> If that was viable, they would have started with Arduino before trying DMA attacks.
I'm not describing something theoretical. I didn't use it for cheating, but I have in fact used an RP2040 to emulate an HID mouse and tried to make it look "invisible" from the other end. I don't have code posted to GitHub, but other people do have similar examples. If you wanted to get obsessive about it, it probably wouldn't even be hard to emulate vendor-specific proprietary behaviors, like whatever custom HID reports are used for things like mouse profiles and firmware updates. It requires a bit of reverse engineering, but nothing that crazy.
You can also just do it with a normal Raspberry Pi too, if it's one that has USB OTG support. This is how a lot of those Raspberry Pi KVM solutions send inputs without needing much external hardware.
Let's say you are still not convinced, and assume that this method is doomed to be detected somehow. One last thing: you could always hack up a real USB gaming mouse and rig the sensor up to a microcontroller or FPGA. That is substantially harder and more expensive, but I'm just trying to illustrate that if people are really driven to cheat and cheating is really a big expensive industry, this can be done.
> One possibility: The Arduino added input latency cancelling out the cheating benefit.
I don't believe so, no. You don't have to stick to a specific hardware platform, but if there's one advantage that microcontrollers have over typical computers like Raspberry Pis it's latency. Even if their USB controllers are too limited or broken somehow, you can probably get away with bitbanging USB, especially on something with a nice and beefy microcontroller like the RP2350.
I believe the actual real reason why this wasn't a thing was because in the beginning for CV-based cheating, the CV part itself wasn't good enough. Now with computers having advanced a few more steps and with ML having advanced many many steps, being able to do very good classifiers in real time seems to be viable.
So really, the clock is ticking. There is no practical reason why the I/O part can't be done, so as long as the actual aimbot part works well enough. I'm willing to put my name on that.
I'm not describing something theoretical. I didn't use it for cheating, but I have in fact used an RP2040 to emulate an HID mouse and tried to make it look "invisible" from the other end. I don't have code posted to GitHub, but other people do have similar examples. If you wanted to get obsessive about it, it probably wouldn't even be hard to emulate vendor-specific proprietary behaviors, like whatever custom HID reports are used for things like mouse profiles and firmware updates. It requires a bit of reverse engineering, but nothing that crazy.
You can also just do it with a normal Raspberry Pi too, if it's one that has USB OTG support. This is how a lot of those Raspberry Pi KVM solutions send inputs without needing much external hardware.
Let's say you are still not convinced, and assume that this method is doomed to be detected somehow. One last thing: you could always hack up a real USB gaming mouse and rig the sensor up to a microcontroller or FPGA. That is substantially harder and more expensive, but I'm just trying to illustrate that if people are really driven to cheat and cheating is really a big expensive industry, this can be done.
> One possibility: The Arduino added input latency cancelling out the cheating benefit.
I don't believe so, no. You don't have to stick to a specific hardware platform, but if there's one advantage that microcontrollers have over typical computers like Raspberry Pis it's latency. Even if their USB controllers are too limited or broken somehow, you can probably get away with bitbanging USB, especially on something with a nice and beefy microcontroller like the RP2350.
I believe the actual real reason why this wasn't a thing was because in the beginning for CV-based cheating, the CV part itself wasn't good enough. Now with computers having advanced a few more steps and with ML having advanced many many steps, being able to do very good classifiers in real time seems to be viable.
So really, the clock is ticking. There is no practical reason why the I/O part can't be done, so as long as the actual aimbot part works well enough. I'm willing to put my name on that.