[MechSkep]

So there was a DARPA initiative to do stuff like this back in the 90's/early 2000's:

http://archive.darpa.mil/darpatech99/Presentations/dsopdf/ds...

Turns out it's easy in controlled environments, and doesn't work at all outside. Once the animal sees something it wants to eat/mate with, it's just going to leave.

The claimed contribution of this work is direct ganglial stimulation as opposed to the antennal stimulation which was used previously. But I'm not seeing anything new in terms of capability/new science, and their data collection methods seem hinky:

The controllability of a roach, in terms of success rate, was subjectively rated based on user observation. If the user was able to successfully have the cockroach turn in a desired direction, then a run was deemed successful.