Inertial parameter identification of bipedal systems : a simulation-based comparison

This thesis presents the implementation of inertial parameter identification using least-squares estimation, comparing four methods to solve the resulting equations. In the first chapters the equation of motion is written linearly in the inertial parameters, which leads to a least squares problem that cannot be solved directly. In the included paper, the least squares problem is scaled using measurement uncertainty in the joints and a prior estimation, and solved using a singular value decomposition. The effects of scaling are examined by comparing the results of the unscaled, and multiple scaled problems. Using simulation data of the Symbitron lower-limb exoskeleton, it is shown that the row- and column-scaled version performs best, despite being sensitive to a prior estimation.