In medical field, the mobile rehabilitation exoskeleton robots arouse wide concern as they can help the paraplegic to regain the ability to walk again. Due to the body specificity of paraplegic patients, as the balance axillary means, crutches are necessary for the paraplegic patients who wear the exoskeleton robot to keep balance. We found that in the case of fixed step length, different support point positions of the crutches have great impact on the gait stable threshold. In the paper, through analyzing the polygon support plane formed when the users wear the exoskeleton robot to walk with crutches, ZMP theory is used to calculate corresponding position of the center of pressure. Thus, the stable threshold expression of quadruped gait is obtained. Based on these, the method of dynamically adjusting step length is proposed, and the fitting surfaces of the crutch support point and step length is obtained, thus the step length can be adjusted properly and in real time. Finally, the gait trajectory is planned using the kinematic model of the exoskeleton robot. A large number of comparison experiments were carried out on the exoskeleton robot developed independently by our team, the results prove that aiming at different crutch support positions, the proposed method can significantly increase the system gait stable threshold and reduce the influence of the crutch position randomness on system stability in each gait cycle.