search

NVIDIA-AI-IOT / jetracer

An autonomous AI racecar using NVIDIA Jetson Nano
MIT License
1.06k stars 319 forks source link

A few small setup issues with new Jetpack 4.5 release #99

Closed zlite closed 2 years ago

zlite commented 3 years ago

Looks like you need "sudo" in front of those git clone commands (sudo does indeed resolve these issues).

Screenshot 2021-03-27 141122

And this directory seems to already exist on the pre-built image: Screenshot 2021-03-27 141258

Also, if you happen to have a monitor attached while doing the bootup, you'll note this error: PXL_20210327_205646826

YaBr11 commented 3 years ago

Same for me. Another issue I detected is that PyTorch is not working inside the JupyterLab environment. When I try to import torch or torchvision I get: OSError: /usr/lib/aarch64-linux-gnu/libgomp.so.1: cannot allocate memory in static TLS block In the terminal using the python3 environment, it works fine on the other hand.

TCIII commented 3 years ago

@zlite,

You were able to get the USB connection to work as I couldn't and had to use the Ubuntu Desktop to setup the WiFi connection?

I originally reported the issues you have experienced and more on the NVIDIA Developers Nano Forum on May 25th.

They acknowledged the fact that there are issues with the latest release, however I don't think that they are in any rush to fix them as the it has been two years since the Jetracer software image was updated until now.

TCIII commented 3 years ago

Observation with the latest Jetracer release for the Nano 4GB B01:

If you successfully get the latest Jetracer release installed you will find that you cannot save any of the Jetracer Notebooks as permission is denied.

TCIII commented 3 years ago

Here is a workaround for the installation of the latest Jetracer release that has worked for me:

1) Burn JP 4.5.1 onto a micro SD card and complete the setup using the Ubuntu Desktop. 2) From a SSH terminal run sudo apt-get update -y followed by sudo apt-get upgrade -y and then a reboot. 3) From a SSH terminal clone and install Jetcard using Option2 Step 2. 4) From a SSH terminal clone and install Jetcam and Jetracer (don't use sudo with clone) per Step 5 in the Jetracer installation guide, but don't clone and install torch2trt as the Jetcard installation has already installed it. 5) You may get a message that Jupyter needs to build/install additional software when you first run the "basic_motion" notebook, however the install will fail as permission is denied. So note the build command and use a SSH terminal prefacing the command with "sudo" to get the build to complete.

Following the above steps I found that I could now save the notebooks without a denial of permission.

Comments?

TCIII commented 3 years ago

Same for me. Another issue I detected is that PyTorch is not working inside the JupyterLab environment. When I try to import torch or torchvision I get: OSError: /usr/lib/aarch64-linux-gnu/libgomp.so.1: cannot allocate memory in static TLS block In the terminal using the python3 environment, it works fine on the other hand.

When I installed the latest release of Jetracer from scratch (see above) torch and torchvision imported without issue when I ran the Model section of the interactive_regression.ipynb notebook.

TCIII commented 3 years ago

@zlite,

FYI

NVIDIA has addressed the issues I pointed out and packaged a new image, partly taking my suggested solutions.

zlite commented 3 years ago

Excellent. Thx!

On Wed, Jun 9, 2021, 6:45 AM TCIII @.***> wrote:

@zlite https://github.com/zlite,

FYI

NVIDIA has addressed the issues I pointed out and packaged a new image, partly taking my suggested solutions.

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/NVIDIA-AI-IOT/jetracer/issues/99#issuecomment-857707590, or unsubscribe https://github.com/notifications/unsubscribe-auth/AAAENFYTSTLQE75M5UP6VXTTR5V77ANCNFSM4Z5JXCTQ .