Abstract: Quadruped bionic robots are favored by development experts because of their broad application prospects, such as interstellar exploration, educational companionship, and social inspections. Quadruped robots were developed and inspired by mammals, which are known to exist in most areas on the earth's land surface. However, quadruped robots cannot achieve such an ideal state due to various reasons. At present, the adaptive problem of quadruped robots under a complex and changeable terrain has made significant progress, as reported in related literature. However, the case of robots that are as flexible as mammals in nature and meet the needs of multi-functional and multi-scenarios are still poorly understood. A quadruped robot is prone to posture instability when dynamically traveling in a ground environment with variable rigidity. This work proposes a real-time adjustment strategy of the active variable stiffness of the legs. This strategy estimates the landing in real time based on the motion state of the fuselage and legs after the robot touches the ground. The coupling stiffness of the legs and the ground and the difference between the coupling stiffness of the front and rear legs and the ground is compensated to the corresponding landing legs. This enables the robot to quickly adapt to the ground with different stiffness characteristics after landing, especially when the ground stiffness differs greatly. The Simulink-SimMechanics simulation platform is established with the diagonal legs on the same stiffness ground and on different ground environments with variable stiffness. The active leg stiffness adjustment strategy combined with conventional attitude feedback control is tested, and results are compared with those using only a conventional attitude feedback control. Results show that through the active variable stiffness of the legs, the quadruped robot realizes the compensation and correction of the pitch and roll angle of the fuselage during the transition between soft and hard ground. Moreover, the control effect is better than that of the conventional attitude feedback control alone.