is there any source (paper/report/code/etc) explaining the exact Visual-Inertial-Odometry (VIO) implementation that is being used in ARCore for tracking the user\'s motion?
