How Accurate is the HoloLens 2? A Robotic Ground-Truth and Sensor Occlusion Evaluation

Spatial tracking in Mixed Reality (MR) relies on the device's sensor fusion and localization accuracy, directly impacting virtual object placement and user interaction. In this paper, we measure the minimum achievable projection error of Microsoft's HL2 with an industrial-grade robotic manipulator as ground truth. Specifically, we use a reproducible methodology to measure the device's highest achievable spatial accuracy and precision under controlled motion trajectories and tracking scenarios, including static positioning, sinusoidal 1D dynamic motions, and circular 2D planar movements at varying tangential velocities. Additionally, we investigated the contribution of different device sensors to overall positioning error by manually blocking specific sensor inputs. The analysis revealed an average localization error of approximately 5 mm with a standard deviation of 5.5 mm, representing the device's accuracy and precision, respectively. The goal is to provide realistic expectations of MR readiness levels with respect to its highest achievable accuracy and precision, shedding light on its capabilities and limitations for different applications.