Rubik’s Cube Detection using YOLOv8

This project trains a custom YOLOv8 model to detect Rubik’s Cubes in real-time using a camera feed.
It includes both model training and live object detection code, and can be connected to a robot for further manipulation tasks.

---

Project Structure

cube_yolo/ ├── data.yaml # Dataset configuration ├── best.pt # Trained YOLOv8 model weights ├── robot_cube_detection.ipynb # Training + real-time detection notebook ├── test_realsense.py # Camera test script (for Intel RealSense) ├── yolo_realsence_detect.py # Real-time detection with YOLO + RealSense ├── yolo_realsence_detect_final_position_desktop_plane.py # Detect whether the cube is on the table and return its 3D position.

---

Useage

1. Test camera connection

python test_realsense.py

2. Run YOLO detection

python yolo_realsence_detect.py

---

Results

Precision: 0.997

Recall: 0.965

mAP@0.5: 0.994

mAP@0.5:0.95: 0.809

The trained model can accurately detect both solved and unsolved Rubik’s cubes under varied lighting conditions.

---

Author

Jane (Juan Du) MADS graduate student Volunteer researcher at UVic ACIS Lab Focus: robot vision, human–robot interaction, and visual data analysis