Some projects we cover are simple, while some descend into the sort of obsessive, rabbit-hole-digging-into-wonderland madness that hackers everywhere will recognize. That’s precisely where [Excessive Overload] has gone with the AimBot V3, a target-tracking BB-gun that uses three cameras, two industrial servos, and an indeterminate amount of computing power to track objects and fire up to 40 BB gun pellets a second at them.
The whole project is overkill, made of CNC-machined metal, epoxy-cast gears, and a chain-driven pan-tilt system that looks like it would take off a finger or two before you even get to the shooty bit. That’s driven by input from the three cameras: a wide-angle one that finds the target and a stereo pair that zooms in on the target and determines the distance from the gun, using several hundred frames per second of video. This is then used to aim the BB gun stock, a Polarstar mechanism that fires up to 40 pellets a second. That’s fed by a customized feeder that uses spring wire.
The whole thing comes together to form a huge gun that will automatically track the target. It even uses motion tracking to discern between a static object like a person and a dart fired by a toy gun, picking the dart out of the air at least some of the time.
The downside is that it only works on targets with a retroreflective patch: it includes a 15 watt IR LED on the front of the gun. The camera detects the bright reflection and uses it to track the target, so all you have to do to avoid this particular Terminator is make sure you aren’t wearing anything too shiny.
H E A V Y
I’ll just wait until your arms are tired before shooting you :)
Yeah seems like it could be a lot simpler if it relied more on the human at least getting it in the ballpark and then slightly flexing the end of the barrel, something like that… Instead of this huge gimbal that rotates the whole gun 45 degrees in all directions
You could do that but you could also be unalive well before that happens :-D
I’ve recently got into airsoft and it’s an interesting hobby, with massive potential for technological improvements/open source/startup niche businesses. The use of MOSFETs and digital control units, optical sensors for triggers, and even “proper” RC style battery connectors are a relatively recent arrival in the hobby. Most “modern” offerings are still only using “Deans” connectors, where a molex style “Tamiya” was standard before, very rarely you see that someone has added XT connectors to their setup.
IMO, they’re the best type of “gun”, because despite playing with “toy guns” as an adult initially seeming a bit daft, they’re the only type of gun you can actually shoot at “proper” moving targets that can fight back, e.g. humans (with their implicit permission of course). Static target shooting just doesn’t seem very exciting, the couple of times I’ve done “target practice” at home, I’ve put 10-20 BBs through the bullseye, then end up setting arbitrary challenges like cutting a circle out of the cardboard box it’s taped to.
Now though, pedantry time: “BB guns” generally fire 4.5mm steel bearings at muzzle energies well over 5 joules, whereas this “aimbot gun” is based around an airsoft “engine”, which are used to fire 6mm plastic (mass loaded PLA) balls at between 1J and 3J muzzle energy, depending on local laws (often with a distinction between automatic and single shot), expected engagement distances, and arbitrary site rules.
The “overview” camera isn’t wide angle, it has the same lens and sensor as the others, and uses the whole sensor at a lower framerate. The stereo cameras don’t zoom, but only enable a small section of the sensor at a time and this “region of interest” is able to be actively moved by the connected computer. This allows for much higher framerates as shown in the video.
These types of industrial cameras (GenCam standard) provide raw uncompressed output and the main limitations partially the ADC sample rate of the hardware, but also the communication fabric used. An interesting note from the video is that the author found his gigabit network to not be sufficient!
The custom feeder is a really well engineered part of it, far more complex and well engineered that “some spring wire”. It features a large hopper with an actively driven rotary pusher mechanism with very specificly tuned geometry to prevent jamming. Part of the feed tube is built from a “spring coil tube” for overfeed prevention, feed damping/tension buffering, and uses a piece of non-spring wire to sense when this coil section stretches past a certain distance, cutting off the feed motor.
Polestar mechanisms can fire up to 50 airsoft BBs a second, not pellets (which are aerodynamically profiled metalic projectiles meant for vermin hunting and “plinking”.)
The motion detection doesn’t discern between the nerf projectile and the person shooting it: it is only tracking the IR tracking dot on the projectile and uses the speed of said projectile as a trigger condition. Also the author’s glasses were reflective enough to be considered a target, and eyeballs might too, avoiding reflective clothing isn’t enough to be safe from this thing blinding you!
Criticism over! Would be happy to see the article updated to reflect some of these points, but if not at least anyone interested enough can just read my comment.
Your comment is longer than the article. Brevity is the soul of wit
Great comment, appreciate the detail!
I have to look at airsoft again, seems a lot has changed in the last 15 years. The hop up mechanism used to add so much uncertainty into the system that you’d be lucky to group 5″ at 15 yards, making them useless for an IPSC type backyard training tool.
it actually does use a separate lens for the wide view. I never ended up getting the ROI function to work correctly so the view from the zoomed in cameras is from the entire sensor
Amazing build, but what really cought my eye is the water cooling PCB.
“You’ll shoot your eye out kid!”
-Santa Claus
yeah, the water cooled PCB is a really neat idea. I wonder why I haven’t seen that before? Does it have risks of developing leaks or damaging the substrate?
I naively assume a water and glycol mix shouldn’t cause issues.
I did found this reasearch paper describing the principle: https://www.researchgate.net/figure/High-power-liquid-cooled-3D-PCB-package-28-x-28-x-20-mm-3_fig4_224758911
Looks like the trick should be in the PCB routed edge to be metalized and goldplated, solder rings that keep PCBs together when going through reflow.
But this principle is not seen in the video as he uses only 2 PCBs. At least 3 are needed in the method above in my opinion. I would really like to know more and some real life cases where this could be used.
I see great potential in this method as I have a lot of cases where components cannot be air cooled as they are in sealed compartments reaching over 80°C (example of encolsed compact diesel generator units)
Once that grant money comes in this guy will be doing gov R&D.
I feel like some people came away from watching T2 in a completely different way… Obligatory https://imgflip.com/i/7nkcyn
Fun project! Curious why steppers instead of servos?
They are actually servos
Amazing work. The demo of shooting multiple targets in sequence was particularly cool. I wonder why he didn’t show the boresight camera view while trying to intercept a dart? That would probably show the issue. I’m guessing either the position control loop, or target point calculation was off.
I see you liked my tip then :-D