Observations from testing

[AdamRoden]

For the curious people out there I have done some testing on this and my own aimbots using a GTX 1080ti GPU. This is about as good as you'll find without writing something yourself.

At a high-level:

• Once you get it setup properly it will work to target a humanoid's head.
• It can zero-in extremely fast on an slow moving or idle characters, but it will always lag behind running or jumping targets.
• The children I play with can aim and shoot faster and more accurately than this in 90% of scenarios.

In a deeper-dive:

• I find dxcam to be awesome!
• I modified to use yolov8 and feel the speed and accuracy are superior.
• Using a small PyTorch model I can get 45 fps
• Using a small TensorRT model I can get over 60 fps

[ghost]

@AdamRoden
Using the TensortRT version ( you need to convert the model yourself ) will increase the fps a lot for me from 50 fps to 100 fps, to reduce the lag behind running players increase the speed of the mouse movement since the mouse movement is already bad, it will lag more and be jittery so you also need to find a better way to move the mouse or find a better algorithm, currently, the rookit aimbot is just for educational purposes and just a proof of concept so they didn't do these things to be plug and play cheat
This can be better than memory cheats if you know how to improve it, also you need to train a better model that works on Fortnite players.

[AdamRoden]

@AdamRoden Using the TensortRT version ( you need to convert the model yourself ) will increase the fps a lot for me from 50 fps to 100 fps, to reduce the lag behind running players increase the speed of the mouse movement since the mouse movement is already bad, it will lag more and be jittery so you also need to find a better way to move the mouse or find a better algorithm, currently, the rookit aimbot is just for educational purposes and just a proof of concept so they didn't do these things to be plug and play cheat This can be better than memory cheats if you know how to improve it, also you need to train a better model that works on Fortnite players.

@hafyzwithawhy
Thanks for the reply!
In this ticket I wasn't looking for any improvements. I just wanted to tell people what they could expect from using this in the real world.
I don't understand why you think increasing the speed of mouse movement would help anything when it is not the problem. The real problem is that the AI predicts where an object is from a still frame image, but to match what a real person can do it would need to predict where a dynamic object is going to be and move the mouse to a future coordinate, (like you said: better algorithm.)

[Qfc9]

Great stuff yall. These are great and legit analysis.

The bot in it's current state is more geared for educational usage and is set up to be built upon. We have no plans to add or accept any PRs for adding a GUI like how Lunar has. This is because it takes away for the education value. Since we want to use this to get people interested in programming.

Anyone is free to modify the code base, add a GUI, and even sell it if they want. But in guidance with our license, credit must be given to us for the base code.

YOLOv8 was tested and can give better results, but requires beefier hardware. We have it currently in a lightweight state. Just using a medium size YOLOv5 model would give better performance in game, but requires better hardware.

[AdamRoden]

I'm back with more information. I've converted to yolov8 and feel it is noticeably faster than yolov5, especially for the medium and large models. After converting the model to a TensorRT version I gain a 35% performance boost over straight pytorch models. It is a balancing act with any other programs using your gpu. Overall, I find yolo detection to be hit and miss, sometimes helpful to zero in and other times an obstacle that gets you killed. You may benefit from training a custom model, but that's another road.

It's been a fun learning experience. Even with aim assistance I still suck and shooters.

[Qfc9]

Feel free to submit a Pull Request with the update. You will recieve the Champion role and our active volunteer badge.

https://www.credly.com/org/rootkit/badge/active-volunteer

Additionally, we added a new folder for custom, user submitted models.
@AdamRoden