Hardware Setup Guide

Complete guide to procuring and setting up hardware for Physical AI & Humanoid Robotics development

Overview

This guide covers three hardware tiers:

Minimal Setup ($500-1000): Simulation-only development on consumer hardware
Development Setup ($2000-5000): Basic sensors + simulation on workstation
Production Setup ($10,000-50,000): Full humanoid robot with compute platform

Choose based on your goals:

Learning/Research: Minimal or Development setup sufficient
Prototyping: Development setup with real sensors
Deployment: Production setup for real-world applications

1. Minimal Setup (Simulation-Only)

Purpose: Learn ROS 2, Isaac Sim, VLA concepts without physical hardware

Hardware Requirements

Compute Platform (Choose One)

Option A: Desktop Workstation (Recommended for performance)

CPU: AMD Ryzen 7 5800X or Intel i7-12700K (8+ cores)
GPU: NVIDIA RTX 3060 Ti (8GB VRAM minimum) or better
- RTX 3070/3080: Better for Isaac Sim (more parallel envs)
- RTX 4070/4090: Best performance, supports latest features
RAM: 32GB DDR4 (16GB minimum, 64GB recommended for large scenes)
Storage: 1TB NVMe SSD (500GB minimum)
OS: Ubuntu 22.04 LTS (dual-boot or native install recommended)

Estimated Cost: $1200-2000

Retailers:

USA: Newegg, Micro Center, Amazon
Europe: Scan.co.uk, Alternate.de
Asia: Lazada, Shopee

Option B: Gaming Laptop with NVIDIA GPU

Specs: RTX 3060 Mobile (6GB VRAM) or better, 16GB RAM, i7/Ryzen 7
Pros: Portable, all-in-one solution
Cons: Thermal throttling, harder to upgrade
Estimated Cost: $1000-1500

Models:

ASUS ROG Zephyrus G14/G15
Lenovo Legion 5 Pro
MSI Katana/Sword series

Option C: Cloud Instance (For those without local GPU)

Providers: AWS EC2 (g4dn.xlarge), Google Cloud (n1-standard-4 + T4), Paperspace
Specs: 4 vCPUs, 16GB RAM, NVIDIA T4 or better
Estimated Cost: $0.50-1.00/hour ($40-80/month for part-time use)
Pros: No upfront cost, scalable
Cons: Recurring cost, network latency for GUI

Software Installation

1. Ubuntu 22.04 LTS Setup

# Download from: https://ubuntu.com/download/desktop
# Create bootable USB: https://ubuntu.com/tutorials/create-a-usb-stick-on-windows

# After installation, update system
sudo apt update && sudo apt upgrade -y

# Install essential build tools
sudo apt install -y build-essential cmake git curl wget

2. NVIDIA Drivers + CUDA

# Install NVIDIA drivers (use recommended version)
ubuntu-drivers devices
sudo ubuntu-drivers autoinstall

# Reboot
sudo reboot

# Verify GPU detected
nvidia-smi
# Should show your GPU model and driver version

# Install CUDA 12.1+ (for Isaac Sim)
wget https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2204/x86_64/cuda-keyring_1.1-1_all.deb
sudo dpkg -i cuda-keyring_1.1-1_all.deb
sudo apt update
sudo apt install -y cuda-toolkit-12-1

# Add to PATH (~/.bashrc)
echo 'export PATH=/usr/local/cuda/bin:$PATH' >> ~/.bashrc
echo 'export LD_LIBRARY_PATH=/usr/local/cuda/lib64:$LD_LIBRARY_PATH' >> ~/.bashrc
source ~/.bashrc

# Verify CUDA
nvcc --version

3. ROS 2 Humble Installation

# Add ROS 2 repository
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 $(. /etc/os-release && echo $UBUNTU_CODENAME) main" | sudo tee /etc/apt/sources.list.d/ros2.list > /dev/null

# Install ROS 2 Humble Desktop
sudo apt update
sudo apt install -y ros-humble-desktop

# Install development tools
sudo apt install -y python3-colcon-common-extensions python3-rosdep

# Initialize rosdep
sudo rosdep init
rosdep update

# Add to ~/.bashrc
echo "source /opt/ros/humble/setup.bash" >> ~/.bashrc
source ~/.bashrc

# Verify installation
ros2 run demo_nodes_cpp talker
# Should see "Hello World" messages

4. Isaac Sim Installation (Docker Method)

# Install Docker
curl -fsSL https://get.docker.com -o get-docker.sh
sudo sh get-docker.sh
sudo usermod -aG docker $USER
# Log out and back in for group changes

# Install NVIDIA Container Toolkit
distribution=$(. /etc/os-release;echo $ID$VERSION_ID)
curl -s -L https://nvidia.github.io/nvidia-docker/gpgkey | sudo apt-key add -
curl -s -L https://nvidia.github.io/nvidia-docker/$distribution/nvidia-docker.list | sudo tee /etc/apt/sources.list.d/nvidia-docker.list
sudo apt update && sudo apt install -y nvidia-docker2
sudo systemctl restart docker

# Pull Isaac Sim image (22GB download, be patient)
docker pull nvcr.io/nvidia/isaac-sim:4.0.0

# Create workspace
mkdir -p ~/isaac_workspace

# Run Isaac Sim
xhost +local:docker
docker run --rm -it \
  --gpus all \
  -e DISPLAY=$DISPLAY \
  -v /tmp/.X11-unix:/tmp/.X11-unix:rw \
  -v ~/isaac_workspace:/workspace \
  --network host \
  nvcr.io/nvidia/isaac-sim:4.0.0

# Inside container, launch Isaac Sim
./isaac-sim.sh

Verification Tests

Test 1: ROS 2 Communication

# Terminal 1
ros2 run demo_nodes_cpp talker

# Terminal 2
ros2 run demo_nodes_cpp listener
# Should see messages received

Test 2: Isaac Sim GPU Acceleration

# In Isaac Sim GUI:
# Create > Humanoid Robot > Unitree H1
# Press Play button (bottom-left)
# Robot should simulate in real-time (60 FPS)

# Check GPU usage
nvidia-smi
# GPU Utilization should be 80-100% during simulation

2. Development Setup (Sensors + Workstation)

Purpose: Develop and test perception/manipulation algorithms with real sensors before deploying to robot

Additional Hardware

Intel RealSense D435i Depth Camera

Specs:

RGB: 1920x1080 @ 30 FPS
Depth: 1280x720 @ 30 FPS (stereo IR + depth processor)
Range: 0.3m - 10m
IMU: Accelerometer + Gyroscope
Interface: USB 3.0

Purchase:

Price: $329 USD (official), $250-280 (third-party sellers)
Retailers: Intel Store, Amazon, Mouser Electronics
Link: https://www.intelrealsense.com/depth-camera-d435i/

Alternative: RealSense D405 (shorter range, 0.07-1.5m, better for manipulation): $129

Setup:

# Install librealsense SDK
sudo apt install -y ros-humble-realsense2-camera ros-humble-realsense2-description

# Connect camera via USB 3.0 (blue port)

# Test camera
rs-enumerate-devices
# Should show D435i details

# Launch ROS 2 node
ros2 launch realsense2_camera rs_launch.py enable_depth:=true enable_color:=true

# View RGB stream
ros2 run rqt_image_view rqt_image_view /camera/color/image_raw

# View depth stream
ros2 run rqt_image_view rqt_image_view /camera/depth/image_rect_raw

Calibration (See Appendix C: Troubleshooting, Camera Calibration section)

NVIDIA Jetson Orin Nano/NX (Edge Compute)

Use Case: Deploy trained models on robot for on-device inference (no cloud dependency)

Models:

Jetson Orin Nano (4GB/8GB): $199/$399, 512-1024 CUDA cores, 5-10 TOPS AI
Jetson Orin NX (8GB/16GB): $399/$599, 1024-2048 CUDA cores, 40-100 TOPS AI
Jetson AGX Orin (32GB/64GB): $999/$1999, 2048 CUDA cores, 200-275 TOPS AI

Recommendation: Orin NX 16GB for development (balance of performance + cost)

Purchase: https://www.nvidia.com/en-us/autonomous-machines/embedded-systems/jetson-orin/

Setup:

# Flash JetPack 5.1+ using NVIDIA SDK Manager
# Download: https://developer.nvidia.com/sdk-manager

# On Jetson (after flashing):
# Install ROS 2 Humble (ARM64 binaries)
sudo apt install -y ros-humble-desktop

# Install Isaac Sim for Jetson (lighter image)
docker pull nvcr.io/nvidia/isaac-sim:4.0.0-jetson

# Test GPU
sudo tegrastats
# Should show GPU/CPU utilization

Development Accessories

Mounting Hardware:

Camera Mount: Tripod or robotic arm clamp ($20-50)
Jetson Case: Waveshare metal case with fan ($30)
Power Supply: USB-C PD 65W for Jetson ($25)

Networking:

WiFi 6 Router: For low-latency ROS 2 communication ($80-150)
- Recommended: TP-Link AX3000, ASUS RT-AX55
Ethernet Switch: If using wired (Gigabit recommended) ($20-40)

Total Development Setup Cost: $2500-4000

3. Production Setup (Full Humanoid Robot)

Purpose: Real-world deployment, research, commercial applications

Humanoid Robot Platforms

Option A: Unitree H1 Humanoid

Specs:

Height: 180 cm, Weight: 47 kg
DoF: 19 (legs: 12, arms: 6, head: 1)
Actuators: Unitree motors (peak 360 Nm)
Battery: 15 Ah (2 hours operation)
Compute: Jetson AGX Orin (32GB)
Sensors: RealSense D435i, IMU, force sensors in feet
Connectivity: Ethernet, WiFi, 4G optional

Price: ~$90,000 USD (2024 estimate, contact Unitree for quote)

Purchase: https://www.unitree.com/ (B2B sales, requires application)

Pros:

Full ROS 2 support out-of-box
Open SDK (Python/C++ APIs)
Strong community (research labs worldwide)
Spare parts available

Cons:

Expensive
Long lead time (3-6 months)
Export restrictions (check your country)

Option B: Agility Robotics Digit

Specs:

Height: 160 cm, Weight: 46 kg
DoF: 20 (legs: 12, arms: 4, torso: 4)
Battery: 2-3 hours
Payload: 16 kg
Sensors: LiDAR, cameras, IMU

Price: ~$250,000 USD (lease options available)

Purchase: https://agilityrobotics.com/ (commercial sales)

Pros:

Production-grade reliability
Warehouse-tested (Amazon partnership)
Excellent support

Cons:

Very expensive
Closed software stack (limited customization)
Long-term contracts

Option C: DIY Open-Source Humanoid

Projects:

Open Dynamic Robot Initiative (Max Planck Institute): https://open-dynamic-robot-initiative.github.io/
- Cost: $15,000-30,000 (parts + machining)
- Difficulty: Expert (requires mechanical engineering)
THOR (UCLA): Open-source humanoid design
- Cost: $20,000-40,000
- Difficulty: Advanced

Pros:

Full control over design
Educational value
Community support

Cons:

Requires machining/fabrication access
Time-intensive (6-12 months build)
No warranty or support

Production Compute Setup

Onboard Compute (Robot)

Jetson AGX Orin 64GB: $1999 (AI inference, perception)
Intel NUC 12 Pro (i7, 64GB RAM): $1200 (high-level planning, ROS 2 coordination)

Offboard Compute (Workstation/Server)

Training Server:
- CPU: AMD Threadripper 3970X (32 cores): $2000
- GPU: 2x RTX 4090 (24GB VRAM each): $4000
- RAM: 128GB DDR4: $400
- Storage: 4TB NVMe SSD: $300
- Total: ~$7000
- Use: Isaac Lab RL training, large-scale data processing

Networking

5G/WiFi 6E Router: Low-latency teleoperation ($200-400)
Edge Server: For local LLM inference (avoid cloud latency)
- Mini PC with RTX 4060 Ti: $1200
- Run Llama 3 70B or GPT-4o via API

Total Production Setup Cost: $100,000-300,000 (robot + compute + accessories)

4. Recommended Purchase Order (Development Setup)

Phase 1: Foundations ($1500)

Desktop workstation OR gaming laptop with RTX GPU ($1200)
External SSD for datasets (1TB): $80
USB 3.0 hub (powered): $30
Monitor (if desktop): $150-300

Phase 2: Sensors ($500-800)

Intel RealSense D435i: $329
Camera tripod/mount: $40
Jetson Orin Nano 8GB Dev Kit: $399 (optional, if want edge testing)

Phase 3: Scaling ($1000+)

Upgrade to Jetson Orin NX 16GB: $599
Additional RealSense for stereo vision: $329
LiDAR (RPLiDAR A1): $99 (for navigation testing)
Force-torque sensor (6-axis, Robotiq FT 300): $4000 (advanced manipulation)

5. Maintenance and Upgrades

Regular Maintenance

Software:

Update Ubuntu packages: sudo apt update && sudo apt upgrade (monthly)
Update ROS 2 packages: sudo apt install ros-humble-desktop (quarterly)
Update Isaac Sim: Pull latest Docker image (when new features needed)

Hardware:

Clean GPU fans: Every 6 months (compressed air)
Check cable connections: Monthly (USB 3.0 can become loose)
Calibrate cameras: After any physical movement

Upgrade Path

Year 1-2: Development Setup

Workstation + RealSense + Jetson Orin Nano
Focus: Learning, algorithm development in simulation

Year 2-3: Prototype Hardware

Add: Robotic arm (UR5e: $20,000 or Franka Emika Panda: $25,000)
Purpose: Test manipulation policies on real hardware

Year 3+: Full Humanoid

Unitree H1 or custom build
Purpose: End-to-end system deployment

6. Troubleshooting Hardware Issues

GPU not detected in Docker:

# Verify nvidia-docker2 installed
docker run --rm --gpus all nvidia/cuda:12.1.0-base-ubuntu22.04 nvidia-smi

# If fails, reinstall NVIDIA Container Toolkit
sudo apt purge nvidia-docker2
sudo apt install -y nvidia-docker2
sudo systemctl restart docker

RealSense camera not found:

# Check USB 3.0 connection (blue port)
lsusb | grep Intel
# Should show "Intel Corp. RealSense"

# Check permissions
sudo usermod -aG video $USER
# Log out and back in

# Reset camera
sudo systemctl restart udev

Jetson boot issues:

Use official NVIDIA power supply (65W USB-C PD)
Flash with latest JetPack via SDK Manager
Check SD card/NVMe SSD seating

7. Safety Equipment (Production Setups)

For Humanoid Robots:

Emergency Stop Button (wireless): $200-500
- Mount on robot torso (easily accessible)
- Wired backup on workstation
Safety Enclosure: $5000-20,000 (if testing dangerous behaviors)
- Plexiglass walls, padded floor
Personal Protective Equipment:
- Safety glasses (flying debris from dropped objects)
- Steel-toe boots (robot feet can cause injury)
- Hearing protection (some motors are loud)

Electrical Safety:

Surge protector: $50-100
Uninterruptible Power Supply (UPS): $200-500 (prevents data loss during power outages)

Budget Summary Table

Setup Tier	Hardware Cost	Annual Software/Cloud	Total Year 1
Minimal (Simulation Only)	$1,200-2,000	$0-500 (cloud GPU optional)	$1,200-2,500
Development (Sensors + Workstation)	$2,500-4,000	$200 (OpenAI API, cloud storage)	$2,700-4,200
Production (Full Humanoid)	$100,000-300,000	$2,000 (cloud, API, maintenance)	$102,000-302,000

Recommended Retailers by Region

North America:

Newegg, Micro Center, B&H Photo (electronics)
RobotShop, TrossenRobotics (robotics-specific)

Europe:

Scan.co.uk, Alternate.de (electronics)
RobotShop EU, Generation Robots (robotics)

Asia:

Taobao, JD.com (China)
Akizuki Denshi, Switch Science (Japan)
Lazada, Shopee (Southeast Asia)

Official Vendors (Worldwide):

Intel (RealSense cameras)
NVIDIA (Jetson dev kits)
Unitree Robotics (humanoid robots)

Last Updated: 2025-12-05 Note: Prices are estimates as of 2024-2025 and may vary by region and availability. Always check current pricing and compatibility before purchasing.

Overview​

1. Minimal Setup (Simulation-Only)​

Hardware Requirements​

Compute Platform (Choose One)​

Software Installation​

1. Ubuntu 22.04 LTS Setup​

2. NVIDIA Drivers + CUDA​

3. ROS 2 Humble Installation​

4. Isaac Sim Installation (Docker Method)​

Verification Tests​

2. Development Setup (Sensors + Workstation)​

Additional Hardware​

Intel RealSense D435i Depth Camera​

NVIDIA Jetson Orin Nano/NX (Edge Compute)​

Development Accessories​

3. Production Setup (Full Humanoid Robot)​

Humanoid Robot Platforms​

Option A: Unitree H1 Humanoid​

Option B: Agility Robotics Digit​

Option C: DIY Open-Source Humanoid​

Production Compute Setup​

Onboard Compute (Robot)​

Offboard Compute (Workstation/Server)​

Networking​

4. Recommended Purchase Order (Development Setup)​

Phase 1: Foundations ($1500)​

Phase 2: Sensors ($500-800)​

Phase 3: Scaling ($1000+)​

5. Maintenance and Upgrades​

Regular Maintenance​

Upgrade Path​

6. Troubleshooting Hardware Issues​

7. Safety Equipment (Production Setups)​

Budget Summary Table​

Recommended Retailers by Region​

Overview

1. Minimal Setup (Simulation-Only)

Hardware Requirements

Compute Platform (Choose One)

Software Installation

1. Ubuntu 22.04 LTS Setup

2. NVIDIA Drivers + CUDA

3. ROS 2 Humble Installation

4. Isaac Sim Installation (Docker Method)

Verification Tests

2. Development Setup (Sensors + Workstation)

Additional Hardware

Intel RealSense D435i Depth Camera

NVIDIA Jetson Orin Nano/NX (Edge Compute)

Development Accessories

3. Production Setup (Full Humanoid Robot)

Humanoid Robot Platforms

Option A: Unitree H1 Humanoid

Option B: Agility Robotics Digit

Option C: DIY Open-Source Humanoid

Production Compute Setup

Onboard Compute (Robot)

Offboard Compute (Workstation/Server)

Networking

4. Recommended Purchase Order (Development Setup)

Phase 1: Foundations ($1500)

Phase 2: Sensors ($500-800)

Phase 3: Scaling ($1000+)

5. Maintenance and Upgrades

Regular Maintenance

Upgrade Path

6. Troubleshooting Hardware Issues

7. Safety Equipment (Production Setups)

Budget Summary Table

Recommended Retailers by Region