Abstract: Using monocular optical see-through HMD (helmet-mounted display),the augmented reality system can make an optical illusion for the user,where the virtual object is in front of real objects. Thus the user can not make sure where they should grasp the virtual object. Furthermore,the data got by the system could be affected by many factors such as human operator,environment,and even the hardware itself. Through the human visual awareness,the collision detection based on visual awareness is realized using techniques from robotics and geometrical optics theory. Three knowledge rules about grasping are obtained by analyzing grasping action,and then the rules are fused with the fuzzy theory in evidence deduction. The results show that by using this method,the user can grasp virtual object easily and the process of the interaction is natural.