In responding to the coming of an aging society, the exoskeleton robot is of focus, which provides assistance for people with locomotive problems. As the exoskeleton robot is worn by the user, the interaction in between induces several severe challenges, including adequate controller design. To deal with the nonlinearity involved during motion governing, in this paper, we take advantage of the capability of the adaptive network-based fuzzy inference system (ANFIS), and propose an ANFIS-PID controller for the exoskeleton type of robot. Especially, we apply it to a 2-DOF upper-body exoskeleton robot, HAMEXO-I, developed in our laboratory. With the proposed controller, we intend to let HAMEXO-I adaptable for a wide range of users, and able to serve for simple daily activities. Experiments are performed to demonstrate its effectiveness.