Run robot simulations with Isaac Lab

Isaac Lab is NVIDIA's open-source framework for robot learning. On Olares, it runs as a terminal-based app where you start demos, training tasks, or the bundled Isaac Sim simulator from the command line.

Use this guide to install Isaac Lab, run Isaac Lab workloads, start Isaac Sim, connect to WebRTC streaming, and manage background processes.

Learning objectives

In this guide, you will learn how to:

• Install Isaac Lab on Olares.
• Prepare the Isaac Lab terminal and get the WebRTC endpoint address.
• Run Isaac Lab demos and reinforcement learning tasks.
• Start the bundled Isaac Sim simulator.
• Connect to simulations with the NVIDIA WebRTC Streaming Client.
• Manage background workloads, GPU processes, and runtime logs.

Prerequisites

Before you begin, make sure:

• Olares is running on a machine with an NVIDIA GPU.
• The Olares host uses an AMD64 architecture if you want to use WebRTC streaming. ARM64 hosts (such as DGX Spark) support only non-streaming workloads.
• NVIDIA's WebRTC Streaming Client is installed on your local computer.

Host vs. local computer architecture

The AMD64 requirement applies to the Olares host where Isaac Lab runs, not to the local computer used for streaming.

For example, if Isaac Lab runs on an Olares One with an Intel CPU and NVIDIA GPU, the Olares host is AMD64. Your local computer can use a different architecture, but it must be able to run NVIDIA's WebRTC Streaming Client.

Recommended GPU mode

Isaac Lab and Isaac Sim use significant GPU resources. For better stability, set the GPU mode to App exclusive and bind the GPU to Isaac Lab before running workloads. This helps reduce resource contention with other GPU apps.

Understand Isaac Lab and Isaac Sim

Isaac Lab and Isaac Sim serve different purposes:

Component	Description	Use it when
Isaac Lab	A robot learning framework built on Isaac Sim.	Run predefined demos, reinforcement learning tasks, or training scripts.
Isaac Sim	The underlying simulator bundled with Isaac Lab.	Start the simulator directly for scene exploration, simulation testing, or interactive use.

• To run an Isaac Lab demo or training task, use the commands in Run Isaac Lab workloads.
• To start the Isaac Sim simulator itself, use Run Isaac Sim.

Install Isaac Lab

1. Open Market and search for "IsaacLab".

   

2. Click Get, then Install, and wait for installation to complete.

Prepare Isaac Lab terminal

Whenever you open Isaac Lab or want to switch to a new task, prepare your terminal environment first.

1. Open Isaac Lab from Launchpad to access its terminal interface for running command-line scripts.

   

2. Get the public endpoint address used for WebRTC streaming.

   echo $PUBLIC_IP

   Example output:

   192.168.50.169

   Save the output value. You will need it later to connect the WebRTC Streaming Client.

3. Check the GPU status and look for existing workloads.

   nvidia-smi

   • If no workload is running, the Processes section shows No running processes found.

   

   • If the Processes section lists an Isaac Lab or Isaac Sim process, note its PID before starting a new workload. In this example, the active process uses PID 2314.

   

4. If an existing workload is running, stop it before starting a new one:

   kill -9 <PID>

   Replace <PID> with the actual process ID shown in your terminal.

5. Optional: Clear the existing log file before starting a new workload.

   > nohup.out

Run Isaac Lab workloads

Isaac Lab workloads are started from the terminal. The command you use depends on the architecture of the Olares host where Isaac Lab is installed.

Stop any existing workload before starting a new demo or task.

Commands ending with & run in the background. A message like [1] 580 means the workload has started. The numbers may differ in your terminal.

Run a demo

Run a reinforcement learning task

For more scripts, tasks, and environment configuration, refer to:

• Existing RL scripts
• Environments
• Newton physics integration

Run Isaac Sim

Isaac Lab includes the Isaac Sim simulator. Start Isaac Sim directly when you want to use the simulator itself instead of running a specific Isaac Lab demo or training script.

Stop any existing workload before starting Isaac Sim.

Terminal command reference

Isaac Lab and Isaac Sim commands started with nohup ... & run in the background. Closing the WebRTC Streaming Client does not stop the running workload.

Use these commands for quick terminal management and troubleshooting:

Task	Command
Get the WebRTC endpoint address	echo $PUBLIC_IP
Check GPU processes	nvidia-smi
Stop a background process	kill -9 <PID>
Clear old logs	> nohup.out
Follow live logs	tail -f nohup.out

Troubleshooting

WebRTC Client cannot connect

Check the following:

1. Make sure the workload is still running:

   nvidia-smi

2. Confirm that you entered the correct PUBLIC_IP value:

   echo $PUBLIC_IP

3. Make sure only one WebRTC Streaming Client is connected.

4. Check the logs for fatal errors:

   tail -f nohup.out

Look for errors such as:

Error
Traceback
CUDA out of memory
Segmentation fault
Killed

If none of these errors appear and the process is still running, wait for the logs to stabilize, then reconnect or click Reload in the WebRTC Streaming Client.

The streaming view turns gray

When loading a complex scene, the streaming view might turn gray or stop responding temporarily. Wait for the terminal logs to stabilize. Depending on the scene complexity, this may take one to several minutes.

After the logs become stable, click View > Reload in the WebRTC Streaming Client to reconnect to the stream.

Logs show warnings during startup

Warnings from Isaac Sim, Omniverse, or PyTorch may appear while a workload is starting. These warnings do not always mean the workload has failed.

If the process remains running in nvidia-smi and no fatal error appears in the logs, continue with the WebRTC connection step.

Learn more

• Isaac Lab documentation: Official Isaac Lab documentation, including tutorials, environments, reinforcement learning workflows, and API references.
• Isaac Sim documentation: NVIDIA Isaac Sim user guide.