OS Testing Rig for OMX

[roboteer]

Since the recent ROBOTIS announcement of the OMX, I was reading up on the Setting Up of the OMX at its web link:

ROBOTIS

Right now the “official” setup is using a PC running on Ubuntu 24.04, but “at first look”, I did not see any constraint “against” using OMX with SBCs such as RPi5 or Orin Nano, instead. Furthermore, at present, I only have the needed critical components (XL-330s, XL-430s and OpenRB-150s) and not the actual frame parts for the OMX. Thus, I realized that, for now, I can build a compact test rig that can be a simplified physical model of the OMX that I can use to test to see if the OMX tool chain created by ROBOTIS for the OMX can be applied to an RPi5 or an Orin Nano.

The proposed OMX test rig is shown below as a top view:

1. The Leader-RB150 (on the right in this picture) is set up to use USB Power and has the Arduino sketch “usb_to_dynamixel.ino” uploaded to it so that it can pass-thru DXL packets to the 6 XL-330-M077s (operating at 5V) that I used for the Leader Pseudo-Arm. The OMX officially uses XL-330-M288s but I do not have enough of the M288s for the Leader Arm, so we’ll see if this switch-out will mess me up later with the software tool chain or not.
2. The Follower-RB150 (on the left in this picture) is also set up to use USB Power and the Arduino sketch “usb_to_dynamixel.ino” also. Please note that the 3 XL-330s (ID=14, 15, 16) are connected right into the Follower-RB-150, so they are powered at 5V and controlled directly by this RB-150. However, the XL-430s are powered at 12 V via an SMPS2Dynamixel module (see more details in the picture below)

So how did I prevent the 12V PS from “frying out” the XL-330s and RB150?

In the following picture, the connection between the XL-430 group (ID=13) and the XL-330 group (ID=14) is made using a “special” X3P cable that has its VDD wire cut-off (labeled with some white electrical tape), thus the 5V circuit (RB-150 & XL-330) and 12 V circuit (SMPS2Dynamixel & XL-430) are isolated from each other.

However, because the DATA Line of this special X3P cable is still intact, the Follower-RB150 can “see and control” both groups of servos of the Follower arm.

Using DXL-Wizard 2 tool to connect to the Leader-RB150 shows that the control of the 6 XL-330s is as it should be:

While DXL-Wizard 2 groups the XL-330s and XL-430s separately (because they have different Control Tables anyway)

So the OS Testing Rig is working out good, and I also recently have my RPi5 on Ubuntu 24.04 ready at this post:

A new and easier way to boot with various drives for RPi5

Update: 9/16/2025 I finally got to the right hardware for my NVMe setup to work properly, and I should have known better! Always use “genuine” hardware! [Rpi5-16_NVMe] The Disk Benchmark test is quite decent (SSD set to PCIe Gen 3): [DiskBenchMark]

So in the next post, I will report about my success or failure in implementing the ROBOTIS OMX tool chain to an RPi5 set up on Ubuntu 24.04, along with this OMX Test Rig.

So wish me LUCK!

[roboteer]

@YunWonho
I followed the OMX instructions from ai.robotis.com and I managed to install and checked out the Open Manipulator Docker Container OK for the physical setup shown below:

After git-cloning the Physical AI tools OK, I proceeded to the step when I needed to start the Physical AI Container.

It looked like that the RPi5 was pulling in the ai_server and ai_manager OK (for more than 500s, but then the daemon showed a runtime error about a shim task and something to do with NVIDIA? But I used an RPi5, so is this “my problem”?

But the OMY uses an RPi5 too?

Thank you in advance for any hint.

[roboteer]

I redid the installation for the 3 Docker containers: Open Manipulator, Physical AI Manager, and Physical AI Server onto my RPi5, and still the same issues with Physical AI Server.

So at present I got Physical AI Manager and Open Manipulator to work for me on the RPi5. Below is a video clip showing that I can use Teleoperation on all the Joint Servos except for No. 6 (i.e. the Gripper), as the default script sets up a very stiff actuator for No. 6. All the other ones #1 through #5 are working as intended.

In summary, for the RPi5 2, 2 out of 3 tools are working as “documented” for the OMX Test Rig.

Additionally, I did test the OMX installation procedure on a Desktop PC with NVIDIA GPU and Ubuntu and all 3 tools are installed and working OK.

The interesting result was that on the PC, the Physical AI container was pulling 49 components, while the RPi5 was trying to pull 85 components - see previous picture for the RPi5 results. So, I am wondering why??? Did my RPi5 pull the right container previously?

So the next logical step is to try this OMX Test Rig with an Orin Nano!

[roboteer]

9/22/2025
This time I tried to install the OMX Containers into an Orin Nano (Jetpack 6.2 - i.e. Ubuntu 22.04) and they installed OK (Open Manipulator then the Physical AI tools), then I rebooted the Nano to get back to a clean start. AND the NANO just would not restart, the Display just stayed Black. So the OMX containers sure did a number on the Orin Nano. So it looks like that I am going to have to wait until Jan-March 2026 when NVIDIA would actually put the ORIN NANO on Ubuntu 24.04 before trying all this again with the Orin Nano. I got the hunch that for now the Open Manipulator Container will be OK - so just Classical Robotics is OK for now. Imitation Learning has to be done with a computing platform with Actual Ubuntu 24.04 and NVIDIA GPU as “advertised” on the ai.robotis.com web site

Lastly, I got an AMD type of mini PC, a GMKTec EVO-X1 (i.e. no NVIDIA GPU) to test out the OMX containers and the final results were just like for the RPi5. “Open Manipulator” and “AI Manager” containers ran fine, but the “AI Server” has that "trim + CDI drivers’ error so it won’t run.

And once again the Motion Duplication demo from Leader to Follower ran fine (except for ID=6 just like for the RPi5 case).

[roboteer]

9/23/2025
Today, I went back to the Orin Nano, and using SDK MANAGER, I tried to install ALL available packages to the O-Nano. Most of the “run-time” packages got installed OK, but the SDK packages encountered errors during installation.

Then I went through the process of installing all the ROBOTIS Containers, and this time things went through OK. Below is a screen picture showing that I could get all 3 Docker applications running in its own terminal: Camera, Open Manipulator and even the Physical AI Server.

But now Firefox does not know how to pull up AI Server web page “http://localhost” (bottom right screen).

And Chromium used to work previously, but since morning today was acting up!

In summary, now there is a chance that we can put the OMX system on a mobile platform with O-Nano.

[roboteer]

Late 9/23/2025

Later in the day, Firefox is working with the O-Nano OK, but I could not get the USB camera to work yet with ROS2 + AI Server.

Lastly, I figured out why my O-Nano’s just would not boot up properly as mentioned in a previous post. For some reasons, the O-Nano will not boot up properly if my OS Testing Rig is connected to the O-Nano via the USB ports when it is rebooted.

[roboteer]

9/30/2026

Finally today I got everything to work properly on my Orin Nano and my OMX Test Rig.

So the instructions on the web site ai.robotis.com for the OMX does work for an Orin Nano but the key operation will be to make the JetPack 6.21’s core components and all the optional runtime packages to be installed properly on the O-Nano before installing the Robotis Docker containers.