TongjiSuperPower/sp_vision_25

What it solves

This project provides a complete auto-aiming and fire-control system for robotic combat (specifically for the RoboMaster competition). It addresses the difficulty of accurately tracking and hitting moving targets (armor plates) by predicting target trajectories and calculating ballistic solutions, reducing the time required to eliminate a target and increasing hit rates.

How it works

The system uses a modular pipeline consisting of perception, decision-making, and control:

• Perception: Uses a camera and neural network-based four-point detection models to identify armor plates and an Extended Kalman Filter (EKF) to estimate the target's motion state.
• Decision-Making: Implements a "Trajectory-based Auto-Aim Theory." Instead of simple rule-based logic, it uses a trajectory planner (utilizing TinyMPC for quadratic programming) to optimize the shooting trajectory based on the gimbal's maximum acceleration, ensuring the target is followed smoothly even during armor plate switches.
• Control: Sends target positions, velocities, and accelerations as feed-forward terms to the embedded controller, which uses computed torque control to drive the gimbal motors.

Who it’s for

• Robotics developers and students participating in RoboMaster or similar combat robotics competitions.
• Engineers interested in computer vision-based target tracking and real-time trajectory planning for actuators.

Highlights

• No ROS Dependency: Designed to be accessible for beginners without requiring ROS knowledge.
• Trajectory Planning: Replaces empirical decision logic with a planner that considers physical gimbal constraints to minimize overshoot and lag.
• High Performance: Achieves hit rates up to 39.6% and kill times as low as 8-10 seconds for a 300HP target at 2 meters.
• Modular Architecture: Includes a full workflow from development and calibration (camera/hand-eye) to deployment and testing.