Week 1–2: Foundations of Physical AI & Embodied Intelligence

Learning Objectives

By the end of this section, you will be able to:

• Define Physical AI and distinguish it from purely digital AI systems
• Explain the concept of Embodied Intelligence and why it matters for robotics
• Identify key sensor modalities used in humanoid robots (LiDAR, cameras, IMUs, F/T sensors)
• Explain why the humanoid form factor is strategically important for real-world deployment
• Describe the basic perception-reasoning-action loop of a robotic agent

What Is Embodied Intelligence?

Embodied Intelligence is the idea that true intelligence cannot exist independently of a physical body interacting with the world. Humans learn by touching, moving, and manipulating objects — not just by reading about them. Robots with embodied intelligence learn similarly.

Key insight: The world is the best model of itself. Instead of building a perfect internal simulation, an embodied agent continuously queries the real world through sensors and updates its beliefs in real time.

Why Humanoids?

The world is built for humans — doorknobs at hand height, stairs at human stride length, tools designed for human hands. A humanoid robot can:

• Navigate human-scale environments without modification
• Use existing tools and infrastructure
• Collaborate naturally alongside human workers
• Generalize learned behaviors across diverse tasks

Major investments from Boston Dynamics (Atlas), Figure AI, Agility Robotics (Digit), and Tesla (Optimus) reflect the industrial confidence in humanoid robotics.

Sensor Systems Overview

Sensor	Type	Use Case
RGB Camera	Visual	Object detection, face recognition, navigation
Depth Camera (RealSense, ZED)	Visual+Depth	3D mapping, obstacle avoidance
LiDAR	Laser Range	Precise mapping, SLAM
IMU (Inertial Measurement Unit)	Inertial	Balance, orientation, fall detection
Force/Torque (F/T) Sensors	Tactile	Grasping, contact force control
Microphone Array	Audio	Voice commands, sound localization

The Perception-Reasoning-Action Loop

# Simplified Physical AI agent loop (conceptual)
import rclpy
from rclpy.node import Node

class PhysicalAIAgent(Node):
    def __init__(self):
        super().__init__('physical_ai_agent')
        self.sensor_sub = self.create_subscription(
            SensorData, '/sensors/all', self.perception_callback, 10
        )
        self.action_pub = self.create_publisher(
            ActionCommand, '/robot/action', 10
        )

    def perception_callback(self, sensor_data):
        # 1. PERCEIVE: Process raw sensor data
        world_state = self.perceive(sensor_data)

        # 2. REASON: Plan next action
        action = self.reason(world_state)

        # 3. ACT: Execute in the physical world
        self.action_pub.publish(action)

Key Concepts Review

Key Concept

Embodied Intelligence = AI + Physical Body + Real-World Interaction. The body is not just a container for the AI — it is integral to how intelligence develops and operates.

The three pillars of Physical AI:

1. Sensing — Rich, multi-modal perception of the physical world
2. World Modeling — Maintaining an internal representation of the environment
3. Closed-Loop Control — Continuous feedback between sensing and acting

---

→ Next Module: Module 1 – The Robotic Nervous System (ROS 2)