Abstract: Biologically-inspired control theory is introduced to improve human-robot interaction flexibility and multi-joint coordinated autonomous control in walking assist, and a new hybrid control method is explored. The hybrid control method consists of CPG (central pattern generator) based hip joint control, knee joint hierarchical impedance control and a hip-knee joint linkage control. Firstly, CPG self-excited oscillation and its external communication characteristics are utilized to obtain the desired active/passive trajectories of the hip joints. In addition, a symmetrical inhibitory network of CPGs is built to maintain anti-phase of the left and right hip joints, and thus to realize stable walking in the complicated HRI (human-robot interaction) environment. Secondly, a high and low impedance hierarchical control law is designed according to gait requirements to obtain the knee joint torque in stance and swing phases respectively. Finally, a linkage mechanism of hip-knee joints is implemented to connect the hip joints CPG control to the knee joints impedance control, and thus to establish natural movement of hip and knee joints in walking. The stability of the hybrid control method is analyzed using Lyapunov stability theory. Computer simulation analysis and walking experiments are carried out to demonstrate that the proposed control framework is effective in generating natural hip and knee joint movement in walking.