Large Unmanned Store System

Undergraduate project at Hanyang Univ.

This project proposed a cost-efficient unmanned store system by centralizing heavy perception on a main vision server and keeping each cart lightweight. For full project scope and team-level implementation details, see the project repository.

My Contribution

1. Infrastructure Perception

   • Calibrated multiple ceiling cameras and transformed each camera coordinate to a single global reference frame.
   • Built multi-camera person detection and converted detections into global target positions for cart following.

2. Cart Autonomy

   • Implemented a local safety layer with front/rear LiDARs to reduce blind spots and react to nearby obstacles in real time.
   • Estimated cart pose from encoder odometry and corrected accumulated drift using ArUco marker observations.

Outcome

• Demonstrated person-following and obstacle-aware cart operation in an indoor retail-like setup.
• Reduced cart-side compute burden and achieved a prototype cart hardware cost below KRW 800,000.