The human-assisting robot can be helpful for improving the life quality of the disabled and elderly. As Electromyography (EMG) is a physiological signal generated during muscle contraction, it implicates, to certain extent, the human intention for movement, and is thus very suitable to serve as the control signal for the assisting robot. In this paper, we develop an upper-arm EMG-based robot control system, which provides a natural and intuitive manipulation. To satisfy the demand of real-time control, we propose a simple and effective method for the mapping between the upper-arm EMG signal and corresponding movement. However, due to the fuzziness inherent in the EMG signals, which are time-varying and highly nonlinear, the tuning for system parameters is not that straightforward. We therefore employ the concept of fuzzy system to find proper parameters that better suit for the individual users. To provide better adaptive capability, we propose using the adaptive neuro-fuzzy inference system (ANFIS) to realize the fuzzy system. We perform a series of experiments to demonstrate the effectiveness of the proposed adaptive upper-arm EMG-based robot control system.