TX2 Wifi problems

TL;DR : The Jetson’s wifi cuts randomly and messes up with our robots (using ROS)

Hi guys,
I’ve come to here to seek for ya’ll knowledge and help.

We’ve ordered 6 Jetsons for our company and received them, all working fine. I’ve installed the latest software with CUDA9 with jetpack on it and have been cloning the jetsons between each other using This: Jetson/TX2 Cloning - eLinux.org

Everything worked out fine and the robots my company makes use a jetson, a lidar (sick-tim) and wheels.
The software of the robot runs primarily on ROS, and, as some of you may know, ROS works with nodes connected through an ip adress (ROS_IP and ROS_MASTER_URI.).
When the wifi cuts, everything stops (The robot, literally) because the IP adress that ROS is set to is the wifi adress. We’ve got around that probem by simply disabling the wifi and/or setting the adresses to 127.0.0.1 but we need the robot to be connected to internet and we need to have access to the ROSIP to use rviz and ssh in the robot for ease of use.

The 1 million dollar question is: Anyone have that issue ? How did you solve it ? Is there anyway to make it so the wifi doesnt cut out randomly ?

Jules for unsupervised.ai

Seluj78,

It would be better to share more detail and your kernel log when disconnection happens if possible. That is, please share your wifi setting. How do you connect your devices? Wifi connection between each device as p2p? AP mode? or through external ethernet connection?

Hi,

What kind of kernel log do you want? (what files ?)

As for network details, It’s a wifi grid protected with WPA-WPA2 enterprise, Authentification using protected EAP, No CA certificate, automatic PEAP version, inner auth using GTC.

The robot itself is only connected to the wifi. The computer is also connected through wifi and ssh on the jetson. The jetson ony has the two antennas setup on the board

I am not sure if ROS is able to read the kernel driver log. In L4T, command “dmesg” would show the log from kernel driver.

Here’s the raw log from /var/log/kern.log. Everytime the wifi cut, the log was displaying something. I will sort through it now and put some comments and send you a pastebin ASAP

http://termbin.com/m7y7

I remove all the unecessarey previous boot from the log.

The comments will be signified with // So you can easily search for them. I will do my best in this edited log file to SEPARATE (not remove) parts of the log that i understand to make it easier to understand for you

Mind you, I also modified the kernel a little bit by doing these steps: Kernel modification for jetson - Pastebin.com

Somehow I cannot access your link. Could you share it in comment with code block?

Removed, too long and cut. See below for file

It looks to big to be accepted in full
How can you not access this link ?

https://pastebin.com/raw/BrkPxMQc

Hi Seluj78,

There is a function button looks like “</>” in the reply window. Please use it to wrap your log.

That’s what I used, it’s called


and it doesn’t work with this log as it is too big and that the forum cuts the text

You can attach a log file in your previous comment. There should be a button to add attachment.

Right, I didn’t think about that. Here ya go
tx2_wifi_cut_kernel_log_commented.txt (347 KB)

Thanks for sharing. Could you also see such error when you put your module on a nvidia development kit?

It would be better for us to debug if we could reproduce such issue on our board.

I don’t really see what you mean by module :/

I don’t think you get our setup. We’re making robots with Jetson TX2s. So we are using the dev kit. I’ve flashed the jetson with jetpack 3.2. Cuda 9 is installed and working correctly. The kernel has been changed such as so we can get an xbox controller working (using GitHub - jetsonhacks/buildJetsonTX2Kernel: Build the NVIDIA Jetson TX2 Kernel on the device itself). There is no such module and you should be able to reprocude the problem If i send you the 30Gb image? either it’s software or hardware. If it’s hardware, we surely can send you back a jetson that is not working

Ok… so this setup is already on devkit. Sorry for this. Then could you share a full steps to reproduce your disconnection issue?

We had some customers having wifi connection problem. However, their issue is not able to reproduce on our device, so we failed to debug from our side.

Moreover, because I cannot access your pastebin, no patch in previous comment is available to me. Could you share as attachment?

Thanks!

Here is the patch I applied on the jetson kernel. These are the exact steps I followed after flashing the jetson with jetpack

Once I had done that, I changed the user to jetson-uai and modified some system settings as I wanted the image to be ready to be flashed to other robots.

Once again, If it turns out to be necessary, I am willing to ship you one of our jetsons or send you the 30Gb file (img)
kernel_modification_for_jetson.txt (554 Bytes)

Ad for the full steps, When we take a new jetson out of a box (we had 5, so this is done on all the jetsons we have), they are flashed with what we call SV1, which stands for Software Version 1 which corresponds to the image that is already on the other jetsons. This image has automatic login, jetson-uai user, a zsh-bash rc ready for ros and node, a catkin workspace is setup and everything should be ready to be installed on a robot.

That’s pretty much it. I’ve linked the readme that i’ve made for other devs to setup their workspace to work in our company. This should be opened with markdown (preferably in github.)
readme_installing_jetson_or_workspace.txt (15.5 KB)

Seluj78,

I just checked your kernel change, which adds CONFIG_USB_ACM. This should not affect wifi usage.

In fact, what I wanted to ask is if you could use pure jetpack3.2 BSP(without ROS) to reproduce your issue under your network environment setup.

If yes, please share all the detail, for example, what AP configuration you are using.
If no, maybe the issue comes from the package you installed onto Jetson and we may not share much help here.

I guess this would be the second case because you mentioned “From here, I start moving the robot around using the joystick. Suddenly, the wifi cuts and the robot stops”.

Anyway, please give it a try. Thanks.