使用零力矩点反馈的双足机器人惯性参数辨识

Inertia parameter identification of biped robot using ZMP feedback

为解决基于关节力矩的双足机器人参数辨识方法辨识精度不高,基于完整的足底力信息和运动捕捉数据的辨识方法对实验条件要求较高的问题,提出基于ZMP(zero moment point)数据的双足机器人惯性参数辨识方法。将理论ZMP与实际ZMP的位置偏差作为目标函数,考虑参数范围和机器人总质量两类约束条件,建立只使用双足机器人自身传感器采样数据的惯性参数辨识优化模型。针对所建模型无法拆分成线性形式的问题,推导目标函数关于参数矢量的梯度矢量和海塞矩阵,并给出了基于最速下降法和牛顿法的优化求解算法。使用GoRoBoT-II机器人的双足部分,进行腿部杆件的惯性参数辨识实验,将所提出方法得到的辨识结果与传统基于关节力矩的辨识结果进行对比,发现基于ZMP的辨识方法的结果更接近于三维几何建模得到的参数标称值,且理论ZMP与实际ZMP的偏差均值为4.6 mm,小于传统基于力矩辨识方法的12.4 mm,说明所提出的基于ZMP的惯性参数辨识方法能够得到比传统方法更好的结果。

The existing parameter identification method of biped robot that uses the joint torque has low identification precision.The identification method based on full contact force and motion capture data requires additional equipment,which limits the application in a large range.Regarding this problem,a method for inertial parameter identification of biped robot based on ZMP data is proposed.The objective function is defined as the position deviation of the theoretical ZMP and the actual ZMP.The range of the parameters and the total weight of robot are considered as two constraint conditions.Then the optimization model of inertial parameter identification of biped robot is established,which only needs sample data acquired from the robot itself.Because the built model cannot be split into linear form,the gradient vector and Hessian matrix of the objective function are derived with respect to the parameter vector.Also,the optimization algorithm is given based on steepest descent method and Newton method.Using the biped part of the GoRoBoT-II robot,the inertial parameter identification experiment of the leg links is carried out.The proposed method is compared with the traditional identification method based on joint torque.It is found that the result of the proposed ZMP-based identification method is closer to the nominal value of the parameters obtained by 3D geometric modeling.Also,the deviation between theoretical ZMP and actual ZMP is 4.6 mm,which is smaller than the deviation(12.4 mm)of traditional method,indicating that the proposed ZMP-based parameter identification method can obtain better results than traditional methods.