ut_automata

#

Table of Contents

JetPack on Nvidia DevKit
JetPack on ConnectTech Orbitty Carrier
CUDA + PyTorch
ROS 2 Optional Tools

You will require an xUbuntu 18.04 computer to run the Nvidia SDK Manager. As of the time this documentation was written, sdkmanager is incompatible with 20.04, but it can be run under a VMWare Workstation Player virtual machine running 18.04 - you just need to connect the Jetson USB device to the VM

JetPack on Nvidia DevKit

  1. Run sdkmanager on the host computer, and select both “Host Machine” and “Jetson TX2” to download JetPack to the host. image Select CUDA and additional packages as desired Click next past Step 02, accepting the license along the way.
  2. After Step 03 of the sdkmanager, the Jetpack image will be created, and it will ask you to connect the Jetson to the host computer: image
  3. Connect the devkit using the microUSB OTG port on the devkit and enter recovery mode:
    1. Plug in the power cable to the devkit board, but do not power on yet.
    2. Press and hold the REC (Recovery) button
    3. Press and release the POWERBTN (Power) button
    4. Release the REC button
  4. Select “Manual Setup” in the dropdown option list on the sdkmanager window, and click on the “Flash” button
  5. After JetPack is installed, connect the devkit to a monitor, keyboard and mouse, and complete the Ubuntu setup wizard.
  6. If you selected CUDA and additional packages in step 1, it will ask you for the IP address and login details of the devkit. Continue on to install CUDA + PyTorch

JetPack on ConnectTech Orbitty Carrier

  1. Do not connect the Jetson to your computer at this stage.
  2. Run sdkmanager on the host computer, and select both “Host Machine” and “Jetson TX2” to download JetPack to the host. image Click next past Step 02, accepting the license along the way.
  3. After Step 03 of the sdkmanager, the Jetpack image will be created, and it will ask you to connect the Jetson to the host computer: image
  4. Click on “Skip” and confirm that you want to skip installation to the Jetson: image
  5. At this point, you should see JetPack installed on your host computer at ~/nvidia/nvidia_sdk/: 
     joydeepb@ubuntu:~$ cd nvidia/nvidia_sdk
 joydeepb@ubuntu:~/nvidia/nvidia_sdk$ ls -lthr
 total 8.0K
 drwxrwxr-x 4 joydeepb joydeepb 4.0K May  1 09:16 JetPack_4.5.1_Linux
 drwxrwxr-x 3 joydeepb joydeepb 4.0K May  1 09:19 JetPack_4.5.1_Linux_JETSON_TX2
  6. Go to the Connect Tech support page and download the L4T support package into <JetPack_install_dir>/JetPack_4.5.1_Linux_JETSON_TX2/Linux_for_Tegra/. For example: 
     joydeepb@ubuntu:~$ cd nvidia/nvidia_sdk/JetPack_4.5.1_Linux_JETSON_TX2/Linux_for_Tegra/
 joydeepb@ubuntu:~/nvidia/nvidia_sdk/JetPack_4.5.1_Linux_JETSON_TX2/Linux_for_Tegra$ wget https://connecttech.com/ftp/Drivers/CTI-L4T-TX2-32.5-V001.tgz
 --2021-05-01 09:14:59--  https://connecttech.com/ftp/Drivers/CTI-L4T-TX2-32.5-V001.tgz
 Resolving connecttech.com (connecttech.com)... 104.26.6.94, 172.67.72.40, 104.26.7.94, ...
 Connecting to connecttech.com (connecttech.com)|104.26.6.94|:443... connected.
 HTTP request sent, awaiting response... 200 OK
 Length: 519331075 (495M) [application/octet-stream]
 Saving to: ‘CTI-L4T-TX2-32.5-V001.tgz’

 CTI-L4T-TX2-32.5-V001.tgz                                 100%[===================================================================================================================================>] 495.27M  15.5MB/s    in 32s     

 2021-05-01 09:15:31 (15.4 MB/s) - ‘CTI-L4T-TX2-32.5-V001.tgz’ saved [519331075/519331075]
  7. Extract the archive, and run the install script in the extracted directory: 
     joydeepb@ubuntu:~/nvidia/nvidia_sdk/JetPack_4.5.1_Linux_JETSON_TX2/Linux_for_Tegra$ tar xzf CTI-L4T-TX2-32.5-V001.tgz
 joydeepb@ubuntu:~/nvidia/nvidia_sdk/JetPack_4.5.1_Linux_JETSON_TX2/Linux_for_Tegra$ cd ./CTI-L4T
 joydeepb@ubuntu:~/nvidia/nvidia_sdk/JetPack_4.5.1_Linux_JETSON_TX2/Linux_for_Tegra/CTI-L4T$ sudo ./install.sh
  8. Connect the Jetson to the host computer and enter firmware update mode:
    1. Connect the USB OTG port
    2. Press and hold the recovery button
    3. Press and release the power button
    4. Release the recovery button
  9. Now reflash the Jetson from the Linux_for_Tegra directory: 
     joydeepb@ubuntu:~/nvidia/nvidia_sdk/JetPack_4.5.1_Linux_JETSON_TX2/Linux_for_Tegra$ sudo ./cti-flash.sh

CUDA and PyTorch

  1. Ensure that the Jetson is running, booted into Ubuntu, and accessible via the local network (e.g. WiFi) from the host computer. Let the Jetson’s IP address on the local network be 192.168.10.100
  2. On the host computer, run sdkmanager
  3. Select “Jetson” as the product category, deselect Host Machine, Jetson TX2 as the Target Hardware, and select the correct JetPack version installed on the Jetson.
  4. On step 2, deselect “Jetson OS” from the list of components and select CUDA and additional components: image
  5. Manually enter the IP address of the Jetson computer, and the admin username and password: image
  6. To install pytorch, follow the instructions here: https://forums.developer.nvidia.com/t/pytorch-for-jetson-version-1-8-0-now-available/72048

ROS 2

The current ut_automata code targets ROS 2 Jazzy. Install ROS 2 Jazzy using the official Ubuntu instructions:

https://docs.ros.org/en/jazzy/Installation/Ubuntu-Install-Debs.html

Jazzy targets Ubuntu 24.04. Older JetPack/L4T images for TX2-era hardware are based on Ubuntu 18.04 and are not a native Jazzy target; those systems will need either a newer supported platform image, a containerized build, or a different ROS 2 distribution matched to the OS.

After installing ROS 2, source the environment and build the workspace with colcon:

source /opt/ros/jazzy/setup.bash
cd ~/ut_automata_ws
colcon build --symlink-install
source install/setup.bash

Optional Tools

To monitor CPU and GPU usage, clone and install Jetson Stats image