Robotics-Integration-Group-13

---

Jetson Orin NX

1. System Environment config

Environment	Version
JetPack	6.2.1
Ubuntu	22.04 Jammy
CUDA	12.6.68
cuDNN	9.3.0.75
VPI	3.2.4
Vulkan	1.3.204
Python	3.10.12
OpenCV	4.11.0 with CUDA: YES
PyTorch	2.5.0 (torchvision 0.23.0)
TensorRT	10.3.0

Suggestion: Use jtop to check the system config.

2. Compilation Environment config

Before configuring the environement, run sudo apt-get update and sudo apt-get upgrade at first.

1. Python

• Use miniconda to manage python version.
• Run conda create --name xxx python=3.10 to create virtual env

2. Ultralytics (YOLO)

• Install ultralytics dependencies: pip install ultralytics
• Uninstall torch and torchvision for they're cpu version: pip uninstall torch torchvision (See belows)
• Reinstall numpy for its too-high version: pip install "numpy<2" --force-reinstall

3. OpenCV

• Install nano / orin: sudo apt-get install nano

  Don't worry it's also compatible with orin nx

• Install dphys-swapfile: sudo apt-get install dphys-swapfile

• Enlarge the boundary of CONF_MAXSWAP: sudo nano /sbin/dphys-swapfile

• Restart the nano / orin: sudo reboot

• Check memory: free -m

• Run automator script OpenCV-4-11-0.sh:

  • Grant permissions: sudo chmod 755 ./OpenCV-4-11-0.sh
  • Run: ./OpenCV-4-11-0.sh

• Then check the installation of opencv-cuda-version is okay

  Thanks to Q-engineering I finally did it!

4. Pytorch

• Uninstall cpu-pytorch: sudo apt-get uninstall pytorch
• Install gpu-pytorch from a prebuild wheel:

pip install https://github.com/ultralytics/assets/releases/download/v0.0.0/torch-2.5.0a0+872d972e41.nv24.08-cp310-cp310-linux_aarch64.whl

See Resources Pytorch for Jetson

5. TensorRT

• Install tensorrt from a prebuild wheel:

pip install https://github.com/ultralytics/assets/releases/download/v0.0.0/torchvision-0.20.0a0+afc54f7-cp310-cp310-linux_aarch64.whl

See Resources TensorRT for Jetson

6. ONNX

• The onnxruntime-gpu package hosted in PyPI does not have aarch64 binaries for the Jetson. So we need to manually install this package.
• Install onnxruntime-gpu: With Python3.10 on our jetson orin nx, here we install onnxruntime-gpu 1.20.0 by:

pip install https://github.com/ultralytics/assets/releases/download/v0.0.0/onnxruntime_gpu-1.20.0-cp310-cp310-linux_aarch64.whl

---

Reference

[1] YOLO Official Web: Ultralytics YOLO Docs
[2] Robomaster Support: Robomaster SDK Ultra
[3] Training on Intel Arc GPU: intel-extension-for-pytorch & ultralytics issue #19821

---

⚠️ License: This project isn't open-source. See Details LICENSE.