Abstract: A new humanoid predictive control scheme for biped robot is introduced.To generate adaptive walking pattern，the expected zero moment point（ZMP） can be separated as off-line reference ZMP and variable ZMP，and the preview controller and its inverse system can be used to control the trajectory of the center of mass（CoM）.But the single predictive control system have a big tracking error for variable ZMP such as rectangular perturbation，the humanoid predictive control system overcome these shortcomings because of its strong ability to suppress error.The simulation results verify the effectiveness of the proposed method.

Key words: biped robot, inverse system, humanoid predictive control, variable zero moment point

摘要： 针对双足机器人的稳定行走，提出了一种新的仿人预测控制在线步行模式生成方法。把期望零力矩点（ZMP）分解成离线规划好的参考ZMP和实时变化的可变ZMP之和，通过预测控制和其逆系统共同作用对质心运动进行控制，从而生成具有自适应性的步行模式。但单一的预测控制系统对诸如矩形齿状扰动的可变ZMP的跟踪存在较大的误差，结合仿人智能控制对误差的强抑制能力，设计了与预测控制相结合的仿人预测控制系统。仿真实验验证对矩形齿状扰动的可变ZMP，仿人预测系统也能实现较好的跟踪。

关键词: 双足机器人, 逆系统, 仿人预测控制, 可变零力距点