Abstract
We investigate the effect that visual occlusion plays on users' perception in a prototypical augmented reality post-stroke therapy system. We implemented a reach-based therapeutic exercise using the Microsoft Kinect to enable depth sensing and correct visual occlusion of the upper-limb. Thirty participants evaluated the exercise with three different visual occlusion modes: the correct visual occlusion, a virtual-always-occludes---to date the most commonly used mode in augmented reality---and a mode with semi-transparent virtual objects. The analysis of their reported experience showed that correct occlusion was the most preferred mode for performing the reaching exercise, providing a more tangible and realistic interactive experience.