Skip to main content

Setup Guide: Cloud-Native Development

Learning Objectives

  • Launch GPU-enabled VM instances on AWS EC2 or Azure
  • Install ROS 2 and Isaac Sim on cloud infrastructure
  • Configure remote desktop access for GUI applications
  • Estimate cloud costs and optimize usage for the course

Before You Begin

⏱️ Estimated Time: 120 minutes

Prerequisites: None - this chapter is suitable for beginners.

Estimated Time: 2 hours | Difficulty: Intermediate

Overview​

The Cloud-Native setup uses rented GPU instances from AWS or Azure. This is ideal if you:

  • ✅ Don't have an RTX GPU
  • ✅ Don't want to invest in hardware upfront ($0 initial cost)
  • ✅ Need flexibility to scale resources up/down
  • ✅ Want to access powerful GPUs on-demand

Trade-offs​

ProsCons
No hardware investmentOngoing monthly costs ($50-$200)
Access latest GPUs (A100, V100)Network latency for remote desktop
Scalable (add GPUs as needed)Data transfer costs
Pre-configured images availableRequires internet connection

Cost Estimate​

AWS EC2 (g4dn.xlarge - RTX T4 equivalent)​

ItemCost
Instance (g4dn.xlarge: 4 vCPU, 16GB RAM, T4 GPU)$0.526/hour
Storage (100GB EBS SSD)$10/month
Estimated Usage (40 hours/month for course)~$21/month
Data Transfer (100GB/month)$9/month
Total (3 months)~$90-$120

Azure NC6s_v3 (V100 GPU)​

ItemCost
Instance (NC6s_v3: 6 vCPU, 112GB RAM, V100)$3.06/hour
Storage (256GB Premium SSD)$40/month
Estimated Usage (20 hours/month - more powerful)~$61/month
Total (3 months)~$180-$250

Recommendation: Start with AWS g4dn.xlarge ($90 for course), upgrade to V100 if needed for Module 3-4.

Option A: AWS EC2 Setup​

Step 1: Create AWS Account​

  1. Visit: https://aws.amazon.com/
  2. Sign up (requires credit card, $1 verification charge)
  3. Navigate to EC2 Dashboard

Step 2: Launch GPU Instance​

# Option A: AWS Console (GUI)
1. EC2 Dashboard → Launch Instance
2. Name: "Physical-AI-Workstation"
3. AMI: Ubuntu Server 22.04 LTS (HVM)
4. Instance Type: g4dn.xlarge (4 vCPU, 16GB, T4 GPU)
5. Key Pair: Create new → Download .pem file
6. Storage: 100GB gp3 SSD
7. Security Group:
   - SSH (22) - Your IP only
   - RDP (3389) - Your IP (for remote desktop)
8. Launch Instance

# Option B: AWS CLI
aws ec2 run-instances \
  --image-id ami-0c7217cdde317cfec \
  --instance-type g4dn.xlarge \
  --key-name physical-ai-key \
  --security-group-ids sg-xxxxx \
  --subnet-id subnet-xxxxx \
  --block-device-mappings DeviceName=/dev/sda1,Ebs={VolumeSize=100}

Step 3: Connect to Instance​

# SSH into instance
chmod 400 physical-ai-key.pem
ssh -i physical-ai-key.pem ubuntu@<EC2-PUBLIC-IP>

# Update system
sudo apt update && sudo apt upgrade -y

Step 4: Install NVIDIA Drivers​

# Install NVIDIA drivers (already pre-installed on g4dn)
nvidia-smi  # Should show T4 GPU

# If not installed:
sudo apt install -y nvidia-driver-535 nvidia-cuda-toolkit
sudo reboot

Step 5: Install ROS 2 Humble​

# Follow same steps as Workstation Setup
sudo curl -sSL https://raw.githubusercontent.com/ros/rosdistro/master/ros.key \
  -o /usr/share/keyrings/ros-archive-keyring.gpg
echo "deb [arch=$(dpkg --print-architecture) signed-by=/usr/share/keyrings/ros-archive-keyring.gpg] \
  http://packages.ros.org/ros2/ubuntu jammy main" | \
  sudo tee /etc/apt/sources.list.d/ros2.list

sudo apt update
sudo apt install ros-humble-desktop -y
echo "source /opt/ros/humble/setup.bash" >> ~/.bashrc
source ~/.bashrc

Step 6: Setup Remote Desktop (VNC)​

# Install desktop environment
sudo apt install ubuntu-desktop -y
sudo apt install tigervnc-standalone-server tigervnc-common -y

# Configure VNC password
vncpasswd  # Set password (e.g., "robotics123")

# Start VNC server
vncserver :1 -geometry 1920x1080 -depth 24

# Connect from your local machine:
# ssh -L 5901:localhost:5901 -i physical-ai-key.pem ubuntu@<EC2-IP>
# Then use VNC client (RealVNC, TigerVNC) to connect to localhost:5901

Option B: Azure Virtual Machine​

Step 1: Create Azure Account​

  1. Visit: https://azure.microsoft.com/
  2. Sign up ($200 free credits for new users)
  3. Navigate to Virtual Machines

Step 2: Create GPU VM​

  1. Create Resource → Virtual Machine
  2. Basics:
    • VM name: "physical-ai-vm"
    • Region: East US (cheaper GPUs)
    • Image: Ubuntu Server 22.04 LTS
    • Size: NC6s_v3 (V100, 6 vCPU, 112GB)
  3. Administrator Account: SSH public key
  4. Disks: 256GB Premium SSD
  5. Networking: Allow SSH (22), RDP (3389)
  6. Review + Create

Step 3: Connect and Configure​

# SSH to Azure VM
ssh azureuser@<VM-PUBLIC-IP>

# Install NVIDIA drivers (Azure CUDA image includes drivers)
nvidia-smi  # Should show V100

# Install ROS 2 (same as AWS steps above)

Cost Optimization Tips​

Stop Instance When Not Using​

# AWS: Stop instance (preserves data, no compute charges)
aws ec2 stop-instances --instance-ids i-xxxxx

# AWS: Start instance when needed
aws ec2 start-instances --instance-ids i-xxxxx

# Savings: ~70% reduction in costs by stopping when idle

Use Spot Instances (Advanced)​

  • Spot Instances: 60-90% cheaper than on-demand
  • Risk: Can be terminated if AWS needs capacity
  • Use Case: Training ML models (Module 3-4), not for 24/7 dev
# Launch spot instance
aws ec2 request-spot-instances \
  --spot-price "0.20" \
  --instance-count 1 \
  --type "one-time" \
  --launch-specification file://spot-spec.json

Snapshot Your Setup​

# AWS: Create AMI snapshot after setup
aws ec2 create-image \
  --instance-id i-xxxxx \
  --name "physical-ai-ros2-setup" \
  --description "ROS 2 + Isaac Sim configured"

# Future launches: Use your custom AMI instead of re-installing

Verification Checklist​

  • Instance Running: AWS/Azure console shows "Running"
  • SSH Access: ssh connection succeeds
  • GPU: nvidia-smi shows GPU model and driver
  • ROS 2: ros2 run demo_nodes_cpp talker works
  • Remote Desktop: VNC connection displays Ubuntu desktop
  • Cost Monitoring: Billing alerts configured (AWS CloudWatch/Azure Cost Management)

Example: Running Isaac Sim on Cloud​

# SSH with X11 forwarding for GUI (slower than VNC)
ssh -X -i key.pem ubuntu@<IP>

# OR use VNC for better performance
vncserver :1 -geometry 1920x1080

# Install Isaac Sim (via Omniverse Launcher)
# Note: Cloud instances may have limited internet bandwidth
# Download takes 30-60 minutes for ~50GB

Limitations​

LimitationImpactMitigation
Network LatencyRemote desktop lag (200-500ms)Use VNC, optimize window size
Ongoing Costs$50-$200/monthStop instances when idle, use spot
Data TransferCosts for uploading datasetsUse S3/Azure Blob for storage
Setup TimeInitial configurationCreate AMI/snapshot for future use

Cost Comparison Table​

SetupInitialMonthly (3 months)Total (Course)Best For
Workstation$600-$2,500$0$600-$2,500Long-term investment
Edge Kit$550$0$550Physical robots
Cloud (AWS g4dn)$0$30-$40$90-$120No hardware, budget-friendly
Cloud (Azure V100)$0$60-$80$180-$240Powerful GPUs, RL training

Next Steps​

  1. Setup Billing Alerts: Configure AWS/Azure to alert at $50, $100
  2. Create Snapshot: Save configured instance as custom image
  3. Test GPU: Run nvidia-smi and verify CUDA toolkit
  4. Start Learning: Begin Module 1: ROS 2

Questions? Refer to the Glossary for terminology or consult cloud provider documentation.

Alternatives: