[Jetson] Can I build and run full Autoware on Jetson if edge_auto_jetson_launch already works?

[WintonHsu]

Hi Autoware team,

I'm currently working on an NVIDIA Jetson Orin NX with JetPack 6.2 (L4T R36.4).

I have successfully built and run edge_auto_jetson_launch, including perception and planning modules.

However, when I try to build the full Autoware stack using:

colcon build --symlink-install --cmake-args -DCMAKE_BUILD_TYPE=Release

I encounter multiple build failures related to packages that seem not to support Jetson or aarch64, such as:

• spconv

• cumm

• onnxruntime with TensorRT

• Some CUDA-related components

Questions:

1. Is it possible to build and run the full autoware on Jetson (JetPack 6.2)?

2. Are there any recommended packages to skip or disable when building on Jetson?

3. Should I stick with edge_auto_jetson and selectively add required modules from autoware, or is there a known working setup for full autoware on Jetson?

Thanks in advance for any suggestions or shared experience!

– WintonHsu

[liuXinGangChina]

please list your cuda and tensorrt version here
then @knzo25 kenzo （expert in ai model） san can tell whether it works with your env like ‘spconv’

happy hacking

心刚

[WintonHsu]

📦 JetPack: 6.1.1 (L4T R36.4.3)
🔧 CUDA: 12.6 (V12.6.68)
🧠 TensorRT: 10.3.0.30 (CUDA 12.5+ compatible)
🧩 cuDNN: 9.3.0.75 (CUDA 12.6)
📍 Platform: Jetson Orin / aarch64

[liuXinGangChina]

hi kenzo san @knzo25

dose spconv support orin with cuda version above？

[urusaidog]

https://github.com/autowarefoundation/spconv_cpp/releases
has some arm releases but they probably will not work on a jetson. For that, the library would need to be built in a jetson

[lsteinosk]

This repo has some integrations with spconv from nvidia side that may help out here https://github.com/NVIDIA-AI-IOT/Lidar_AI_Solution