Mini Tank

A real(not sim) self-driving miniature tank. Input from a camera is uesd to make course corrections and stay on path. The tank relies on a Tensorflow model for decision making. The model is trained using data from a human controlled run (supervised learning).

Parts

• Raspberry Pi 3
• Tank Mobile Platform
• USB Webcam
• Xbee for communications
• USB controllers.
• LiPo Battery.
• L298N Motor Drive Controller Board Module Dual H Bridge.
• 5V power bank to power RPi.

Software Setup

1. Install Raspbian on RPi.
2. Install python3 and dependencies: sudo pip3 install -r requirements.txt
3. Install the systemctl service.

sudo cp agent.service /lib/systemd/system/
sudo systemctl daemon-reload
sudo systemctl enable agent.service
sudo systemctl start agent.service

Data Collection

Start controller on PC.

cd desktop_side; python3 controller.py

The car should now be driveable, it will be recording data but it wont commit to disk until you send save. The collected data will appear on the Pi on /home/pi/.

Training

Move collected data to PC. This will be significantly faster than training on the Pi.

cd desktop_side; python3 train.py

Run in self-driving mode.

Move trained model (model.h5) back into RPi. Start the model server on the Pi.

python3 model_serve.py

With the controller send the self-diriving command.

Unit tests

You can run the unit tests by running:

export PYTHONPATH=$PYTHONPATH:./src
python -m unittest discover