This paper addresses the feasibility of an optimization system for a walking assist robot worn on the lower limb, which can reflect the user’s intention to shorten the travel time and walk with less effort. In this optimization system, an evaluation index is defined based on travel time per unit distance and energy consumption of joint motion, and the index is represented as a function of step length and pitch during walking. The optimization system was designed to be real-time by employing a method to mechanically calculate the energy consumption from the measured joint motions. The results of the evaluation indices for walking without and with assist force showed that the optimal gait changes when the robot adds a part of the joint torque required for walking to the user, and that the value of the evaluation index becomes smaller.