Skip to content

Latest commit

 

History

History
50 lines (35 loc) · 2.42 KB

README.md

File metadata and controls

50 lines (35 loc) · 2.42 KB

JetRacer_Pro - Line following by deep learning

image

A. Flash JetRacer Pro AI Kit Image

  1. Please download the JetRacer image through this link:https://drive.google.com/file/d/1bgCAUJ9m16g5FGuYgKy3WmGI-pGjPXN0/view
  2. Prepare a SD card (min 32GB) and flash the image to the SD card by Etcher

image

  1. Insert to Jetson Nano and power it up.

B. Environmnet installation and setup

Please do not type sudo apt-get upgrade into terminal otherwise the camera cannot open and you may need to reflash the image again

Please connect to a display and connect jetson nano to a network. Afterward, the OLED on JetRacer should be able to display the ip address of itself. If yes, you can contine the following step.

image

  1. Open Jupyter notebook with your ip address through browser (e.g. 192.168.0.134:8888) image

  2. Configure power mode of Jetson Nano

$ sudo nvpmodel -m1
$ sudo nvpmodel -q

The response of nano should be MODE : 5W

C. Train your own dataset

  1. Open interactive_regression.ipynb and click the cell sequentially
  2. After click all the cell, the view of the notebook will become like that image
  3. Open teleoperation.ipynb to control the car and move it around the track
  4. Keep clicking on the centre of the track on the left image and move the car via controller to different place to get the pictures
  5. Set epochs to 10 when collect enough data
  6. Click evaluate to view the model result and click save model button to save it

D. Example

  1. Two example code are offered for reference, please refer your own situation to change PID parameter
  2. follower_0.17_pwm.py for having car.throttle = 0.17 and the setting will shown below

image

  1. follower_0.18_pwm.py for having car.throttle = 0.18 and the setting will shown below

image

E. Further develop

  1. The JetRacer Pro can implement with deep reinforcement learning, please refer to https://github.com/masato-ka/airc-rl-agent