Physical AI is quickly becoming one of the most exciting areas in AI, but it can be difficult to know where to begin. This talk is an introduction for anyone who wants to understand what Physical AI is, how these systems work, and how to start building them.
Rather than being a theory-heavy session, this will be a hands-on introduction to Physical AI with live demonstrations. The session will feature LeRobot, Reachy Mini, and OpenDuck Mini, along with the hardware used to build and train them, allowing attendees to experience these systems in action rather than through videos or slides.
I'll share my experience of building and training these robots from scratch, explaining the complete Physical AI workflow from hardware and sensors to data collection, model training, and deployment. Along the way, I'll simplify the core concepts behind Physical AI using real-world examples and practical demonstrations.
My goal is to make Physical AI approachable for beginners, makers, and AI enthusiasts. By the end of the session, attendees will have a clear understanding of the Physical AI ecosystem, the open-source tools available today, and how they can start building and experimenting with Physical AI on their own.
Live Demonstration Robots
Robot Name | Brief Description | Reference Demonstration |
OpenDuckMini | An open-source bipedal walking robot that demonstrates locomotion, balance, and Physical AI. The demo showcases how imitation learning enables the robot to walk, respond to commands, and interact with its environment. | https://www.youtube.com/watch?v=7_LW7u-nk6Q |
ReachyMini | An open-source desktop humanoid robot designed for human-robot interaction. The demo highlights expressive movements, vision-based interaction, voice capabilities, and AI-powered behaviors for building interactive robotics applications. | https://www.youtube.com/watch?v=JvdBJZ-qR18 |
LeRobot | A Physical AI framework featuring robotic arms and hands trained through imitation learning. The demonstration covers teleoperation, data collection, policy training, and autonomous task execution, showing the complete workflow for building intelligent manipulation systems. | https://www.youtube.com/watch?v=oitT8geMat0 |
Talk Breakdown
1. Why Physical AI? (3–4 mins)
What is Physical AI?
Why is everyone talking about it?
How it differs from Generative AI and traditional robotics.
Why this is the next frontier of AI.
2. The Physical AI Ecosystem (5 mins)
The anatomy of a modern Physical AI system.
Hardware, sensors, actuators, cameras, and compute.
The role of open-source robotics.
Introduction to LeRobot, Reachy Mini, and OpenDuck Mini.
3. Building a Physical AI Robot (8–10 mins)
How I built these robots from scratch.
Hardware assembly and software setup.
Collecting demonstrations using teleoperation.
Training robot policies.
Running inference and deploying models on the robot.
Lessons learned and common pitfalls.
4. Live Robot Demonstrations (8–10 mins)
Live demonstrations using LeRobot, Reachy Mini, and OpenDuck Mini.
Walk through the hardware.
Show teleoperation, perception, and task execution.
Explain what is happening behind the scenes as the robots perform tasks.
5. Getting Started with Physical AI (3–4 mins)
Open-source tools and learning resources.
Recommended hardware for beginners.
A practical roadmap for building your first Physical AI project.
Q&A (time permitting).