Deep-dive on ARCore | Fundamentals of designing AR apps with machine learning models
summary
ARCore combines camera feature tracking, a 100+ Hz IMU and visual-inertial sensor fusion to keep poses stable; anchors (hit.createAnchor and .pose) persist through temporary tracking loss and update as scene understanding improves. New ML-enhanced depth gives usable raw depth with far less motion, and a machine-learned IMU integration network trained on hours of real motion data fills multi-frame visual gaps so content no longer drifts. Live View in Google Maps already adopts the approach.
AI-written summary · about
