Kinematic Parameter Calibration for Humanoid Robot Using Relative Pose Measurement in Walking Motion

Robots in use often need to be re-calibrated, due to the mechanical part deformation resulted from unexpected collisions. However, current calibration algorithms are not easy to be applied to humanoids and they require complex experiment setup. This is because humanoids have difficulties in gathering absolute pose measurement since they do not have a fixed frame on its body. On the other hand, the relative pose over time can be easily measured by a camera or a lidar that are usually attached to the humanoids' head. In this paper, we present a convenient and low-cost calibration method for humanoids by using relative pose measurements in walking motion. Our framework starts from the assumption that the sole of the support foot at every single footstep can be thought of as the temporary fixed frame while walking. Then, based on the decision of the supporting foot, we model the forward kinematics of an open chain from the foot to the head. In the description of a kinematic model, we use the local product of exponentials (POE) formula to take advantages of its simplicity and continuity in computation. Then, we solve the least square problem formulated to minimize the error between the actual measurement and nominal value computed from the corresponding joint angle data. Lastly, simulation results will be discussed to verify the feasibility of this framework.