Exoskeletons are expected to be used as wearable haptic devices and power assist robot systems. We have been studying exoskeletons to assist the human motion in daily activity and rehabilitation. The EMG (electromyogram), which directly reflects the human motion intention, has been used to control the exoskeleton without any special control equipment. This paper presents a strategy that realizes the effective adaptation of the EMG-based exoskeleton controller (i.e., robot-human interface) to any persons.