Skip to Main Content
This paper presents methods and experimental results regarding the identification of kinematic and dynamic parameters of force-controlled biped humanoid robots. We first describe a kinematic calibration method to estimate joint angle sensor offsets. The method is practical in the sense that it only uses joint angle and link orientation sensors, which most humanoid robots are equipped with. The basic idea is to solve an optimization problem that represents a kinematic constraint that can be easily enforced, such as placing both feet flat on floor. We then present two methods to identify physically consistent mass and local center of mass parameters even when obtaining enough excitation is difficult, as is always the case in humanoid robots. We demonstrate by experiment that these methods give good identification results even when the regressor has a large condition number. Moreover, we show that gradient-based optimization performs better than the least- square method in many cases.