If you add enough "if's", then surely everything is possible :-).
I don't think we're anywhere near having drones that happily fly above a war zone, detect an interesting hangar, find a way to get inside and select a target inside.
Currently they mostly fly FPV drones manually. The next basic step is to have "terminal engagement", where at some point they can select a target and the drone will fly autonomously to it using CV. But in order to do that, the drone will need processing power, and therefore it won't cost 500$ anymore.
Would you rather go for a drone that costs 5k and can use CV for terminal engagement, or 10 drones that cost 500 and simply stay on their latest vector if no command arrives?
The drone is 500$. If the board is 300$, it makes it a lot more expensive. What's the value of that board? How much does it help? Remember that you need to fly it with FPV to the place where the operator can select the target. So you have a radio link at some point, that can be jammed.
I see on the rpi website [1] that an RPi 5 costs 120$. That's 25% of the price of the FPV drone (500$).
Then you have to integrate that RPi to the drone, both mechanically and software-wise. From YOLO, you need to actually get into commands for the autopilot. There is work involved.
And another question is: can you order millions of RPis like this? If not, what is the cost of alternatives?
Not saying it's impossible, just saying that people tend to underestimate those questions. I know it from experience, I have seen projects fail because of that.
I was trying to state you fly it manually to the hanger. Once it locks onto the hanger (that is not your own hanger) it can fly - even if cv messes up we are at a target so anything destroyd is okay even if not what you want.
Oh, I had misunderstood it. But still, it's a lot harder to do that succesfully than to lock on a target and fly straight into it. Which already requires the compute power to do it, which makes it a lot more expensive than it is without.
Sure, but we can trade off computer power. Pi level computers can fly a straight line inside and explode in the middle (you are likely running several drones so each flys to a pattern hoping to get something useful inside). While flying to the middle they can do some image recognition, and if something looks like a high priority target they can target that, if not exploding in the middle will do something. The more powerful computer you put on (were powerful is often more expensive though not always) the better you can find targets
The important point in this is the drone will explode before it could reach anything not a target. It can sometimes find a better target than a pattern. In the ideal case you might fly it all the way, but if you lose radio over enemy territory anything the drone can find needs to die anyway so it may as well attempt to find something and kill it.
some mid-range iphones or android would even do the trick. especially iphones have tone of processing power nad strong NPU/GPU and lots of cameras, lidar, depth sensor, and plenty of other sensors. second hand phone 13 mini would do the trick and you can get it for less than $500
I don't think we're anywhere near having drones that happily fly above a war zone, detect an interesting hangar, find a way to get inside and select a target inside.
Currently they mostly fly FPV drones manually. The next basic step is to have "terminal engagement", where at some point they can select a target and the drone will fly autonomously to it using CV. But in order to do that, the drone will need processing power, and therefore it won't cost 500$ anymore.
Would you rather go for a drone that costs 5k and can use CV for terminal engagement, or 10 drones that cost 500 and simply stay on their latest vector if no command arrives?