With this sort of tech it only achieves good usability if it is accurate a high percentage of the time. Anything less than 90% accuracy on a whole trip and I likely won't trust it. So that means getting to >99% accurate on each underground train stop. I'm still not sure if this tech reaches that sort of threshold or if this is one of those things that looks like a good idea on paper but falls apart in real world use cases.
I'd assume they would have some opt in, or allowance for either federated learning or sharing of anonymized sensor data.
Google is a huge advantage here, they have both the spatio-temporal intent of the user as well as the physical flow feed. There is so much position data flowing off android phones, that they are able to see the whole topology.
