In the Off Pump Canary Artery Bypass Graphed (CABG) surgery, a robot is usually used to dynamically track the Point of Interest (POI) on the beating heart. To achieve better tracking performance and cancel the relative motion of the heart and the assisted robotics,an adaptive Auto-regessive (AR) linear model of heart motion was built to convert the complicated heart motion tracking problem to a dynamic model following problem. The Kalman filter was used to estimate the motion state of the assisted system and the linear quadratic optimal tracking theory was used to implement the model-following method. The results show that the model-following method which provides estimated future reference enables the robot to improve the relative motion cancellation ability by 30% and decrease the tracking error by 0.25 mm. The comparison tracking result on 3D test bed Phantom robotics is reported, which proves that the model-following method enhances the ability of dynamic relative motion cancellation during the CABG sugary.