Cobot Chess

Chess application for the CORA cobot arm.
The system uses monocular computer vision to estimate rigid-body transforms in ${\mathrm{SE}(3)}$ between the robot, camera, fiducial markers, and a physical chessboard. A CNN-based chess engine selects moves, which are executed via pick-and-place motions using the robot’s Ethernet SDK.

Each agent is trained and biased toward the move distribution of a single human player.

---

Project Structure

assets/
chess/
tfs/
vision/
main.py
tests/
pyproject.toml
requirements.txt

---

Installation

Clone the Repository

git clone https://github.com/C-O-R-A/cora-chess.git

Install Dependencies

pip install -r requirements.txt

---

Dependencies

• numpy
• chess
• codi (networking SDK)
• torch
• pandas
• seaborn

All dependencies are installed automatically when using requirements.txt.

---

Vision and Perception

This section describes the perception pipeline for estimating camera pose, board pose, and opponent moves using rigid-body transforms in ${\mathrm{SE}(3) }$.

---

Camera Calibration

from vision import camera_calibration

# Use camera 0 with a checkerboard pattern
camera_calibration(0, use_checkerboard=True)

---

Marker Placement

---

Basic Usage

python3 main.py

The system will:

1. Detect fiducial markers and the chessboard
2. Estimate all required ${\mathrm{SE}(3) }$ transforms
3. Observe the opponent’s move
4. Select a move using the CNN-based engine
5. Execute the move using the robot arm

---

Playing a Match

Ensure that:

• The camera is calibrated
• All fiducial markers are visible
• The chessboard lies within the robot’s reachable workspace

Once running, the robot alternates turns with the human opponent and physically moves the chess pieces on the board.

---