Skip to main content

Week 1: Foundations of Physical AI

The Embodiment Hypothesis​

Physical AI refers to the synthesis of Artificial Intelligence with physical systems. Unlike Large Language Models (LLMs) that live in the "ethereal" digital space, Physical AI agents must negotiate with the immutable laws of physics: gravity, friction, and inertia.

"Intelligence cannot exist without a body." — Rodney Brooks (MIT)

This hypothesis suggests that true general intelligence requires sensorimotor interaction. A model that has never felt "heavy" cannot impactfuly reason about moving a piano.

The Humanoid Form Factor​

Why are we seeing a surge in Humanoids (Tesla Optimus, Figure 01, Unitree H1)?

  • Infrastructure Compatibility: Our stairs, door handles, and tools are designed for bipeds with two hands.
  • Data Availability: YouTube is full of humans doing things (Ego-Centric Data), which provides a massive training set for imitation.
graph TD;
    A[Digital Intelligence] -->|Sim-to-Real| B[Physical Intelligence];
    B --> C{The Body};
    C -->|Actuators| D[World Interaction];
    C -->|Sensors| E[Perception];
    E --> B;
    style B fill:#f9f,stroke:#333,stroke-width:4px

Lab 1: Establishing the Forge​

Robotics development is notoriously difficult to set up. We will support Linux (Native), Windows (WSL2), and macOS where possible.

Prerequisites​

  • Linux: Ubuntu 22.04 LTS (Standard)
  • Windows: Windows 10/11 with WSL2 enabled.
  • Mac: Apple Silicon (M1/M2/M3) - Note: GPU Simulation support is limited on Mac.

Step 1: System Preparation​

Isolate your environment using Docker to prevent breaking your system Python.

# 1. Update and Install Essentials
sudo apt update && sudo apt install -y curl git build-essential

# 2. Install Docker
curl -fsSL https://get.docker.com -o get-docker.sh
sudo sh get-docker.sh

# 3. NVIDIA Container Toolkit (CRITICAL for Sim)
curl -fsSL https://nvidia.github.io/libnvidia-container/gpgkey | sudo gpg --dearmor -o /usr/share/keyrings/nvidia-container-toolkit-keyring.gpg
sudo apt-get install -y nvidia-container-toolkit
sudo nvidia-ctk runtime configure --runtime=docker
sudo systemctl restart docker

Step 2: Verification​

Let's ensure your GPU is accessible for training neural networks.

test_gpu.py
import torch
print(f"CUDA Available: {torch.cuda.is_available()}")
print(f"Device Name: {torch.cuda.get_device_name(0)}")