Have you ever seen an FPS plug-in that can automatically aim and "accurately hit" without being sealed Now, there is really a CMU graduate student who turns his mouse into a "divine sniper" in FPS game After installing four wheels on it, the mouse can automatically aim and realize one shot sniping according to the feedback of CV algorithm.
Xiao Xiao comes from aofei temple
Before that, although there were some physical plug-ins, they were not automated enough to let the mouse move and complete shooting.
If you choose third-party software modification and other plug-ins, it is very easy to be directly sealed.
After thinking about it, the little brother developed an FPS physical plug-in that can aim at his own target. At present, his performance in the aim Lab of the training ground has exceeded that of some FPS professional players——
The average score of human FPS experts is about 80000 ~ 90000, while it gets 110000 + points.
Some netizens said "it's not easy": for people who have written code, it doesn't sound so easy.
So how on earth did he do it?
Turn the mouse into a "divine sniper"
In order to let the mouse "learn" to aim, the little brother first designed a chassis for the mouse.
In his vision, the chassis should be able to move flexibly with the mouse. The principle is like this:
Finally, the idea of four-way motor control is used, which corresponds to the rear chassis.
In short, Omni wheels in all directions can make the robot move flexibly in any direction, including direct straight line, and these wheels are controlled by different motors.
After the basic mobile operation is realized, the computer vision algorithm is used to make it learn to "find its own prey".
Brother wrote a python target detection algorithm based on OpenCV and trained it to quickly aim at the corresponding target according to the position of the target:
The robot learned to "aim at the nearest point" in the specified distance, and learned to operate the PID algorithm from the specified distance:
When aiming at an object, the robot will automatically trigger the action of mouse "click", so as to complete automatic shooting.
More than some professional players in the test
In order to strengthen the robot's ability, my brother trained it in a first person shooting game simulator called aim lab.
At present, many professional FPS players will train in aim lab, which also includes various types of shooting tasks, such as humanoid moving target, flying target, etc.
The "mouse robot" trained by my brother is mainly aimed at color ball shooting, in which the visual algorithm automatically finds the colored ball, and then the robot follows up to aim and shoot.
After testing the algorithm for two months, my brother finally trained a satisfactory robot. At present, he has got the highest score of 118494 in aim lab, surpassing many professional FPS players.
△ open scores of a professional player
However, at present, it has not broken the highest record of aim Lab (146902 points).
Another reason is that in the process of challenging the highest score, its chassis burned a motor due to too fast speed, so it can't challenge the highest score again (manual dog head).
The little brother who developed this physical plug-in is Kamal Carter. He graduated from CMU with a bachelor's degree. At present, he is a master's student in CMU. At the same time, he also works in Hebi robotics founded by Professor Howie choset.
According to my brother, he began to play with robots in high school. At present, his research interests are mechanical design and CAD.
Some netizens joked that this robot deserves a better mouse:
But some netizens believe that my brother has such ability and should do something useful instead of doing these physical plug-ins.
So, what do you think?
Opencv visual algorithm tutorial: https://docs.opencv.org/4.x/df/d9d/tutorial _ py_ colorspaces. html