Intel Realsense D415 on NVidia Jetson TX1

[LBall1]

Hello,

I have been trying to find a way to use my intel Realsense Camera with my nvidia jetson tx1. I am kind of disappointed that the new intel SDK does not support arm CPUs which kind of defeats the purpos of putting the cameras out for everyone in my eyes.

Intel wants people to make autonomouse robots, but there are no really powerful small intel development boards that are comparable to a jetson.

If anyone has gotten the D400 series of cameras running or has an idea how it could work on the nvidia Jetson I would be happy if you could help out !

[MartyG]

The 400 Series cameras work with any Intel or ARM processor. Although there have been a couple of recent hiccups in regard to using the SDK 2.0 software with ARM-based devices such as ODROID, these are getting smoothed out. A new Preview build of the SDK (version 2.10.3) was released today with some bug-fixing for ARM.

[ARyba]

lballenweg, DXXX cameras do work with Jetson. There're some USB3 specific instabilities, which hopefully will be fixed by Intel by adding USB2 support in the next firmware release(s). You want to compile the librealsense2 library yourself.

https://github.com/IntelRealSense/librealsense/issues/1424 Installing D435 on Jetson Tx2 · Issue # 1424 · IntelRealSense/librealsense · GitHub

-albertr

[LBall1]

Well the problem I have is when adding librealsense to the repository list i get an error when doing sudo apt-get update which says that the arkitecture Armhf us not supported...

[MartyG]

Apologies in advance that my ARM knowledge is very limited, so I learn as I go!

Armhf seems to be a rather complicated subject. If I read my research correctly though, it seems to boil down to this:

- Armhf is a 32-bit version of ARM that has floating-point support.

- It is older than ARM64 architecture

- The Nvidia Jetson TX1 has a 64-bit ARM processor

- It started off with a 32-bit CUDA and updated later on.

Do you have the latest CUDA on your Jetson, one with 64-bit support? Nvidia seem to recommend using CUDA 8 on the TX1.

https://devtalk.nvidia.com/default/topic/1008072/jetson-tx1/the-best-cuda-version-for-tx1/ https://devtalk.nvidia.com/default/topic/1008072/jetson-tx1/the-best-cuda-version-for-tx1/

[LBall1]

I am not sure if I am doing the install correctly...

I used the repository and tried to install librealsense over it, and I got a similar error, only this time with arm64.

The one other option I think i have is building it from scratch, but I do not have any experience doing that.

Maybe the cause of the Problem is simple, but I do not really know to much about installing such package.

[LBall1]

I think I have cuda 6.0 on it... it is the version that came with jetpack 3.1

[MartyG]

The information I read indicated that the 64-bit support came in shortly after CUDA 7.5. So CUDA 8, if it is possible to update to that, may be helpful to you.

[LBall1]

Okay I will try that, it would make sense if that is the cause... i belive jetpack 3.2 has support for the jetson tx1 or is it only the tx2 ? If not i will have to install cuda just so...

[MartyG]

It seems that Jetpack 3.2 supports TX1. The Jetpack website says:

"JetPack 3.2 with L4T R28.2 is the latest production software release for NVIDIA Jetson TX2, Jetson TX2i and Jetson TX1. It bundles all the Jetson platform software, including TensorRT, cuDNN, CUDA Toolkit, VisionWorks, GStreamer, and OpenCV, all built on top of L4T with LTS Linux kernel.

https://developer.nvidia.com/embedded/jetpack JetPack | NVIDIA Developer

[PSnip]

Hi,

I have been successfully using D4xx on TX2 Dev-kit. Since it works fine on TX2, I see no reason why it should not work on TX1.

So, most probably it is just a configuration issue, and not the feasibility issue.

[LBall1]

Well, I tryed loading JetPack 3.2 to my Jetson, did all of the setup in the loader correctly, and now it does not recognise any ports anymore. I must add it sayd in the logs that the install of everything was succesful, but my new problem is not related to this forum. So thePaintedSnipe which version of JetPack or cuda are you runnuing ?

[PSnip]

Hi,

I am running Jetpack 3.1 - release L4T R28.1 on TX2 Devkit. And though I have not tested 3.2 myself, I am aware of people having tested D4xx on Devkit with Jetpack 3.2 - L4T R28.2

From your last message, it is not very clear how you have installed librealsense. I have cloned the repo from below and used it https://github.com/IntelRealSense/librealsense/ https://github.com/IntelRealSense/librealsense/

https://github.com/IntelRealSense/librealsense/ GitHub - IntelRealSense/librealsense: Intel® RealSense™ SDK

I have tested tag 2.10.0 but later versions should as well work fine.

Build and install procedure is very easy, you should be able to follow instructions from doc/installation.md

If you get any build errros, you can discuss here or on the github.

Please let us know if you are able to install librealsense correctly or not.

[LBall1]

So I managed to build the SDK now and everything works fine. I saw earlier today that there also is a .sh file on the Jetsonhacks Github account, but it currently only fully supports the Jetson TX2. One thing that I noticed is that the RealSense Viewer allways uses one core and has it up to 100% load at all time. Does anyone know of a solution to that problem or will I just have to deal with it ? Also is there a way to add support for Cuda ?

Thanks for the Help with solving my main Problem, the other questions are just little things I was thinking about.

[idata]

Check this out: https://www.jetsonhacks.com/2018/04/25/now-cuda-intel-realsense-d400-cameras-nvidia-jetson-tx/ Now with CUDA! Intel RealSense D400 cameras – NVIDIA Jetson TX | JetsonHacks

[MartyG]

Great, TK, thanks for sharing!

If you are a Jetson user, you may also be interested in the announcement in the past week by Nvidia of Jetson Xavier, which is more than 20 times faster than TX2.