Abstract: Minimally invasive coronary artery bypass grafting (MICS-CABG) is an important development in the surgical treatment of severe coronary artery disease, but MICS-CABG is difficult and risky, and trainees cannot directly train surgical skills in actual patients. The virtual reality-based surgical simulation training can effectively improve the training effect and avoid surgical risks. For the characteristics of MICS-CABG, we propose a real-time simulation method of minimally invasive coronary artery bypass grafting operation based on XPBD framework. A tetrahedral exoskeleton model is used to realize the physical deformation of the heart, and a cylindrical geometry is used to represent the physical model of the surgical instrument. Based on the GPU-accelerated XPBD simulation framework, we established the Cosserat constraints based on XPBD, using the coronary artery mid-axis to achieve the coronary artery deformation simulation, and used tetrahedral volume constraints to achieve the heart deformation simulation. Furthermore, based on the collision detection between the surgical instrument and the physical model of the heart, non-embedding constraints and attachment constraints were introduced to generate the feedback force for the interaction between the tool and the soft tissue. The experimental results show that the proposed method has high simulation efficiency and stability in real-time interaction between the surgical instrument model and soft tissue model. In the simulation scenario containing 100 000 physical units, the simulation rate can reach 60 frames/s. with the force feedback simulation frequency over 1 kHz, which can maintain a high-quality visual effect.