RoboMaster Auto-Aim

Overview

In RoboMaster, robots score by hitting the armor plates on enemy robots. Auto-aim lets the robot find those plates in a camera image, predict where they will be, and point the gimbal there on its own. This matters most against robots that spin in place to dodge.

What I did

Our team had no vision code, so I deployed Tongji University's open-source auto-aim system on our robot and integrated it with my embedded firmware. The pipeline has three stages:

1. DetectionYOLO finds armor plates in the camera image
2. Pose estimationa calibrated camera turns them into 3D positions
3. Prediction & controlMPC plans where the gimbal should aim
4. Firmwarethe STM32 drives the gimbal and fires

I calibrated the camera and tuned the system on our hardware. At this point I can deploy and tune it, but I don't yet fully understand the theory behind every stage. Learning that is my next step.

Results

• ~70% hit rate on a robot spinning at 1 rev/s, 3 m away.
• ~25% hit rate on randomly moving robots in real matches.