1.2 Installation and Setup
##Introduction
This section guides you through installing NVIDIA Isaac Sim 2023.1 LTS on your local machine or cloud GPU instance. We target the 2023.1 LTS (Long-Term Support) release for stability and compatibility with RTX 2060+ GPUs. Migration notes for Isaac Sim 2024.x users are provided at the end.
System Requirements
Minimum Hardware
| Component | Minimum | Recommended |
|---|---|---|
| GPU | NVIDIA RTX 2060 (6GB VRAM) | NVIDIA RTX 3080 (10GB+ VRAM) |
| RAM | 16GB | 32GB (for large scenes or RL training) |
| CPU | Intel i7 / AMD Ryzen 7 (6 cores) | Intel i9 / AMD Ryzen 9 (8+ cores) |
| Storage | 50GB free SSD space | 100GB+ SSD (for assets and logs) |
| OS | Ubuntu 22.04 LTS or Windows 10 | Ubuntu 22.04 LTS (best performance) |
Software Prerequisites
- NVIDIA Driver: Version 525+ (check with
nvidia-smi) - Python: 3.10+ (Isaac Sim 2023.1 uses Python 3.10)
Installation Methods
Choose one of the following installation methods:
- Local Installation (Ubuntu/Windows) - Full control, best performance
- Cloud GPU (AWS G4/G5) - No local GPU required, pay-per-use
- NVIDIA LaunchPad (Free Trial) - Free exploration, limited hours
Method 1: Local Installation (Ubuntu 22.04)
Step 1: Install NVIDIA Driver
# Check current driver version
nvidia-smi
# If version < 525, install updated driver
sudo ubuntu-drivers list
sudo ubuntu-drivers install nvidia-driver-525
# Reboot to load new driver
sudo reboot
# Verify driver installation
nvidia-smi
# Expected output: Driver Version: 525.xx or higher
Screenshot Placeholder: nvidia-smi-driver-check.png - Terminal showing NVIDIA driver 525+ with GPU info
Step 2: Download and Install Omniverse Launcher
- Visit NVIDIA Omniverse
- Click "Download Omniverse Launcher" (requires free NVIDIA account)
- Download for Ubuntu:
omniverse-launcher-linux.AppImage
Screenshot Placeholder: omniverse-launcher-download-page.png - Omniverse website download page
- Make executable and run:
chmod +x omniverse-launcher-linux.AppImage
./omniverse-launcher-linux.AppImage
- Sign in with your NVIDIA account
- Accept terms and complete installation
Screenshot Placeholder: omniverse-launcher-signin.png - Launcher sign-in screen
Step 3: Install Isaac Sim 2023.1 LTS via Launcher
- In Omniverse Launcher, navigate to "Exchange" tab
- Search for "Isaac Sim"
- Find "Isaac Sim 2023.1.1" (LTS release)
- Click "Install" button
- Choose installation directory (default:
~/.local/share/ov/pkg/isaac_sim-2023.1.1) - Wait for download (~20GB, 15-60 minutes depending on connection)
Screenshot Placeholder: isaac-sim-installation-progress.png - Launcher showing Isaac Sim download progress
Step 4: Configure Nucleus Server (Local)
Nucleus server is optional for single-user setups, but useful for asset management.
Option A: Local Nucleus (Lightweight)
- In Omniverse Launcher, go to "Nucleus" tab
- Click "Add Local Nucleus Service"
- Set administrator password
- Click "Create"
- Nucleus starts at
localhost:3030orlocalhost:3009
Option B: Skip Nucleus (Use Local Files)
You can work entirely with local USD files (file:///home/user/my_scene.usd) without Nucleus. This is fine for learning and single-user projects.
Screenshot Placeholder: nucleus-local-setup.png - Nucleus configuration in Launcher
Step 5: Launch Isaac Sim and Verify Installation
# Navigate to Isaac Sim installation directory
cd ~/.local/share/ov/pkg/isaac_sim-2023.1.1
# Launch Isaac Sim GUI
./isaac-sim.sh
# Alternative: Launch headless (no GUI, for testing)
./isaac-sim.sh --headless
First Launch: Isaac Sim may take 2-5 minutes to initialize shaders and cache. Subsequent launches are faster (~30 seconds).
Screenshot Placeholder: isaac-sim-first-launch.png - Isaac Sim main window with default scene
Step 6: Load Sample Scene for Verification
- In Isaac Sim window, go to File → Open
- Navigate to
Isaac/Samples/Robots/(in Nucleus browser or local samples) - Open
simple_robot.usdorfranka.usd - Click Play button (▶️) to start simulation
- Verify physics simulation runs smoothly (robot responds to gravity)
- Check viewport FPS (should be 30-60 FPS for simple scenes)
Measure Real-Time Factor (RTF):
- Run simulation for 60 seconds (sim time)
- Measure wall-clock time elapsed
- Calculate RTF = 60 / wall_clock_time
- Target: RTF ≥1.0 (e.g., 60 sim seconds in 30 wall-clock seconds → RTF = 2.0)
Screenshot Placeholder: isaac-sim-sample-robot-simulation.png - Franka robot simulation running
Step 7: Verify Python API Access
# Isaac Sim includes standalone Python interpreter
./python.sh
# In Python shell:
>>> import omni.isaac.core
>>> print(omni.isaac.core.__version__)
# Expected: Version string like "2023.1.1"
>>> exit()
Installation Verification Checklist:
- ✅ NVIDIA driver 525+ installed
- ✅ Omniverse Launcher running
- ✅ Isaac Sim 2023.1.1 installed and launches
- ✅ Sample robot simulates with RTF ≥1.0
- ✅ Python API accessible
If all checks pass, your local installation is complete!
Method 2: Cloud GPU Installation (AWS EC2)
If you don't have a local NVIDIA GPU, AWS EC2 provides on-demand and spot instances with RTX-class GPUs.
Step 1: Choose Instance Type
| Instance Type | GPU | VRAM | vCPUs | RAM | On-Demand Price | Spot Price (est.) |
|---|---|---|---|---|---|---|
| g4dn.xlarge | NVIDIA T4 | 16GB | 4 | 16GB | ~$0.526/hr | ~$0.15/hr |
| g4dn.2xlarge | NVIDIA T4 | 16GB | 8 | 32GB | ~$0.752/hr | ~$0.22/hr |
| g5.xlarge | NVIDIA A10G | 24GB | 4 | 16GB | ~$1.006/hr | ~$0.30/hr |
| g5.2xlarge | NVIDIA A10G | 24GB | 8 | 32GB | ~$1.212/hr | ~$0.36/hr |
Recommendation: Start with g4dn.xlarge for learning (sufficient for single robot RTF ≥1.0). Upgrade to g5.xlarge for RL training with 2048 parallel environments.
Step 2: Launch EC2 Instance
# Using AWS CLI (or use AWS Console web interface)
# 1. Create key pair for SSH access
aws ec2 create-key-pair --key-name isaac-sim-key --query 'KeyMaterial' --output text > isaac-sim-key.pem
chmod 400 isaac-sim-key.pem
# 2. Launch instance with Ubuntu 22.04 Deep Learning AMI
# (Deep Learning AMI includes NVIDIA drivers pre-installed)
aws ec2 run-instances \
--image-id ami-0c55b159cbfafe1f0 \ # Ubuntu 22.04 DL AMI (us-east-1, check for your region)
--instance-type g4dn.xlarge \
--key-name isaac-sim-key \
--security-group-ids sg-xxxxxx \ # Configure security group with SSH (port 22) access
--block-device-mappings '[{"DeviceName":"/dev/sda1","Ebs":{"VolumeSize":100}}]' # 100GB storage
# 3. Get instance public IP
aws ec2 describe-instances --query 'Reservations[0].Instances[0].PublicIpAddress'
# 4. SSH into instance
ssh -i isaac-sim-key.pem ubuntu@<PUBLIC_IP>
Cost Optimization: Use Spot Instances for 70% discount (terminates if AWS needs capacity back, but you can save checkpoint and resume). For training jobs:
aws ec2 request-spot-instances \
--spot-price "0.30" \ # Max price you're willing to pay
--instance-count 1 \
--type "one-time" \
--launch-specification file://spot-spec.json
Step 3: Install Isaac Sim on Cloud Instance
# On EC2 instance (via SSH):
# 1. Verify GPU
nvidia-smi
# Should show T4 or A10G GPU
# 2. Download Omniverse Launcher (headless)
wget https://install.launcher.omniverse.nvidia.com/installers/omniverse-launcher-linux.AppImage
# 3. For headless installations, use Isaac Sim Docker container (recommended)
# Pull Isaac Sim Docker image from NVIDIA NGC
docker pull nvcr.io/nvidia/isaac-sim:2023.1.1
# 4. Run Isaac Sim in Docker with GPU passthrough
docker run --gpus all -it \
-v ~/isaac_sim_data:/workspace/data \ # Mount local directory for data persistence
nvcr.io/nvidia/isaac-sim:2023.1.1 \
/bin/bash
# Inside Docker container:
./isaac-sim.sh --headless # Run headless Isaac Sim
X11 Forwarding for GUI (Optional): If you want to see Isaac Sim GUI from cloud instance:
# On your local machine, SSH with X11 forwarding:
ssh -X -i isaac-sim-key.pem ubuntu@<PUBLIC_IP>
# On EC2 instance, install X11 libraries:
sudo apt-get update
sudo apt-get install -y xorg
# Launch Isaac Sim with GUI (rendered on cloud, displayed locally):
./isaac-sim.sh
Warning: X11 forwarding over SSH is slow for interactive use. Use headless mode for training/testing, and download videos/screenshots for review.
Step 4: Stop Instance When Not in Use
# Stop instance (preserves storage, stops compute charges)
aws ec2 stop-instances --instance-ids i-xxxxxxxxx
# Start instance later
aws ec2 start-instances --instance-ids i-xxxxxxxxx
# Terminate instance (deletes storage, stops all charges)
aws ec2 terminate-instances --instance-ids i-xxxxxxxxx
Cost Management:
- Stop instances when not actively training/testing
- Use Spot instances for long-running experiments (can save checkpoints and resume if interrupted)
- Set billing alarms in AWS Console to avoid unexpected charges
Method 3: NVIDIA LaunchPad (Free Trial)
NVIDIA LaunchPad offers free GPU access for limited hours per month, perfect for exploring Isaac Sim before committing to paid options.
- Visit NVIDIA LaunchPad
- Sign up for free account
- Request access to "Isaac Sim" lab
- Once approved, launch cloud instance (pre-configured with Isaac Sim)
- Access via browser (NVIDIA provides streaming GPU desktop)
Limitations:
- 8-16 hours/month free tier (varies by promotion)
- Cannot save data persistently (download results before session ends)
- Shared GPU resources (performance may vary)
Best For: Trying Isaac Sim before local installation, completing Chapter 1 tutorials, testing compatibility.
Troubleshooting
Issue 1: NVIDIA Driver Version Mismatch
Symptom: Isaac Sim fails to launch with error "NVIDIA driver version 525 or higher required"
Cause: Outdated NVIDIA driver (< 525)
Solution:
# Ubuntu:
sudo ubuntu-drivers install nvidia-driver-525
sudo reboot
# Verify:
nvidia-smi
Issue 2: Isaac Sim Installation Errors
Symptom: Omniverse Launcher shows "Installation failed" or download stalls
Cause: Insufficient disk space (< 50GB free), network issues, or corrupted download
Solution:
# Check disk space:
df -h
# If low, free up space or choose different installation directory
# Clear Omniverse cache and retry:
rm -rf ~/.local/share/ov/Cache
rm -rf ~/.local/share/ov/pkg/isaac_sim-* # Remove partial install
# Relaunch Omniverse Launcher and reinstall Isaac Sim
Issue 3: USD Import Failures
Symptom: URDF files fail to import with error "URDF Importer extension not found"
Cause: URDF Importer extension not enabled
Solution:
- In Isaac Sim, go to Window → Extensions
- Search for "URDF Importer"
- Toggle switch to ON
- Retry URDF import
Issue 4: ROS 2 Bridge Not Publishing Topics
Symptom: ros2 topic list shows no topics from Isaac Sim
Cause: ROS 2 bridge extension not enabled, or ROS 2 not installed
Solution:
- Enable "ROS2 Bridge" extension in Isaac Sim (Window → Extensions → Search "ROS2 Bridge" → Toggle ON)
- Verify ROS 2 Humble installed on your system:
source /opt/ros/humble/setup.bash
ros2 topic list
- Ensure Isaac Sim is publishing to ROS 2 (check extension settings for topic names)
Issue 5: RTX Rendering Slow / Low FPS
Symptom: Viewport FPS < 10, simulation feels sluggish
Cause: GPU cannot handle real-time ray tracing at high resolution
Solution:
- Reduce viewport resolution: Edit → Preferences → Rendering → Set resolution to 1280x720 or lower
- Disable real-time ray tracing: Change render mode from "RTX - Interactive" to "Iray - Iterative" (or use rasterization mode for faster preview)
- Use headless mode (
./isaac-sim.sh --headless) when rendering is not needed (maximizes RTF)
Issue 6: PhysX Simulation Instability (Robot Explodes/Jitters)
Symptom: Robot limbs fly apart, jitter violently, or simulation crashes
Cause: PhysX solver parameters too low for complex articulated robot
Solution:
- Increase solver iterations: Physics → Settings → Set "Position Iteration Count" to 16-32 (default is 4)
- Reduce simulation timestep: Set "Physics Update Rate" to 120 Hz (from default 60 Hz)
- Increase contact stiffness: Adjust material physics properties for robot links
- See Section 1-4 for detailed PhysX tuning guidance
Migration Notes for Isaac Sim 2024.x Users
If you're using Isaac Sim 2024.x (latest release) instead of 2023.1 LTS:
API Changes:
- Some
omni.isaac.coreAPI methods renamed or deprecated (check release notes) - PhysX 5.3 → 5.4 parameter changes (contact defaults adjusted)
Compatibility:
- Code examples in this chapter target 2023.1 API
- Most examples work with 2024.x with minor adjustments
- Look for comments like
# Isaac Sim 2024.x: Use new_method() insteadin example code
Recommendation: Stick with 2023.1 LTS for this course to match tested examples. Migrate to 2024.x after completing Module 3.
Verification Checklist
Before proceeding to Section 1-3, ensure:
- Isaac Sim 2023.1.1 installed and launches successfully
- Sample robot simulation runs with RTF ≥1.0
- Python API accessible (
import omni.isaac.coreworks) - NVIDIA driver version ≥525
- (Optional) Nucleus server configured or local USD file workflow confirmed
- (If cloud) SSH access to EC2 instance working, GPU verified with
nvidia-smi
Next: 1.3 URDF to USD Conversion → - Convert your Module 1 humanoid robot to Isaac Sim USD format!