The novel part here appears to be doing shape matching against road maps to improve accuracy. The accelerometers and gyroscope in consumer devices are so inaccurate that any track will be mostly useless after a few minutes, but if they can periodically tie that data back to known map points then accuracy improves a lot.

I'd like to see navigation apps make use of this approach when then lose GNSS signals, like in a tunnel.

Interesting, the augmentation reminds me of the Etak Navigator [1] which took a similar approach to remove accumulated errors.

[1] https://www.fastcompany.com/3047828/who-needs-gps-the-forgot...

"Snap to map" is often a setting.

It's a relatively well studied topic:

https://en.wikipedia.org/wiki/Map_matching

As the other reply says, this has been done for in-car systems at least since the late 90s. Heck, I had a company in the gnss space ask me this as an interview question probably 7 or 8 years ago.

In-car navigation systems do that, because they have access of stearing angles and speed.