Jetson Nano

I know some teams have been working with the TK series.

This looks like it will help more teams start experimenting with AI stuff.
$99 vs $500+

2 Likes

I’m very intrigued. I might pick one up for my side projects at home.

NVIDIA Jetson Nano modules will be available from distributors worldwide starting June 2019.

In case anyone else was wondering why it all said pre-order and is blind like me.

This does look interesting though. Would make a very nice upgrade to a Raspberry Pi if we somehow overrun what we can do with that.

I think that is an understatement. 128 CUDA cores. Of course, the GPUs have so far been not useful for the (usual) simple image processing done in FRC, but it should make encoding to H.264 really easy.

1 Like

I don’t know where the $500 number is coming from but the Jetson TX2 is available for $299 for FRC Teams, which is cheaper than the RoboRIO for a heck of a lot more processing power.

Some other fun stuff about the nano platform though:

NVIDIA has created a JetBot educational platform that should be a good starting point: https://github.com/NVIDIA-AI-IOT/jetbot

It will have support for AWS RoboMaker out of the gate too: https://www.therobotreport.com/nvidia-jetson-platform-now-supports-aws-robomaker/

1 Like

In our experience, it already is easy - just use OpenMAX instead of software encoding. :wink:

Just saw it too and immediately thought of FRC applications. The size of it looks like the key factor but I haven’t really compared it to the Jetson TK series.
I always wandered, as my team never used a co-processor for image processing, is it better than using a Raspberry Pi or are most teams could be using either one for the tasks of an FRC game?

Link for those looking:

1 Like

FYI: Jetson Nano’s ship next Monday: order here.

They’ve been shipping for a bit, I think they’re just production constrained.

1 Like

Can confirm.

I’ll also mention that everyone interested in the details of getting running on a new Jetson can best find information and ask questions not on CD but rather on the official NVIDIA Jetson and Embedded Systems Forum.

Just be mindful that NVIDIA isn’t likely to have answers to FRC specific questions but as these developer kits are VERY new, it’s important to know where to get the latest info and get questions answered.

And as always, if you’re wanting to know something about Jetsons and FRC then we’re always willing to help however we can, get in touch at support@team900.org.

1 Like

I heard WPI will make the same image for it as the raspberry pi

I gotta say - I struggled all weekend to get allwpilib to compile on the nano with no luck.

I re-read the README-CMAKE again tonight and have a list of things to try, but the errors I was getting just trying to do a ‘cmake’ and not even getting to the point of compilation make me think there’s more issues that will be solved by simply reading more carefully.

Anyone had success compiling allwpilib for nano yet?

bob - FRC Team 1967 - The Janksters

I noticed the WITHOUT_JAVA option wasn’t working
(no idea why) and after manually hardcoding it to false in the various CMakeLists.txt files for the projects I’ve got wpilib compiled and installed.

Also make sure you’re clearing our your cmake build folders between config changes, that tripped me up a bunch.

Can you be more specific? N00b to Jetson here.
What do we download?
What needs to be in place before trying to compile?
Any details would be helpful for the community

Thanks Woolly - we did wind up getting a successful build of allwpilib without Java on the nano. And we managed to get our camera capture code from last year up and running on it. PHEW! We re-organized our code tree a little bit to accomodate the allwpilib piece and added it as a submodule this year to our repo which looks super clean and will automatically follow the changes in allwpilib too. Slick. BTW, I think we did get tripped up SEVERAL times by not clearing out our ‘buildcmake’ folder prior to re-doing a ‘cmake’ again.

I believe we did:
cd allwpilib ; mkdir buildcmake ; cd buildcmake
cmake -DWITHOUT_JAVA=on …
make

Then in our vision code I think we needed to add a build option of -std=c++17 as well.

bob

BTW - I dont know why, but when I look at the formatted output of my message above, it looks like there are THREE ‘dots’ after the cmake -DWITHOUT_JAVA=on when in reality, there should be TWO ‘dots’.

You’ll want to clone or download the allwpilib repository and follow this readme https://github.com/wpilibsuite/allwpilib/blob/master/README-CMAKE.md to compile and install wpilib onto your jetson, making sure to use the WITHOUT_JAVA option when running cmake on allwpilib.

Thanks! This worked for me, just like bob it looks like.
Now… I need to get some code to capture/stream video. I’m hoping to be able to use the nano to do the inferencing much like the rPi + GoogleAddOnThing

This topic was automatically closed 365 days after the last reply. New replies are no longer allowed.