HN user
This comment is spot on. Actually if you have a device with ARKit on it, you can see for a fact that they do have relocalization: cover the lens for a few seconds until tracking is lost, then return to the point of origin, you will see that the intial cube (placed at the origin) snaps back to the correct place. What they probably don't do is global mapping of the entire session.
It seems that ARkit is actually using SLAM as the core technology (in parts, the same technology is being used in autonomous cars and navigating drones). Getting it to work on mobile phones is quite challenging and had been the subject of much research in recent years, some good starting points (with code): ORB-SLAM - https://github.com/raulmur/ORB_SLAM2, https://www.youtube.com/watch?v=T-9PYCKhDLM DSO - https://github.com/JakobEngel/dso, https://www.youtube.com/watch?v=C6-xwSOOdqQ VINS - https://github.com/HKUST-Aerial-Robotics/VINS-Mobile, https://www.youtube.com/watch?v=0mTXnIfFisI&feature=youtu.be
(the last one being released just a few days ago...)
Of course there are a lot more resources out there and a lot of the cutting edge research is involved with deep learning techniques (starting with e.g. https://github.com/alexgkendall/caffe-posenet).
You may also want to watch out for this conference http://cvpr2017.thecvf.com/ scheduled for the end of this month and is usually a good source of cutting edge research...