四足机器人凭借离散的立足点和腿身解耦的运动特性，可满足崎岖地形下良好的通过性，在军事、野外勘探、灾害搜救作业等应用方面具有广阔的发展前景。受启发于有脊椎四足动物丰富的运动模式和灵活的环境适应能力，基于仿生学的思想，本文研究了中枢模式发生器(CPG)的数学模型，针对现有CPG模型的不足，提出可实现四足机器人在线全方位运动规划的3D-CPG振荡器组模型，并基于该模型开展了自研四足机器人样机对角步态下的柔顺控制与反射运动规划方法的研究。论文主要工作与创新点：完成了可实现轮足切换的腿部结构参数设计，进而利用标准D-H方法建立样机运动学模型，完成了腿部逆运动学求解，并引入浮动基运动学模型实现了机身姿态控制与足端轨迹规划的解耦。针对现有CPG振荡器输出维度不足以控制三自由度腿的问题，提出主从双振荡器耦合的方法，完成基于足端工作空间CPG的极限环轨迹从二维相平面到三维相空间的扩展，进而实现了三维足端轨迹的在线规划和trot基准步态生成，并基于3D-CPG振荡器组模型设计机身自转步态，从而实现了样机的偏航控制。针对四足机器人在行走过程中的足地交互问题，本文引入了基于末端力反馈的导纳控制器对足端接触力进行柔顺以提高机身姿态的稳定性。此外针对腿部传动机构非线性强模型复杂度高的问题，建立了简化的腿部动力学模型，利用力雅可比矩阵和驱动电机电流环的力矩信息计算支撑腿的足端接触力，降低控制运算量。针对四足机器人在非结构化地面上的运动适应性需求，利用3D-CPG振荡器组的动态特性构建了基于振荡器参考相位与足端接触状态的反射控制策略，并通过有限状态机来管理反射回路的触发信号和相应的执行动作。此外利用卡尔曼滤波方法建立融合振荡器相位信息与足力数据的触地概率计算模型，实现了高准确性的足端接触状态估计。完成物理样机的装配和控制系统硬件的选型和集成，并在V-REP动力学仿真软件中建立了高仿真度的虚拟样机模型，在Simulink中搭建了虚拟样机的控制系统模型，通过联合仿真完成了运动控制方法的验证，并后续将Simulink中的控制系统模型迁移到ROS框架下，在物理样机上完成控制算法的验证。关键词：四足机器人；中枢模式发生器；轨迹规划；主动柔顺控制；状态估计

With the discrete foothold and the movement characteristics of leg decou-pling,the quadruped robot can meet the good passability under rough terrain, making it have broad development prospects in military, field exploration, disaster search and rescue operations and other applications. Inspired by the rich movement patterns and flexible environmental adaptability of vertebrate quadrupeds, this paper studies the mathematical model of the central pattern generator (CPG) based on the idea of bio-mimicry, and proposes a 3D-CPG oscillator group model that can realize the online all-round motion planning of quadruped robots for the shortcomings of the existing CPG model. Based on this model, the study of soft control and reflection motion planning method under the tret gait of the self-developed quadruped robot prototype was carried out. The main work and innovation points of the thesis:The design of the leg structure parameters that can realize the wheel-foot switch was completed, and then the prototype kinematic model was established by the standard D-H method, the inverse kinematics solution of the leg was completed, and the floating base kinematic model was introduced to realize the decoupling of fuse-lage attitude control and foot trajectory planning. Aiming at the problem that the output dimension of the existing CPG oscillator is insufficient to control the three-degree-of-freedom leg, this paper proposes a mas-ter-slave dual oscillator coupling method to complete the expansion of the limit ring trajectory based on the foot-end working space CPG from the two-dimensional phase plane to the three-dimensional phase space, and then realizes the online planning of the three-dimensional foot-end trajectory and the generation of the trot reference gait, which is based on 3D-CPG oscillator set model designs the fuselage rotation gait, which enables yaw control of the prototype. Aiming at the foot-to-ground interaction problem of the quadruped robot during walking, this paper introduces a guide controller based on the end force feedback to soften the contact force of the foot to improve the stability of the fuselage posture. In addition, aiming at the problem of high complexity of the nonlinear strong model of the leg transmission mechanism, a simplified leg dynamic model is established, and the foot contact force of the support leg is calculated by using the Jacobian matrix and the torque information of the current ring of the driving motor to reduce the amount of control operation. Aiming at the motion adaptability requirements of the quadruped robot on the unstructured ground, a reflection control strategy based on the contact state of the os-cillator reference phase and the foot end is constructed by using the dynamic charac-teristics of the 3D-CPG oscillator group, and the trigger signal and corresponding execution action of the reflection circuit are managed by means of the finite state machine. In addition, the Kalman filtering method is used to establish a ground-to-ground probability calculation model that fuses the phase information of the oscillator and the foot force data, which realizes the high-accuracy estimation of the contact state of the foot end. Completed the assembly of the physical prototype and the selection and integra-tion of the control system hardware, and established a high-degree virtual prototype model in the V-REP dynamic simulation software, built a control system model of the virtual prototype in Simulink, completed the verification of the motion control method through joint simulation, and subsequently migrated the control system mod-el in Simulink to ROS under the framework, complete the writing of the computer software on the physical prototype.