I'm curious about the high level, non-obvious gotchas in making them really useful to match current best offerings. Things like smoothing of gps (specific to the use case), better estimation techniques for current position/speed between gps updates, best practices around time estimation to next turn given current (estimated) position/speed, how best to determine when to reroute (how far is too far off route, how long to wait until rerouting), cheats people use to paper over bad data, etc.
Thanks!