Skip to main content

Digital Twin Workstation, Jetson Edge Kit, and Sensors

This section describes the hardware components recommended or used for practical implementations and exercises throughout this textbook. These components are crucial for developing and testing Physical AI systems, from high-fidelity simulations on powerful workstations to real-world deployments on edge devices and robots.

Digital Twin Workstation (RTX + Ubuntu)​

A high-performance workstation is essential for running complex physics simulations and AI model training, especially when working with NVIDIA Isaac Sim and large datasets.

  • Processor: Intel Core i7/i9 or AMD Ryzen 7/9 (latest generation)
  • RAM: 32 GB or more
  • GPU: NVIDIA RTX series (e.g., RTX 3080, 4090) with at least 10 GB VRAM. This is critical for GPU-accelerated simulations and AI inference.
  • Operating System: Ubuntu (LTS version, e.g., 20.04 or 22.04). Ubuntu provides the best compatibility and performance for ROS 2 and NVIDIA developer tools.

This workstation serves as the primary development environment for designing digital twins, running high-fidelity simulations, and training deep learning models.

Jetson Orin Edge Kit​

For deploying AI models and running ROS 2 applications on a physical robot or at the edge, NVIDIA Jetson platforms are highly recommended. The Jetson Orin series offers significant AI performance in a compact, power-efficient form factor.

  • Model: NVIDIA Jetson Orin Nano Developer Kit or Jetson AGX Orin Developer Kit.
  • Purpose: On-robot AI inference, sensor data processing, and local control.
  • Software: JetPack SDK, including CUDA, cuDNN, TensorRT, and pre-built ROS 2 packages optimized for Jetson.

The Jetson Orin is ideal for bridging the gap between simulation and real-world deployment, enabling edge computing for robotics.

Sensors (Intel RealSense, IMU)​

Accurate perception is fundamental for any intelligent robot. The following sensors are commonly used and recommended:

  • Intel RealSense Depth Camera: Provides high-quality RGB-D (color and depth) data, essential for 3D perception, object detection, and SLAM.
    • Models: D435i, D455 (integrated IMU in 'i' models is a bonus).
  • IMU (Inertial Measurement Unit): Crucial for robot localization, pose estimation, and balance control.
    • Integration: Often integrated into RealSense cameras or as standalone units (e.g., Adafruit BNO055, Xsens MTi series).

These sensors provide the necessary inputs for the robot's perception pipelines, allowing it to understand its surroundings and its own state.

Robots (Unitree Go2, G1)​

This textbook uses conceptual humanoid robot examples, but for physical implementations, quadrupeds and humanoids from Unitree Robotics are excellent platforms for experimentation.

  • Unitree Go2: A versatile quadruped robot suitable for locomotion, navigation, and manipulation tasks. Its robust design makes it an excellent platform for learning physical AI concepts.
  • Unitree G1: A research-grade humanoid robot, offering advanced capabilities for bipedal locomotion and complex manipulation, ideal for pushing the boundaries of humanoid AI.

These robots provide the physical embodiment necessary to test and validate the algorithms developed in simulation.

Cloud-Native Lab (AWS Omniverse)​

For scalable AI training, collaborative simulation, and remote development, cloud platforms integrated with Omniverse can provide significant advantages.

  • AWS EC2 with NVIDIA GPUs: For training large AI models that require more computational power than a local workstation.
  • NVIDIA Omniverse Cloud: Enables collaborative 3D workflows and scalable simulation environments accessible from anywhere, fostering distributed development and testing.

This cloud integration supports advanced research and large-scale deployment scenarios, providing flexibility and scalability beyond local hardware constraints.

💬