Street is danger, why not making autonomous toy?
Autonomous Remote Controlled (toy) car using deep learning. Raspberry Pi, PiCar, Python
You can find hardware on here, I'm too lazy to build from scratch.
Inspired by Robot Operating System (ROS) each components is a Node and run in it own process. A Node can be sensors, controller, motion planner, camera, Deep Learning pilot... which talk to each other via pub/sub like system.
Current implementation using Python's built in multiprocessing manager that allow sharing objects, data between Processes. Pros: Utilize CPU cores, avoid PIL, run nodes over network from different machine. Cons: Data are being pickle/unpickle multiple time and send over network which is slower than thread. May consider using more decend message queue like Redis.
 |
 |
|---|---|
| Paper road | Duct tape border |
- Raspberry Model 2/3 or greater
- Python 3.5
- OpenCV (optional) or PyGame camera (just for capture video)
- SD card 8gb or more
- Car kit (robot HAT, mortor controller... mine just use picar)
- Webcam or Picamera
- Setup Hardware
- Setup software
- Run, record
- Training
Hit record button while driving. A pair of image and json file including steering angle, speed will be produce for each frame. Record folder located at
noahcar/autorc/training-set/copy to your PC for training.
- Write new model: Noahcar.ipynb
- Transfer leaning
- Autopilot
cd noahcar
. env/bin/activate
python manage.py start
# Or
./manage.py -p <profile_name> start