IoT & Robotics: AI in the Physical World

20 slides on what happens when AI gets a body — the sensors, robots, and humanoids reshaping the physical economy, and exactly how to start building for under $200.

20 lessons · ~4 min · free, no signup.

Lessons

Full course text

1. IoT (Internet of Things)

Physical objects — sensors, meters, machines, vehicles — connected to the internet, streaming data and taking commands. IoT is how AI gets eyes and hands in the real world.

Source: IoT Analytics — State of IoT

2. Connected devices, worldwide

Connected IoT devices number around 20 billion globally and are forecast to roughly double by 2030. Every one is a data source an AI model can learn from — and increasingly, act on.

Source: IoT Analytics, State of IoT (2025)

3. Physical AI / Embodied AI

AI that perceives and acts in the physical world through sensors and actuators — robots, drones, autonomous vehicles. The 2026 breakthrough: robots running foundation models instead of hand-coded rules.

Source: NVIDIA — Physical AI

4. Industrial robots installed

542,000 industrial robots were installed worldwide in 2024 — more than double the number a decade earlier. Asia took 74% of deployments; the installed market value hit an all-time high of $16.7 billion.

Source: IFR, World Robotics 2025

5. Cobot

A collaborative robot designed to work safely beside humans — no cage required. Cobots are the fastest-growing robot class because they deploy in weeks, cost a fraction of traditional cells, and can be re-tasked across a shop floor.

Source: IFR — collaborative robots

6. Humanoid platforms available

Commercially available humanoid robot platforms jumped from 3 in 2024 to 12 in 2026, priced from ~$28K (torso-only) to ~$245K (full biped) — with lease programs at $3,500–8,000/month. Humanoid market projected: ~$2B (2024) → ~$18B by 2030.

Source: Silicon Valley Robotics — State of Robotics 2026; MarketsandMarkets

7. VLA (Vision-Language-Action)

Foundation models that take camera input + a plain-English instruction (‘put the red parts in bin 3’) and output robot movements. VLA adoption tripled in a year — now in ~40% of new robot deployments. Programming is becoming prompting.

Source: State of Robotics 2026

8. Edge AI

Running AI models on the device itself — a camera, a controller, a sensor hub — instead of the cloud. Why it matters: millisecond decisions, works offline, keeps sensitive data local. A $250 Jetson-class board now runs serious vision models.

Source: NVIDIA Jetson — Edge AI

9. Digital twin

A live virtual replica of a physical asset — machine, warehouse, city block — fed by IoT sensors. Teams simulate changes and train robot AI in the twin before touching the real thing, collapsing risk and cost.

Source: IBM — What is a digital twin?

10. Predictive maintenance pays

AI-driven predictive maintenance cuts equipment downtime by ~45% and maintenance costs by ~25%. It’s the most reliable first ROI in industrial AI: sensors + history + a model that hears failure coming.

Source: Deloitte / McKinsey analyses

11. “Robots that use AI to work independently are becoming more common.”

— The IFR’s #1 robotics trend for 2026. Analytical AI predicts failures; generative AI lets robots learn tasks and generate their own training data in simulation.

Source: IFR, Top 5 Robotics Trends 2026

12. U.S. robot installs, 2025

U.S. robot installations rose 11% in 2025 to 38,000 units — with growth led not by automotive but by food and non-manufacturing sectors. Chronic labor shortages, not layoffs, are the main driver companies cite.

Source: IFR (June 2026)

13. Getting started: under $200

Raspberry Pi 5 or ESP32 microcontroller + a few sensors (temperature, motion, camera) + Python. First project: a sensor that streams readings to a dashboard and texts you on anomalies. You’ll touch the whole IoT stack — device, network, data, alert.

Source: Raspberry Pi Foundation — Projects

14. Getting started: robotics

Hugging Face’s open-source LeRobot gives you trainable robot AI, and the SO-ARM class of hobby arms costs ~$100–300. Teleoperate the arm, record demonstrations, train a policy, watch it repeat the task. That loop IS modern robotics.

Source: Hugging Face — LeRobot

15. MQTT & telemetry

MQTT is the lightweight publish/subscribe protocol most IoT devices use to talk. Telemetry is the continuous data they emit. Business angle: telemetry is proprietary training data no competitor can scrape — the most defensible dataset you can own.

Source: MQTT.org

16. Simulation first

NVIDIA Isaac Sim and open-source Gazebo let you build and train robots in physics-accurate virtual worlds — free — before buying hardware. Imitation learning (learning from human demonstrations) overtook reinforcement learning as the main training method in 2026.

Source: NVIDIA Isaac Sim; State of Robotics 2026

17. The convergence

Forrester flags ‘physical AI’ — agents coordinating robots, sensors, and supply chains in real time — as a defining 2026–27 shift. Digital agents plus edge hardware is the highest-impact combo in manufacturing and logistics.

Source: Forrester Predictions 2026

18. Myth: robotics is only for factories

2024 alone: professional service robots grew 9%, consumer service robots 11%, and medical robots 91% year-over-year. Agriculture, hospitals, hotels, and last-mile delivery are the growth frontier.

Source: IFR, World Robotics 2025 — Service Robots

19. The business entry point

Don’t start with a robot; start with a measurement problem. Instrument one process with cheap sensors, collect 90 days of telemetry, let AI find the pattern (waste, downtime, drift), then automate the response. Data before actuators, always.

Source: McKinsey — IoT value creation

20. AI has a body now. You know how it works.

Sensors in, models in the middle, action out — from a $200 Pi rig to a $245K humanoid, it’s the same loop. Final course: assembling everything into a durable business system.

Source: Continue → Advanced AI Architecture

All courses · Robauto