources in order to improve parameter identification robustness to uncertainties. A robotic application example that showcases the differences between approaches is developed as well.