Abstract: Lower limb exoskeletons enable paraplegics to regain some degree of locomotion ability, i.e., standing and walking. As a human machine system, the human machine interface (HMI) play an important role. The ideal lower limb exoskeleton for paralyzed people can move following the intention of pilot. To achieve it, many kinds of HMI systems are designed. However, for many exoskeletons, manual operation is still necessary for controlling the exoskeleton. In this paper, we designed and implemented an intention recognition method which is able to detect human motion intention. With the detected intention, exoskeleton can be controlled automatically as will of pilot. In this method, zero moment point (ZMP) is chosen as one of features of human intention and an online machine learning algorithm (online SVM) is used to learn intention online. Experiments in real systems show the effectiveness and advantages of our proposed method.