The CPU is slightly better in the Pi 4. The Pi 4 has built-in WiFi, which I have to add with a USB dongle on the Nano. There is a larger community of hobbyist developers, most of whom are probably less experienced than you are.
There ends the advantages for the Pi 4. The Nano has better camera support circuitry and hardware encoders. The Nano has a much better GPU. The Nano has a very easy time installing reasonably-modern "deep" learning models, and also has sample code for things like visual odometry. And, most importantly, most of the Nano software stack scales up to the Xavier, as well as Mini-ITX systems with full-fledged GPUs. (There are some minor differences between "Jetson" vs "Desktop" platforms, but they're not that big.)
If I were to start another rover today, I'd probably use the Xavier NX. It's the same size as the Nano, but it has a much more modern CPU, with more cores, and more modern GPU, with more cores, as well. (In fact, my current rover has a Xavier on it, but the "full fat" version, that's a bit bigger, and has EVEN MOAR CPU and GPU.) It also supports two CSI cameras. Compared to the cost of all the bits that need to actually go on a rover these days, the additional cost for the NX part over the Nano part is totally worth it IMO.
For the low-level interfacing, I'd almost surely use one of the Teensys. The 4.1 is out, and is awesome. Or, if I still need 5V, and don't want to bother with level translators, the 3.5 is still a solid choice.
Sincerely,
Jon Watte
--
"I find that the harder I work, the more luck I seem to have." -- Thomas Jefferson