Trouble is GPS isn't always reliable in heavily wooded areas, so there might need to be improvements in GPS before that is really viable.
GNSS accuracy is poor in forests, canyons, and dense urban environments due to line-of-sight obstructions and multipath reflections. The latest Navstar Block III satellites should improve that a little, but any real solution would require deploying additional satellites in geosynchronous orbit like the Japanese QZSS.
I feel like this should be solvable on the software side, especially for medium or long distance runs. If I look at the traces, I can see that the recorded zig-zag is obviously wrong, and a bit of smoothing should be able to fix it.
Are there any running apps that are accurate?
Accelerometer, magnetometer, and gyroscope sensor data can help a little to sanity check GNSS inputs and fill in brief gaps. But those tiny sensors have terrible drift which makes them nearly useless for sustained position tracking.
This frustrated me enough that I eventually got a Stryd footpod. The pace/distance tracking is extremely accurate and I use it to override what the watch records. So the GPS track still bounces all over the place but the recorded pace/distance data is correct.
There's nothing about geosync orbit that would help with the tree/canyon effect. Satellites directly overhead contribute the least to horizontal positioning - essentially only helping to establish the "true" clock value but not 2D position.
Multi-constellation tracking can be helpful, simply because there are more satellites to use.
https://www.ion.org/publications/abstract.cfm?articleID=9679
Unfortunately the state of localization isn't ready for passive (aka non-radiative) reliable sub meter localization over anything more than a few meters squared in a well lit indoor space.
Going to be a while likely before we get to "wear everyday" AR glasses with sub meter accurate camera position/localization.