The driving force of our current research is the development of medical training systems using augmented reality techniques. To provide multimodal feedback for the simulation, haptic interfaces are integrated into the framework. In this setting, high accuracy and stability are a prerequisite. Misalignment of overlaid virtual objects would greatly compromise manipulative fidelity and the sense of presence, and thus reduce the overall training effect. Therefore, our work targets the precise integration of haptic devices into the augmented environment and the stabilization of the tracking process. This includes a distributed system structure which is able to handle multiple users in a collaborative augmented world. In this paper we provide an