Abstract: Many XR devices such as AR-headsets, smartphones, as well as a growing number of VR-HMDs rely on visual-inertial tracking for motion estimation and environment mapping. Tracking accuracy is thus strongly affected by properties of the environment, such as number of visual features. In order to systematically assess such environmental influences on the tracking accuracy of XR devices, we have designed a setup where the XR device is attached to a moving robotic arm which itself is situated in a Virtual Reality CAVE. In our experiments, the high-precision robot arm repeatedly moved along the same trajectories in synthetically generated virtual scenes where the number of visual features was systematically increased. As expected, the measured tracking accuracy varied for different XR devices and virtual scenes. At one end of the spectrum, tracking completely failed in very feature-poor scenes. At the other end, sub-millimeter tracking accuracy was achieved for a synthetic scene with a very high number of features in case of a high-end XR device, outperforming the tracking accuracy achieved in a real-world environment with the same robotic setup. We conclude from these results that the proposed Robot-in-a-CAVE setup is in principle well-suited for assessing the tracking accuracy of XR devices.
0 Replies
Loading