Welcome to the DNRacing Repo! This repository is for a scaled V2V autonomous motorsport project involving two 1:8 scale RC cars. All of the work was done with ROS Kinetic and Keras. The project was recently accepted to the upcoming IEEE ITSC 2018 conference. The project build is written below.
<<Hardware>>
Note: I am selling a big portion of my hardware; please let me know if you are interested in purchasing!
Computing Unit
Nvidia Jetson TX2 Developer Board : The TX2 is futureproof, but the TX1 will also run the 3D_CNN architecture with 2 stacked images.
Arduino Uno : Arudino Megas are also a good option.
FPV RX/TX Interface
Amimon Connex : I tried UDP wifi video streaming, but the latencies were too high. The wireless HDMI transmitter from Amimon Connex has no latency, but it is expensive. An analogy FPV TX/RX setup will also work for those who are on a budget.
Futaba RX/TX : The trainer port must have trainer/master mode capability for the PPM-to-USB (Compufly V2) cable to stream the signals back to the PC and receive commands from the steering wheel.
Sensors
Ximea XiQ CMOS Camera with 1.8mm fisheye lens : Fish eye lens are highly recommended for a wide field of view (FOV). I mounted the cameras really high with a long mobile phone stand because I wanted a larger FOV.
Chassis
Vehicle #1: OFNA DM1-Spec E
Vehicle #2: Hong Nor X3-GT
*Note: The vehicles above were chosen because they were fast. However, other 1/10th scale RC cars with standard servos and ESCs will also work.
Drive Train
Tekin RX-8 ver.2 Electronic Speed Controller : High end ESCs are recommended for quality low speed control with brushless motors. Sensored motors are also highly recommended.
11T Steel Pinion Gear & 48T Steel Spur Gear : Both the spur gears and pinion gears should be steel because plastic ones tend to chip off after prolonged use.
TrackStar 2650Kv Brushless Motor
PowerSupply
GoalZero Sherpa 100 Battery Pack : Although it is possible to buy other alternative batteries, remember that they should at least carry a power output of 30W.
5000mAh 11.1v LiPo Battery Pack
Other
USB 3.0 7-port Expansion Hub
Logitech G25/G27/G29 : The G25 is the best choice if you are on a budget.
Compufly V2 USB-to-PPM USB Cable : This converts the Logitech Steering Wheel commands into PPM signals for the Futaba transmitters to read.
VRC-3NT PWM-to-USB USB Cable : This converts the PWM signals from a standard RC receiver into signals readable by the ROS joy pkg.
<<Software>>
First, make the ROS Catkin_PKG_Car package and solve any dependecy issues. The race_robot folder contains the launch file (sixthtest.launch) for data collection. To start collecting images while driving, type
roslaunch race_robot sixthtest.launch
The controller folder contains the G27/G29 joystick commands which will be recorded in the rosbag. To start recording the images and controll commands, type
rosbag record controller controller_two camera/left/image_raw
<CNN_Model>
The end-to-end, behavioral cloning 3D CNN model uses a stack of images with specified jumps.
data_preprocessing_v2.py processes the .csv file which contains throttle, steering, brake inputs correlated with the corresponding image.
deploymodel.py deploys the keras model within ROS using the joy catkin package
model.py contains the actual 3D CNN model implementation
run_3d_cnn.py creates the 3D CNN model with the .csv file and the image folder
<Catkin_PKG_Car/src>
This folder contains all the ROS catkin packages used by the DNRacing Car.
<USBtoPPM>
This Java application is used to convert the USB HID commands sent from the Logitech G27 Steering Wheel into PPM signals.
<Useful_Tools>
usbfinder.sh is a script used to find all the usb serial devices connected to the PC.
imageconverter.py is a script which converts the images obtained from the XiQ cameras. (e.g. Downsampling)
<Past_Work>
This folder contains scripts used previously (e.g. for AHRS IMUs and MarvelMind IPS) which may be of some use.