We have been developing wearable power assist suits for walking and lifting up a heavy object. This paper describes the method of estimation of walking intention without using bioelectrical signals such as surface electromyogram signals for walking assist of non-handicapped persons. We estimate walking intention using the on-off signal of foot switches and hip joint angles. As a result of the experiment of walking on a model course including flat ground and stairs, accurate walking intention estimation was confirmed at the beginning of the start of walking for 2 to 3 steps and at the beginning of the end of walking for 1 to 2 steps.