I guess I don't understand what precision is needed when dropping a grenade. I've never had a grenade thrown at me. I sort of suspect, but have no first hand knowledge, that a half a second is a big deal. yes, the drone will go to the location and drop the grenade. But there's no opportunity for dynamic reaction for changing conditions. I think, but do not know, that a few feet are a big deal with small explosives. Well, I kinda know from bottle rockets and such, but they aren't that scary.

You can cover a lot of small errors with a bigger explosion. but that means more payload, and more expensive.

denying that elegant control of small explosive in a precise location seems like a winner. yeah, you can go to position X but maybe I'm a little to the left of that. of course it's not free, but me picking up my shit and moving 10 feet is pretty cheap. cheaper than a drone for sure.

Maybe it doesn't matter, maybe tensor flow says that's a person near the target, adjust location to drop the explosive on the person.

from the point of a know nothing observer, it sure seems like there is some wiggle room. I'm not convinced an inexpensive drone has the capacity to solve these problems. (but you could send like 36 and drop grenades in a 6x6 grid, but that seems expensive).

Screwing around with RF seems like step 1.

you're underestimating the control software. with consumer grade cameras and simple image processing, you can write "go to x, find the nearest person and drop on them". also, just having the camera in the air is enough to make it much easier to launch a rocket at a target for juicier targets.

See [https://youtu.be/m89bNn6RFoQ]

Not just field trials, your average joe has been able to program waypoint missions for a long time. They generally require reliable GPS after take off, but I imagine rough intertial referential sensors would be enough for warfare. It's enough for commercial aviation.

Consumer grade IMUs drift into unacceptable garbage in about two minutes of flight without GPS corrections[0]. You would need visual odometry to work in GNSS-denied environments

[0] https://discuss.ardupilot.org/t/indoor-mission-plan-no-gps/7...

And visual assistance has been well supported in most commercial drones for years.

2 minutes is a pretty long time for a lot of use cases.

Yup - and consumer drones have had reasonable inertial sensors for quite awhile now. Not good enough for nap of the earth, but not far off.