E2E Driver (Self Driving Library)

Using:

• Jetson AGX Xavier JetPack 5.0.2 --- r35.1.0
  
• Arduino Nano
  

Configuration 1:

• Custom 3D Printed Body on top of modified Wltoys 124017 chassis
  
• ZED Mini Stereo Camera
  
• HC-020K Optical Encoder
  
• Turnigy Evolution Radio Control System
  

 

Configuration 2:

• Traxxas chassis
  
• ZED Stereo Camera
  
• Flysky FS-i6X Radio Control System
  
• VESC
  

 

Features

- Currently using behavioral cloning for driving autonomously
- Modular architecture. New parts can be added easily
- Parts works at high speed and low latency thanks to threading and multi threading,
 all parts can run min @100 Hz.
- Easy to set up and configure.
- Fast training and inferance
- TensorRT support
- TensorBoard support
- DAgger(Dataset Aggregation) support
- Data Augmentation support
- Cruise Control
- Merge multiple datasets with ease
- Reduced data sampling
- Early stopping
- Web interface

Web Interface

Low Latency Live Data:

 

Data Editing:

 

Getting Started

Pull docker image:

sudo docker pull mekala02/e2e-driver:developer

Run docker image:

nvidia-docker run -it --rm --privileged --network host --ipc=host -v $(pwd):/root mekala02/e2e-driver:developer