This video details the creation of an "anti-cammer eye replacement exoskeleton device" to improve performance in the aim trainer game, Aimlabs, under the humorous pretense of a severe eye infection. The project involves building a three-component system: an exoskeleton for controlled mouse movement (wrist), a finger-clicking exoskeleton, and a computer vision "eyeball" powered by an Nvidia Jetson AGX Orin for target detection. The creator navigates numerous hardware and software debugging challenges, particularly related to motor control, encoder wiring, and achieving ultra-low latency. Through an iterative design process, including optimizing the computer vision model with Tensor RT and upgrading to a high-speed CSI2 camera, the system's latency is reduced from an initial ~50 milliseconds to an impressive 17 milliseconds. The final exoskeleton successfully aids the creator in achieving a significantly higher score, landing him in second place on the Aimlabs leaderboard.
Building an Exoskeleton for Enhanced Gaming Performance
0:00 Project Premise: The creator humorously claims to have an eye infection and expensive eye drops, leading him to build an "anti-cammer eye replacement exoskeleton device" to achieve a high score in Aimlabs.
0:55 Goal and Baseline: The objective is to engineer a device that can improve his personal best Aimlabs score of 135,000.
1:12 Three Key Components: The exoskeleton needs three main parts:
Robotic assistance for mouse movement (wrist).
A finger-clicking exoskeleton for precise firing.
A computer vision program ("eyeball") to locate enemies and control the exoskeletons.
1:38 Wrist Movement Exoskeleton Design: The creator designs a forearm-mounted motor system using a gimbal motor, DIY skateboard speed controller, and Kevlar lines to pull the wrist left and right, targeting horizontal aiming.
2:28 Limp Motor Control: The motor needs to allow natural wrist movement while providing assistance, achieved by using a speed controller that can fully disengage the motor.
3:35 Finger Clicking Exoskeleton: A solenoid is modified and attached to a finger-tracking glove to provide rapid clicking assistance when the crosshair is on target.
4:11 Initial Assembly & Debugging: Basic functionality testing reveals significant issues with the motor encoder, including incorrect wiring and a potentially faulty unit.
5:10 Encoder Challenges: The encoder, which provides motor position feedback, was initially miswired and later discovered to be installed upside down according to documentation, leading to several days of troubleshooting.
6:44 Computer Vision Hardware: An Nvidia Jetson AGX Orin, designed for robotics, is chosen to handle the computer vision tasks and control the H-bridge for the solenoid.
7:19 Microcontroller for Jetson Control: It's discovered that the Jetson's GPIO pins cannot directly control the H-bridge for the solenoid, requiring an intermediate microcontroller.
7:45 Desk Sponsorship: The video is sponsored by Colomy Lumidesk, highlighted for its advanced cable management, adjustable height, robust modular design, and built-in features like a footrest.
9:26 Latency Target: Unlike neuromuscular aim assist (200ms reaction time), game aim assistance requires sub-25 milliseconds latency for effectiveness.
9:52 Anti-Cheat Strategy: A physical camera pointed at the screen (instead of screen capture) is used to avoid anti-cheat detection, adding 11ms per frame for a 90fps camera.
10:31 Computer Vision Model: A YOLO vision model is manually trained to detect blue orb targets in Aimlabs, then converted to a Tensor RT engine file for optimized performance on the Jetson.
11:25 First Exoskeleton Test: The initial setup achieves wrist movement and solenoid firing, providing noticeable assistance.
12:02 Mechanical Breakdowns: The Kevlar line frequently comes loose due to untied knots, leading to repeated repairs and adjustments, including looping the line between fingers with a rubber band.
13:22 Adapting to Assistance: The creator finds he needs to "learn" to use the exoskeleton by keeping his wrist softer and allowing the system to take over.
14:00 Initial Score Improvement: The first full test results in a 3% score increase, beating his high score, but with a total latency of ~50ms, falling short of the sub-25ms goal. The solenoid also fails to register clicks consistently.
15:13 Latency Optimization: To achieve lower latency and stronger actuation, the creator upgrades to a 120fps global shutter camera with a CSI2 connection (bypassing the CPU for direct feed to the tensor model) and increases the voltage for both motors.
16:35 Optimized Performance: The optimized system achieves a program-to-wrist latency of 9ms, totaling 17ms, significantly improving responsiveness.
17:51 Second Place on Leaderboard: With the improved exoskeleton, the creator achieves a score of 244,000, securing second place on the Aimlabs leaderboard.
18:18 Future Potential: The creator muses about combining neuromuscular and exoskeletal aim assist for even greater performance.