Architect would be much more knowledgeable than me on this, for sure, but for pathfinding itself, I think there’s plenty of processing power, especially in a board like the Raptor.
2-wheel balancing needs to make very fine adjustments many times a second (both reading sensors, and adjusting servos) so if someone has gotten that to work, then pathfinding (which, it seems to me, needs an order of magnitude less processing power) should be a piece of cake.
There -was- a “line following” type of gadgeteer module available awhile ago, that was used with even the slower boards:
It was fast enough to prevent bots from going off tables, supposedly, so I am assuming some people got that to work. (it is discontinued now though)
If you are looking for pathfinding along the lines of mapping an area in decent resolution with distance sensors, and object avoidance navigation that remembers where obstacles are, you’ll probably need a board with more ram, like the raptor, to store where you sense objects, and encoders on your wheels to determine more accurately where you are, and maybe a compass module for confirming direction, but it should all be very doable in C# with the gadgeteer boards, and I am sure many people have done it.
I’m actually working on something similar. It hasnt been speed or ram or processing power that have been my issues. Its been getting parts to get my Vex Mecanum wheels to mount nicely on some decent motors, lol.