Skip to main content
You’ve installed the CLI and logged in. This page takes the shortest path to a robot you can program: a cloud simulator session. No hardware, about five minutes, and you end up inside a live workspace where real robot code runs.

​
1. Start a sim session

​
2. Pick a robot and a scene

The wizard opens with a New Sim Session card that fills in as you pick. First, the robot — tabs group them by form factor (←/→ to switch tabs):
Pick Franka Panda. It runs on one simulator, so the wizard jumps straight to scenes (robots available on more than one simulator get a simulator step in between):
Pick tabletop_objs, then review. The card now shows every choice plus a generated session name ({robot}-{scene}-{date}-{time}) — press enter to launch. If you want to see exactly what you’re deploying, press e at any step: the session config opens as YAML in your editor, and your edits carry into the launch.

​
3. Watch it boot

Your cluster allocated a GPU, booted Isaac Sim next to a browser IDE, and opened the workspace. Two things to notice in the shell:
  • The statusline at the bottom now shows ● with your session name — green means this shell has a live session attached.
  • The UI link is durable. Close the tab, reopen the link — same session.

​
4. It’s alive

The workspace is VS Code in your browser, running on the session pod. On first open it asks “Do you trust the authors of the files in this folder?” — this is your own session workspace, so choose Yes, I trust the authors. The session workspace: the seeded welcome notebook open in the browser IDE, with the first cell connected to the simulated Franka arm What you’re looking at:
  • sessions/<session-name>/notebook.ipynb — a welcome notebook seeded for the robot you picked, already open, kernel ready.
  • Sample Notebooks (bottom of the Explorer) — ready-made notebooks for common tasks; open any of them to explore.
  • Sim Streaming and Visualization panels — the live scene view and camera/sensor streams for this session.

​
5. Do one real thing

Run the notebook’s first cell (shift+enter):
It goes green in about a second. That’s not a mock — your code is now holding a connection to the Franka in the running scene. The second cell connects the other half of GRID, cloud models:
get_info() lists the models actually deployed in your cluster, and client.help("<model>") shows usage for any of them — see the model catalog and the grid-cortex-client reference for what you can call.

​
6. Housekeeping

Back in the shell (or any new terminal), your session shows up for the whole org:
Selecting the active row reconnects — it reopens the workspace UI for that session. When you’re done, stop it (release the GPU — sessions don’t stop themselves):
With no argument, session stop lets you pick from the active sessions and confirms before stopping. One timing note: the GPU takes ~20 seconds to return to the pool, so if an immediate restart reports no GPU available, wait a moment and retry. If anything misbehaves, three commands have your back:
  • error — replay the last failure in full detail.
  • diagnostics — bundle redacted logs + config into a tarball for support.
  • debug on — verbose logging with inline stack traces for this shell.

​
Where next

You have a live robot and a workspace — the next two pages are the same staircase with more weight on it:
  • Write and save real robot code: Your first skill
  • Point it at hardware: robot add onboards your robot, then session start robot opens this same workspace wired to the physical machine — see Deployment
  • Every command in the shell: the CLI reference

Related topics

Get startedYour first skillThe Intelligence Grid for Physical AICommon FAQsMDP Config