The joint of the upper limb rehabilitation robot, which is designed and built in our lab, is driven by pneumatic muscles (PMs) in an opposing pair configuration. Each PM drives the robotic joint through a steel wire with a flexible sleeve and a tension device, which causes delay and various frictions as disturbances to the robotic joint system. These factors make the rehabilitation robotic joint very complex to model and control. Especially in position control, the overshoot is difficult to deal with when the directions of the friction forces are changing. The main purpose of this paper is to enhance the position control performance of the robotic joint by neuron PI and feedforward. Neuron PI control has a powerful capability of learning, adaptation, and tackling nonlinearity, and feedforward control demonstrates good performance in dealing with frictions, which cause overshoot. The results of the experiments indicate that the proposed controller, which combines neuron PI and feedforward, can enhance the performance in position control of the robotic joint, especially on dealing with overshoot.