This is perhaps the first vehicle test of the SmartRTL mode on ArduPilot Rover 3.2 (currently in beta testing). Peter Barker and I ported the SmartRTL feature over from Copter (video) last week. Just like copter, SmartRTL in Rover stores up to 500 points in memory constantly simplifying and pruning those points as required to save space.
When the vehicle is switched into the new mode, it follows the path back to the location where it was first armed. You might notice that the vehicle is a bit jerky on it's return journey. This is a natural outcome of the point simplification which turns multiple points in a line into a single straight line. So curves tend to become straight lines. There are parameters to control the accuracy of the simplification and pruning which can make it smoother but at the expense of storing more points. I suspect we can also smooth out the path without requiring more points with a bit more work.
For developers out there interested in writing your own drive modes, we've written up a wiki page on how to do it here.. In fact, Peter Barker and I wrote this mode last week during a live streamed development session. I won't post the link to the video here because it took us 5 hours (:-)) but we plan to do another simpler mode in a week or two and will certainly post a link then.