Hardware Prerequisites

This document outlines the hardware requirements for developing and deploying Physical AI and Humanoid Robotics systems. Specifications are categorized by use case and budget level.

Development Workstation

RTX Workstation Specifications

Component	Minimum	Recommended	High-End
GPU	NVIDIA RTX 3060 (12GB)	NVIDIA RTX 3080 Ti (12GB)	NVIDIA RTX 4090 (24GB)
CPU	Intel i5-12600K / AMD Ryzen 5 5600X	Intel i7-13700K / AMD Ryzen 7 7700X	Intel i9-13900K / AMD Ryzen 9 7950X
RAM	32 GB DDR4	64 GB DDR5	128 GB DDR5
Storage	500 GB NVMe SSD	1 TB NVMe SSD	2 TB NVMe SSD + 4 TB HDD
PSU	650W 80+ Gold	850W 80+ Gold	1200W 80+ Platinum
Network	Gigabit Ethernet	Gigabit Ethernet + WiFi 6	10GbE + WiFi 6E

GPU Requirements by Module

Module	Minimum GPU	Recommended GPU	VRAM Required
ROS 2	Integrated / GTX 1650	RTX 3060	4 GB
Digital Twin	RTX 3060	RTX 3080	8 GB
NVIDIA Isaac	RTX 3060 (12GB)	RTX 4090 (24GB)	12 GB
VLA Training	RTX 3090 (24GB)	A100 (40-80GB)	24 GB

NVIDIA GPU Compatibility

Supported GPU Architectures:
├── Ampere (RTX 30xx, A100)
│   ├── Full support for all modules
│   └── Recommended for Isaac Sim
├── Ada Lovelace (RTX 40xx)
│   ├── Best performance for Isaac Sim
│   └── DLSS 3.0 support
├── Hopper (H100)
│   ├── Datacenter training
│   └── VLA model training
└── Jetson Orin (Edge)
    ├── Edge deployment
    └── Real-time inference

Jetson Edge Deployment Kit

Jetson Orin Series Comparison

Specification	Orin Nano	Orin NX	AGX Orin
GPU	Ampere (1024 CUDA)	Ampere (1024 CUDA)	Ampere (2048 CUDA)
Tensor Cores	32	32	64
CPU	6-core ARM Cortex-A78AE	6-core ARM Cortex-A78AE	12-core ARM Cortex-A78AE
Memory	4-8 GB LPDDR5	8-16 GB LPDDR5	32-64 GB LPDDR5
Storage	microSD / NVMe	microSD / NVMe	NVMe / eMMC
Power	7-15W	10-25W	15-60W
AI Performance	20-40 TOPS	70-100 TOPS	100-275 TOPS
Price	~$199	~$399	~$1999

Recommended Jetson Kit

Jetson Edge Development Kit:
├── Jetson Orin NX 16GB (Recommended for most applications)
├── Carrier Board with:
│   ├── 2x Gigabit Ethernet
│   ├── 4x USB 3.2
│   ├── 2x CSI Camera interfaces
│   ├── GPIO / I2C / SPI / UART
│   └── M.2 Key M for NVMe SSD
├── Power Supply: 19V/65W
├── Cooling: Active heatsink with fan
└── Enclosure: IP65 rated for robotics

Peripheral Requirements

Peripheral	Specification	Purpose
Camera	Intel RealSense D435i / OAK-D	Depth perception, VLA input
LiDAR	Ouster OS0-16 / Velodyne VLP-16	Navigation, mapping
IMU	Xsens MTi-300 / Bosch BMI088	State estimation, balance
Motor Controller	ODrive v3.6 / T-Motor AK Series	Joint actuation
Battery	6S LiPo 22000mAh / LiFePO4 25.6V	Power system

Robot Hardware Specifications

Humanoid Robot Platform

Component	Specification	Notes
DOF	20-40 degrees of freedom	Full humanoid
Height	1.2 - 1.8 meters	Human-scale
Weight	30 - 80 kg	Depends on payload
Payload	5 - 20 kg	End-effector capacity
Actuators	Brushless DC with harmonic drives	High torque density
Sensors	Force-torque, IMU, encoders	Proprioception

Sensor Suite

Recommended Sensor Configuration:
├── Vision
│   ├── RGB-D Camera (head): Intel RealSense D455
│   ├── Stereo Cameras (eyes): OV9281 global shutter
│   └── Event Camera (optional): Prophesee Gen4
├── Proprioception
│   ├── IMU: Xsens MTi-670 (head)
│   ├── Joint Encoders: Absolute magnetic (all joints)
│   └── Force-Torque: ATI Mini40 (ankles, wrists)
├── Exteroception
│   ├── LiDAR: Ouster OS0-16 (torso)
│   ├── Ultrasonic: HC-SR04 (collision avoidance)
│   └── Tactile: GelSight / tactile fingers
└── Audio
    ├── Microphone Array: ReSpeaker 4-mic
    └── Speaker: 3W mono

Development Environment Setup

Software Prerequisites

# Operating System
Ubuntu 22.04 LTS (Jammy Jellyfish)

# ROS 2 Distribution
ROS 2 Humble Hawksbill

# Python Version
Python 3.10+

# CUDA Version
CUDA 11.8 / 12.1 (for Isaac Sim and VLA training)

# Docker
NVIDIA Container Toolkit (for Isaac Sim containers)

Environment Variables

# ~/.bashrc configuration

# ROS 2
source /opt/ros/humble/setup.bash
export ROS_DOMAIN_ID=0

# Isaac Sim
export ISAACSIM_PATH="${HOME}/.local/share/ov/pkg/isaac-sim"
export LD_LIBRARY_PATH="${ISAACSIM_PATH}/exts/isaac-sim:${LD_LIBRARY_PATH}"

# CUDA
export CUDA_HOME=/usr/local/cuda
export LD_LIBRARY_PATH="${CUDA_HOME}/lib64:${LD_LIBRARY_PATH}"
export PATH="${CUDA_HOME}/bin:${PATH}"

# Jetson (if applicable)
export JETSON_STATS=/home/nvidia/jetson-stats

Network Requirements

Multi-Robot Communication

Requirement	Specification	Notes
Latency	< 10 ms	Real-time control
Bandwidth	> 100 Mbps	Camera streams
Jitter	< 1 ms	Deterministic control
Protocol	UDP/TCP with QoS	ROS 2 DDS

Recommended Network Setup

Network Architecture:
├── Management Network (1 GbE)
│   ├── SSH access
│   ├── File transfer
│   └── Monitoring
├── Real-time Network (10 GbE or TSN)
│   ├── Robot control
│   ├── Sensor data
│   └── Low-latency communication
└── WiFi 6E (optional)
    ├── Mobile robot communication
    └── Remote monitoring

Budget Estimates

Development Setup Costs

Tier	Workstation	Jetson Kit	Sensors	Total
Entry	$1,500	$400	$500	$2,400
Professional	$3,500	$800	$2,000	$6,300
Research	$8,000	$2,500	$10,000	$20,500

Cloud Alternatives

For teams without local GPU resources:

Provider	Instance	Hourly Rate	Use Case
AWS	g5.2xlarge (A10G)	~$1.00/hr	VLA training
GCP	a2-highgpu-1g (A100)	~$3.00/hr	Large-scale training
Lambda Labs	1x RTX 4090	~$0.50/hr	Isaac Sim, training
RunPod	1x RTX 3090	~$0.40/hr	Cost-effective training

Troubleshooting

Common Hardware Issues

Issue	Symptom	Solution
GPU Memory	OOM errors in Isaac Sim	Reduce scene complexity, use smaller textures
Thermal	Throttling during training	Improve cooling, reduce ambient temperature
Network	High latency in ROS 2	Use wired connection, check QoS settings
Jetson Power	Unexpected shutdowns	Use adequate power supply, check current limits

Resources

• NVIDIA GPU Support Matrix
• Jetson Developer Guide
• ROS 2 Hardware Guide

Next Steps

After verifying your hardware setup, proceed to:

1. Module 1: ROS 2 Fundamentals
2. Module 2: Digital Twin Simulation
3. Module 3: NVIDIA Isaac Platform
4. Module 4: Vision-Language-Action Models