NVIDIA _1-JetsonNano-开箱大全

桌面系统

lightdm桌面系统比gdm3可以少用500m内存

usb端口号别名

https://www.clearpathrobotics.com/assets/guides/kinetic/ros/Udev%20Rules.html

依据
主板存在多个USB，每个USB注册在操作系统固定的设备树路径，可用如下命令查看，得到下图。

$ ls -l /sys/bus/usb-serial/devices

1）串口编号随设备插入USB的时间先后分配序号；
2）路径则相对操作系统固定

查询usb设备连接属性
得到ATTRS参数列表（见后图）,从父子结构属性树的某一层，找出一组：
KERNELS,SUBSYSTEMS,DRIVERS

$ udevadm info --name=/dev/ttyUSB0 --attribute-walk

配置一个规则文件可以,独用或混合使用ATTRS,ENV

$ sudo vi /etc/udev/rules.d/99-usb-serial.rules
SUBSYSTEM=="tty", SUBSYSTEMS=="usb", DRIVERS=="cp210x", KERNELS=="1-2.2:1.0", MODE="0666", SYMLINK+="wheeltec_base"
生效规则文件, 测试(验证是否和上面同样命令的执行结果相同)
$ sudo udevadm trigger
# 测试
$ udevadm info --name=/dev/wheeltec_base --attribute-walk

常用手册

https://github.com/jetsonhacks/jetsonUtilities
1）常用命令

先安装jetson_stats
https://github.com/rbonghi/jetson_stats

$ sudo jtop 
$ jetpack版本
$ cat /etc/nv_tegra_release

opencv版本

$ cd /usr/bin
$ ./opencv_version
* cuda版本
$ cd /usr/local/cuda/bin
$ nvcc -v
$ gst-inspect-1.0 --gst-version

排错

Failed to fetch ubuntu bionic arm64 packages
https://github.com/tuna/issues/issues/595#issuecomment-613789709

$ sudo vi /etc/apt/source.list
...
deb https://mirrors.aliyun.com/ubuntu-ports xenial main restricted universe multiverse
deb https://mirrors.aliyun.com/ubuntu-ports xenial-security main restricted universe multiverse
deb https://mirrors.aliyun.com/ubuntu-ports xenial-updates main restricted universe multiverse
deb https://mirrors.aliyun.com/ubuntu-ports xenial-proposed main restricted universe multiverse
deb https://mirrors.aliyun.com/ubuntu-ports xenial-backports main restricted universe multiverse
...
$ sudo apt-get update

Package nvidia-l4t-xusb-firmware is not configured yet.
因将/usr/bin/python指向python3导致问题
https://forums.developer.nvidia.com/t/errors-on-nvidia-l4t-xusb-firmware-and-nvidia-l4t-initrd/159209

$ sudo vi /usr/sbin/l4t_payload_updater_t210
#!/usr/bin/python
to
#!/usr/bin/python2

安装 jetson-inference

参考
https://github.com/dusty-nv/jetson-inference
https://www.rs-online.com/designspark/hands-on-with-the-nvidia-jetson-nano

$ sudo apt update
$ sudo apt install git cmake libpython3-dev python3-numpy
$ git clone --recursive https://github.com/dusty-nv/jetson-inference
$ cd jetson-inference
$ mkdir build
$ cd build
$ cmake ../
$ make
$ sudo make install
$ sudo ldconfig

BLAS和LAPACK的openblas安装

http://jungle.webhop.me:81/admin/index.html#/posts/edit?postId=273

安装pcl

包括pcl的依赖，vtk/flann/glew/boost更新/等等
http://jungle.webhop.me:81/admin/index.html#/posts/edit?postId=143

安装opencv

opencv安装，包括ffmpeg，gstreamer等依赖库安装
http://jungle.webhop.me:81/admin/index.html#/posts/edit?postId=81

ffmpeg安装

参考
https://forums.developer.nvidia.com/t/hardware-accelerated-video-playback-with-l4t-ffmpeg/154019/7

- 删除老版本
$ sudo dpkg -l | grep ffmpeg #查看安装
$ sudo dpkg -r ffmpeg 
$ sudo apt-get purge '*ffmpeg*'
# 下方删除要谨慎
$ sudo find /usr -name "*ffmpeg*" -exec rm -rf {} \;
- 安装
$ echo "deb https://repo.download.nvidia.com/jetson/ffmpeg main main" |  sudo tee -a /etc/apt/sources.list
$ echo "deb-src https://repo.download.nvidia.com/jetson/ffmpeg main main" |  sudo tee -a /etc/apt/sources.list
$ sudo apt update
$ sudo apt install ffmpeg
- 检查
$ ffmpeg -decoders | grep nvv4l2

安装TensorFlow.1.x

参考
https://docs.nvidia.com/deeplearning/frameworks/install-tf-jetson-platform/index.html
https://docs.nvidia.com/deeplearning/frameworks/install-tf-jetson-platform-release-notes/tf-jetson-rel.html

# Install system packages required by TensorFlow:
$ sudo apt-get update
$ sudo apt-get install libhdf5-serial-dev hdf5-tools libhdf5-dev zlib1g-dev zip libjpeg8-dev liblapack-dev libblas-dev gfortran

# Install and upgrade pip3.
$ sudo apt-get install python3-pip
$ sudo pip3 install -U pip testresources setuptools==49.6.0 

# Install the Python package dependencies.
$ sudo pip3 install -U numpy==1.19.4 future==0.18.2 mock==3.0.5 h5py==2.10.0 keras_preprocessing==1.1.1 keras_applications==1.0.8 gast==0.2.2 futures protobuf pybind11
$ sudo pip3 install --pre --extra-index-url https://developer.download.nvidia.com/compute/redist/jp/v45 'tensorflow<2'

测试

$python3
>>>import tensorflow as tf

g）测试cv2

$ workon pyenv_ros2
$ python
>>> import cv2
>>> print(cv2.__version__)

安装pytorch/LibTorch

手册
https://qengineering.eu/install-pytorch-on-jetson-nano.html(包括：pytorch/LibTorch/Caffe2/TorchVision)
https://forums.developer.nvidia.com/t/pytorch-for-jetson-version-1-9-0-now-available/72048（TODO，好像是CPU版本）

$ python
>> import torch
>> torch.__version__
>> torch.cuda.is_available()
True  <- 检查

安装TorchVision

https://qengineering.eu/install-pytorch-on-jetson-nano.html

检查报错

$ python3
> import torchvision as tv
...
PIL.Image Has No Attribute Image

解决：

#删除在/usr/lib/python3/有个pillow 5x版本
$ sudo rm -rf /usr/lib/python3/dist-packages/Pillow-5.1.0.egg-info
$ sudo rm -rf /usr/lib/python3/dist-packages/PIL
# 新安装pillow
$ sudo pip3 install -U pillow
将安装与/usr/local/lib/python3.6/dist-packages

Torchvision is a collection of frequent used datasets, architectures and image algorithms.

编译onnxruntime

https://medium.com/vitrox-publication/build-onnxruntime-from-source-on-windows-10-e85ea5fe71db
https://elinux.org/Jetson_Zoo#ONNX_Runtime
http://article.docway.net/details?id=6046eea04da5fa50e15091ca

- 下载最新
$ git clone https://github.com/microsoft/onnxruntime.git -b v1.8.2
- 编译
$ cd onnxruntime

- 修改cmake/CMakeLists.txt
option(onnxruntime_BUILD_UNIT_TESTS "Build ONNXRuntime unit tests" OFF)

$ ./build.sh --config Release --skip_submodule_sync --build_shared_lib --build_wheel --parallel --use_openmp --use_cuda --skip_tests --cudnn_home "/usr/local/cuda-10.2" --cuda_home "/usr/local/cuda-10.2"

- 安装
$ cd build/Linux/Release/dist
$ sudo python -m pip install onnxruntime_gpu-1.8.2-cp36-cp36m-linux_aarch64.whl

- 验证
>> import onnxruntime as rt
>> rt.get_device()
'GPU'

安装vscode

1）下载
https://github.com/futurejones/code-oss-aarch64/releases
2）安装
https://code.visualstudio.com/docs/setup/linux

$sudo dpkg -i code-oss_1.44.2-1587955852_arm64_ubuntu-18.04.deb

安装arduino

https://www.instructables.com/To-Install-Arduino-Software-IDE-on-Jetson-Nano-Dev/

增加系统交换空间

https://www.jetsonhacks.com/2019/11/28/jetson-nano-even-more-swap/

$ git clone https://github.com/JetsonHacksNano/resizeSwapMemory
$ cd resizeSwapMemory
$ ./setSwapMemorySize.sh -g 8

深度学习起步

https://gilberttanner.com/blog/jetson-nano-getting-started

MPI Solutions for GPUs

https://developer.nvidia.com/mpi-solutions-gpus
https://www.open-mpi.org/faq/?category=buildcuda

配置opencv到pyenv_ros2（不需要，这个方法可以了解）

$ cd ~/.pyenvs/pyenv_ros2/lib/python3.6/site-packages
$ ln -s /usr/local/lib/python3.6/site-packages/cv2/python-3.6/cv2.cpython-36m-aarch64-linux-gnu.so cv2.so

从硬盘运行系统

1）准备
a）主要参考文章（视频）
https://www.jetsonhacks.com/2019/09/17/jetson-nano-run-from-usb-drive/
（老文章https://www.jetsonhacks.com/2019/04/25/jetson-nano-run-on-usb-drive/）
b）Linux内核主要相关的概念：
Jetson bootloaders|an initrd (initial ramdisk)|Linux kernel|USB driver|USB firmware
c）相关概念参考文章
https://blog.csdn.net/seek_0380/article/details/106563825
https://stackoverflow.com/questions/10603104/the-difference-between-initrd-and-initramfs
d）下载脚本
https://github.com/JetsonHacksNano/rootOnUSB

$ cd
$ git https://github.com/JetsonHacksNano/rootOnUSB.git
$ cd rootOnUSB

2）动手了，只需4步
///////////////////////////

网络配置

查看当前网络配置
$ ll ll /etc/NetworkManager/system-connections
重启网卡服务
$ sudo /etc/init.d/networking restart

桌面共享

https://www.fatalerrors.org/a/jetson-nano-configuration-vnc.html
https://github.com/overclock98/Jetson_Nano_true_Headless_setup_without_hdmi_display

安装
完善nano 4g一个bug（nano_2g_L4T 32.6.1正常）

$ vi /usr/share/glib-2.0/schemas/org.gnome.Vino.gschema.xml
# 添加（若缺少）
    <key name='enabled' type='b'>
      <summary>Enable remote access to the desktop</summary>
      <description>
        If true, allows remote access to the desktop via the RFB
        protocol. Users on remote machines may then connect to the
        desktop using a VNC viewer.
      </description>
      <default>true</default>
    </key>
# 重新编译
$ sudo glib-compile-schemas /usr/share/glib-2.0/schemas

配置远程访问
选择vino-server（XRDP占用内存高）

通过命令行配置

# Disable encryption for the VNC server
$ gsettings set org.gnome.Vino require-encryption false
$ gsettings set org.gnome.Vino prompt-enabled false

# Find the UUIDs of your connections
$ nmcli connection show

# Replace UUID with yours as a comma separated list in the square brackets of the configuration line
$ dconf write /org/gnome/settings-daemon/plugins/sharing/vino-server/enabled-connections "['UUID']"

# 6-Enabling Automatic Login (otherwise cant start from ssh): steps
$ sudo nano /etc/gdm3/custom.conf
添加或启用
AutomaticLoginEnable = true
AutomaticLogin = yourname

# 配置/etc/X11/xorg.conf
$ vi /etc/X11/xorg.conf
在文件末尾添加
Section "Screen"
    Identifier    "Default Screen"
    Monitor       "Configured Monitor"
    Device        "Tegra0"
    SubSection "Display"
        Depth   24
        Virtual 1920 1600 # Modify the resolution by editing these values
    EndSubSection
EndSection

# 重启，启动vino server
$ sudo reboot
$ export DISPLAY=:0 && /usr/lib/vino/vino-server
若在有物理显示器的终端，则直接启动
$ /usr/lib/vino/vino-server

通过GUI配置(若有显示器)
1）打开窗口 : 系统设置 > desktop sharing

2）设置控制参数
$ sudo apt-get install dconf-editor
$ dconf-editor

3）自动启动

打开设置程序: Start the application

去掉桌面环境GUI

https://lunar.computer/posts/nvidia-jetson-nano-headless/

连接PCA9685 PWM driver

1）更新GPIO库
1）本系统和树莓派的RPi.GPIO的API完全相同（开发的应用程序模块可兼容）
2）下载安装GPIO库
https://github.com/NVIDIA/jetson-gpio
...
3）安装PCA9685 servo library
pip3 install Adafruit_PCA9685
4）引脚连线（左侧为PCA9685模块）
VCC <-> 3.3v(1)
SDL <-> SDL(3)
SCL <-> SCL(5)
GND <-> GND(9)
图示见：http://docs.donkeycar.com/guide/build_hardware/#step-4-connect-servo-shield-to-raspberry-pi

Jungle Blog