Preparing an RTX 4090 Robotics Workstation
Before relying on the robot as my only development machine, I prepared a separate Ubuntu workstation around an NVIDIA RTX 4090.
The idea is important to the architecture I want to test: the Jetson is physically attached to the robot, but that does not mean every development, visualization, simulation, or AI workload should run there.
I set up a dedicated Ubuntu environment and started bringing up the ROS 2 / NVIDIA robotics toolchain, including ROS 2 Jazzy and Isaac-related navigation components.
The longer-term architecture I want to explore is:
M3 Pro / Jetson
sensors, motors, arm, robot-side ROS 2
|
| ROS 2 / DDS
v
Ubuntu workstation / RTX 4090
RViz, development, simulation,
debugging, heavier GPU workloads
The first goal is much simpler than distributing AI models: prove that the workstation can reliably participate in the same ROS 2 system and consume the robot's topics directly.
If it works, I should be able to run tools such as RViz locally on the RTX 4090 workstation while receiving TF, odometry, LiDAR, camera, and other robot state from the Jetson.
That would be better than treating the robot as a desktop PC that I happen to view through VNC.
Why this matters beyond convenience
This is also my first concrete edge-compute question:
What belongs on the robot, and what belongs on a nearby larger GPU?
Eventually that boundary may involve latency, network failure, bandwidth, privacy, safety, and inference cost. For now, the experiment starts with something much less glamorous: can two ROS 2 machines communicate reliably enough for everyday development?