Exoskeletons have been widely used as walking assistive devices for military and rehabilitation purposes. Many exoskeleton control systems detect users' intention by directly sensing their biological signals such as EMG, EEG, force, etc. Therefore, additional sensors are required, resulting in high cost and inconvenience of putting on and taking off the exoskeleton. In this paper, the user's intention is detected by estimating the torque exerted by the user based on exoskeleton dynamics as well as the current and angle of each joint motor; hence, requirements of sensors are diminished. Moreover, we conduct compliance control by modifying joint velocity commands according to the estimated torque and predefined mechanical admittance of each joint. Experiments are carried out to demonstrate that satisfactory estimation of the user's torque can be achieved by the proposed method.