Abstract: An assisted-walking stability identification and speed optimization method considering the human-robot interaction force is proposed, aiming to address the issue of neglecting of human-robot interaction requirements in the lower-limb exoskeleton walking stability criterion. Firstly, the inverted pendulum models for human walking and human-robot cooperative walking are established and analyzed, and the human-robot interaction force which is difficult to measure is indirectly expressed by integrating the centroid acceleration and plantar pressure information, and further the mathematical expression of CoM-CoP-CoA (center of mass-center of pressure-composition of acceleration) is given. Here the influence of human-robot interaction on the stability of exoskeleton walking is explored. Secondly, an exoskeleton walking stability criterion based on matrix inequality is proposed by using the above mathematical expression and the regional comparison method. Furthermore, a stability margin evaluation function is proposed by the improved stability margin description, in order to examine the relationship between the walking speed and the stability margin, and to optimize the walking speed of exoskeleton. Finally, comparative experiments of human-robot cooperative walking at three walking speeds are conducted, the experimental results are analyzed, and the optimal walking speed of human-robot cooperative walking is given. Experimental results demonstrate that the proposed stability criterion and evaluation function can effectively discriminate the stability of exoskeleton walking and optimize the walking speed.