Self-Driving RC-Car project

ROS Software for performing autonomous driving using monocular vision on an RC car.
This package was designed to run on an Nvidia Jetson TX2 using a PS4 controller. It provides two driving mode:

• manual: The default mode. In this mode, you can drive the RC Car manually using the paired PS4 controller. You can also recording driving sequences to assemble new training data. Pressing L1, you can retrain a model from scratch.
• autonomous: The self-driving mode. This mode enables the trained residual network to perform steering and throttle inference from input images.

RC Car PS4 Controls

Getting started:

Software

To install the ros package run:

cd $CATKIN_WS/src/  
git clone https://github.com/germain-hug/Autonomous-RC-Car.git  
cd .. && catkin_make

To install dependencies run:

pip install -r requirements.txt

You will also need to have OpenCV installed.

Hardware

Here is a non-exhaustive list of the hardware used in this project:

• Nvidia Jetson TX2
• Orbitty Carrier Board
• Traxxas 4WD RC-Car
• Pololu Micro Maestro Controller
• Spinel 2Mp / 140° FOV USB Camera
• 4000mA 11.1V LiPo Battery
• Playstation Dualshock 4 Controller
• LiPo Battery Charger

Hardware assembly should be fairly simple, and additional ressources can be found here

Running the software

To launch the software, run:

roslaunch rccar.launch

Node description

• mode_management.py: Mode manager, activates manual and self-driving modes through /mode and allows for model retraining.
• manual_driver.py: Publishes PS4 manual commands on /cmd, enables data capture mode.
• self_driver.py: Streams from camera, and publishes inferred throttle and steering commands on /cmd
• actuator.py: Receives throttle and steering commands through the /cmd topic, pre-processes them transmits them to the servo-controller.