Robotics PlatformProjects

I built RoboOps around NVIDIA Isaac Sim because I wanted to explore what fleet operations feel like when there is a real simulated robot in the loop. For this setup, Isaac Sim runs through an NVIDIA Isaac launchable on an L40S GPU.

On the platform side, RoboOps is running on AWS as a small fleet backend. I clone the project into the Brev environment, start the Isaac workflow, and bring up the simulation. Now I launch a sidecar next to Isaac Sim. This is the robot-near part of the prototype: it watches the simulation and sends telemetry back to the fleet platform.

The dashboard is not just showing mock data. It is receiving state from a simulated robot through the same kind of boundary I would expect in a real system. Here you can see Isaac Sim and the RoboOps operator UI side by side. I click Go to pose in the dashboard, and then the Nova Carter robot starts moving in the warehouse.

This is the happy path: the operator sends a command, the robot reacts, and the UI reflects what is happening. The important part is the control loop: command, motion, telemetry, and operator feedback.

But I did not just want to show the happy path, because robotic systems are rarely that clean. Here I disconnect the robot while the operator UI is still running. After a short moment, telemetry becomes stale. The UI makes that visible instead of hiding it behind logs or silent failure.

Then I hit play again in Isaac Sim. The robot reconnects, telemetry starts flowing again, and the operator view recovers. That is the kind of behavior I care about in operator tooling: not just showing that everything works, but making the system understandable when something goes wrong.

RoboOps is still a prototype, not a finished product or safety system, but it was a great exercise in thinking through the messy parts of robotics tooling: stale telemetry, reconnects, recovery, and how to make those states visible to the operator. You can find the full repo on my website, mrza.ch.