This thesis addresses the issue of model identification of an advanced modular and reconfigurable robot system, the IRIS (Institute for Robotic and Intelligent Systems) Facility. Theoretical insights into the methodology of parameters estimation are developed, The theory is experimentally applied to the IRIS Facility.

In the theoretical part, the classical Least Squares (LS) algorithm is presented. It is found that the LS estimates contain bias when measurement noise is present. To compensate the bias in LS estimates, two modified LS algorithms are presented:​ (i) the Bias-Compensating LS algorithm (BCLS);​ (ii) the Instrumental Variables (IV) method. The applicability of these two algorithms to robot parameter estimation is discussed and evaluated by simulation examples.

Using the proposed algorithms, the dynamic parameters of the IRIS Facility are determined experimentally. The experimental results have confirmed the theoretical analysis that the modified LS estimates are more accurate than the original LS estimates. The experimental work also provides a set of reliable parameters for the facility users and control system designers.

Finally, recommendations are given for improvement of the estimation process, and for related issues that need to be addressed.