A major challenge in the development of an assistive exoskeleton robot is to design appropriate control algorithms. These algorithms should be trajectory-independent and require a minimum number of sensors to work in any intended motion and to be easily implementable. As a simple assistive strategy with all promising features, Delayed Output Feedback Control (DOFC) is shown to be effective in assisting the wearers in different types of motion. In this method, the assistive torques are defined in proportion to delayed feedback from the angle difference between the two legs. The authors have recently suggested an intelligent version of DOFC, in which a Deep Q-Network (DQN) was used to adjust the feedback delay according to the speed of the motion. Simulation studies were used to investigate the idea. By conducting some real-world experiments, the present paper extends the results to practical conditions. The provided results clearly verify that if the time delay is adjusted according to the walking speed, the DOFC method can effectively help the users in their motions of any speed. The results also indicated that a fixed or an inappropriate value of the delay may result in resistance against the user motion.