Robots are widely used in industry in the repeating tasks to free human from the tedious labor. There are some tasks that are either complex itself or simple but interacting with complex environment. In these cases, human are not replaceable and human/robot interactions are inevitable. Design and developing human performance enhancing robotic devices is applicable to not only industrial assembling robots and moving/carrying assist devices, but also can be extended to areas like medical surgical robots, exo-skeleton etc. The benefit is to utilize the human perceiving and analyzing capability to the difficult tasks and environments, thereby the main purpose and challenge of such system is to implement the intuitivity of the operator in the system control. The admittance control approach is often adopted in human/robot interaction control. It is noticed that the control designed with a fixed pair of virtual mass and damping cannot reach desirable performance of handling. From literature review, the variation in admittance adapting to operator’s velocity and force can improve the performance. In this paper, an iterative learning method is proposed for variable admittance control design. A cost function is used to let controller learns while operator follow a predefined trajectory. It is applied to a four-degree-of-freedom assist device. The simulation and test results are used to calibrate and validate the control algorithm.