What is Physical AI?
Physical AI refers to AI systems that can perceive and interact with the physical world — through sensors that read real-world data (temperature, motion, sound, light) and actuators that take physical action (motors, speakers, displays, valves).
The key insight: Physical AI is defined by what it connects to, not where it runs. The AI brain can live on the cloud, on an edge server, or directly on the device — as long as it is communicating with the physical world, it qualifies as Physical AI.
How Physical AI Works
Physical World
│
[Sensors] ← reads temperature, motion, sound, images…
│
[AI System] ← can be cloud, edge server, or on-device
│
[Actuators] ← controls motors, alarms, displays, valves…
│
Physical World
Even a cloud-hosted AI that only reads sensor data (without controlling anything) still counts as Physical AI — it is grounded in real-world inputs rather than purely digital ones.
TinyML vs. Edge AI vs. Physical AI
These three terms are often confused. Here is how they differ:
| TinyML | Edge AI | Physical AI | |
|---|---|---|---|
| What it means | ML running on a tiny microcontroller | AI running on or near the device | AI that interacts with the physical world |
| Where AI runs | On the microcontroller itself | On the device / local server | Anywhere — cloud, edge, or on-device |
| Internet needed? | No | Usually not | Optional |
| Sensors / Actuators | Yes (core to TinyML) | Usually yes | Yes — this is the defining feature |
| Compute power | Milliwatts, kilobytes of RAM | Moderate (Raspberry Pi, Jetson, phone) | Varies widely |
| Example | Keyword detection on Arduino | Face recognition on a Jetson Nano | A cloud robot that reads a camera and moves an arm |
Quick way to remember
- TinyML → tiny device, ML runs locally
- Edge AI → AI runs close to the data source, not in the cloud
- Physical AI → AI with eyes, ears, and hands in the real world — regardless of where the brain lives
Physical AI Does Not Have to Run on the Edge
This is an important distinction. Consider a smart irrigation system:
- A soil moisture sensor sends readings over Wi-Fi to a cloud server
- A cloud AI model analyses the data and decides whether to water
- It sends a command back to a motorised valve (actuator) that opens or closes
This is Physical AI — it reads the physical world and acts on it — but the AI itself runs entirely in the cloud. No edge device, no microcontroller required.
Real-World Examples
| Example | Sensors | Actuators | Where AI Runs |
|---|---|---|---|
| Smart thermostat | Temperature sensor | Heating/cooling relay | Cloud or on-device |
| Self-driving car | Camera, LiDAR, radar | Steering, brakes, throttle | On-vehicle (edge) |
| Industrial robot arm | Vision camera, force sensor | Servo motors | On-device or edge server |
| Remote weather station | Temp, humidity, wind | Satellite uplink | Cloud |
| Voice assistant (Alexa) | Microphone | Speaker | Cloud |
| TinyML gesture wristband | IMU (accelerometer) | Haptic motor | On-device (TinyML) |
Why Physical AI Matters
Bridges the Digital–Physical Gap
Most AI today lives in the digital world — processing text, images, and numbers. Physical AI closes the loop between software intelligence and real-world action.
Enables Autonomous Systems
Robots, drones, autonomous vehicles, and smart buildings all depend on Physical AI to sense their environment and respond in real time.
Expands Where AI Can Run
Because the AI can live anywhere — cloud, edge, or device — Physical AI enables use cases where the sensor/actuator node is cheap and simple, while the intelligence stays in a powerful server.
How TinyML Fits Inside Physical AI
TinyML is a subset of Physical AI. When you run a machine learning model directly on a microcontroller like the Arduino Nicla Q, you get the benefits of Physical AI (real-world sensing and actuation) plus the benefits of on-device inference: no internet required, near-zero latency, and ultra-low power consumption.
Physical AI (broad — AI connected to the real world)
└── Edge AI (AI runs near the data source)
└── TinyML (AI runs on a tiny microcontroller)
Next Steps
→ What is Edge AI? — understand where AI runs at the edge
→ What is Machine Learning? — the foundations of ML
→ Workshop Overview — see the full agenda