FUBAR Labs Autonomous Racing Vehicles

Autonmous Vehicle Project at Fubar Labs for the Autonomous Powerwheels Racing compeition.

• Autonmous Powerwheels Racing event will be Makerfaire NYC 2017
• Autonmous Vehicle Competition via Sparkfun at Denver Makerfaire 2017

Quickstart

Car Code

Prepare your PI

sudo apt-getupdate
sudo apt-get install python3-pip python3-dev

Get TensorFlow for PI installed (http://www.instructables.com/id/Google-Tensorflow-on-Rapsberry-Pi/)

wget https://github.com/samjabrahams/tensorflow-on-raspberry-pi/releases/download/v1.0.1/tensorflow-1.0.1-cp34-cp34m-linux_armv7l.whl

Install Tensorflow for Python 3.4

sudo pip3 install tensorflow-1.0.1-cp34-cp34m-linux_armv7l.whl

Fix a potential error message:

sudo pip3 uninstall mock
sudo pip3 install mock

Raspberry PI open cv installation:

Training system needs openCV for image review. It's a nice to have for the PI.

TODO: custom compilation information

The actual car code:

git clone https://github.com/fubarlabs/autonomous
cd autonomous
virtualenv auto -p python3 
source auto/bin/activate
pip install -r requirements.txt

Note for Arduino

Code is installed from the Raspberry PI using PLatform IO

sudo pip install platformio

Platform IO is only Python 2.7 but it can program the Arduino. In our chase it's the Fubarion SD board.

Fubarino SD / Arduino Code

Arduino code location: ./autonomous/arduino

cd arduino

MOTTO: Small RC Car Collection Code: MOTTOServoDtaSampleDelay.ino

cd MOTTOServoDataSampleDelay
pio run -t upload

Full Auto Code: MOTTOFullAutoDrive.ino

cd MOTTOFUllAutoDrive.ino
pio run -t upload

OTTO: Power Wheels Autonomus Collection Code: NewOTTOFullAutoDrive.ino

cd NewOTTOFUllAutoDrive.ino
pio run -t upload

Full Auto Code: NewOTTOFullAutoDrive.ino

cd NewOTTOFUllAutoDrive.ino
pio run -t upload

Data Collection Code

Data collection is done as a Raspberry PI service. The folder services contains:

1. ottologger.py

2. ottologger.service

3. Copy ottologger.py /usr/local/bin

4. register ottologer.service as a system service

5. Switch on pin 4 enables and disables collection

6. LED on pin 11 shows the status of collection

Training code

More documentation at the wiki

Autonomous Project Documenatation

Code details

Simple model in basic_model.py. Currently linear with mean squared error loss (summed over outputs?)

Inputs

• Webcam image
• Current Accel
• Current Speed

Future Inputs

• Current Distance from rangefinder
• Current Steering wheel angle

Outputs

• Steering Wheel angle
• Maybe speed?

Data sources

• DeepDrive
  • Compressed subset of DeepDrive

Notes

Driving model is in current_model.py. Weights are on Google Drive. Line 74 of the model will have to be changed to reflect the true location/name of the weights file.