embodied artificial intelligence

This is not science fiction, this is Tuesday's factory floor.

Jensen Huang says, "Physical AI is having its ChatGPT moment," and my guest ran the division that would know about that.

He's an expert in robotics, and physical AI just had their strongest VC quarter on record, $16 billion across roughly 500 deals in Q1 alone, apparently.

And the tone has shifted from, "Does this technology work?" to, "Can this be manufactured at scale and sold?" And it's interesting, you know, we've got Hyundai, they've got humanoids in production settings.

Audi and BMW are piloting them on the line.

Actually, it's, it's, has helped other companies to improve their humanoid robots.

And we just, today is the last day of the World Humanoid Robot Games that took place in Beijing for, like, four d- four, uh, days.

So it started August 22nd, finished today, and it was really the visible demonstration of China's strategy around physical AI, as you just mentioned, or what we call embodied AI.

Welcome back to Embedded Executive Live here at Embedded World North America.

I'm Ken Briota, Editor-in-Chief of Embedded Computing Design, and I'm so excited to be here in California for this show and sitting down right now with my good friend Pramit Nandy from Microchip, and we're going to talk a little bit about physical AI.

But before we jump into that, why don't you tell the folks a little bit about yourself and what you do at Microchip?

Sure.

And I want to talk about robotics.

I want to talk about the intersection of robotics.

sort of technologies that are enabling what robotics is becoming because I like thinking of this sort of system of systems rather than talking about physical AI because I think the term is bad.

But the confluence of connectivity, processing, storage, motor controls, automation, all that together is really a hard engineering challenge.

It exists inside massive data centers, surrounded by stable power grids and aggressive cooling systems.

But there is a major engineering effort underway to push this intelligence out of the data center and into what engineers call physical AI.

The goal is to move AI away from just answering text prompts on a screen and put it inside machines that interact directly with the messy, chaotic physical world.

The problem is you cannot rely on the cloud to run a physical machine.

To eliminate that delay, the computing power has to be placed directly on the machine itself.

But you can't exactly strap a delicate, high-powered server rack onto the back of a moving vehicle or a robotic arm.

So making physical AI a reality demands two contradictory engineering feats.

We have to compress an enormous digital brain, and we have to build it an indestructible physical skull.

Uh, Brian's the president of US Operations for GeeMax Robotics.

They are a physical AI company.

They're developing what they describe as terminal plus brain robotics platforms.

Uh, and their initial commercial focus, it includes this really cool bon vivant kitchen robot, uh, through its culinary AI line.

Right.

We've, we've introduced the audience to the principle of embodied AI, where we're basically, you know, taking AI Uh, that learns and acts, you know, through some physical body.

What you're doing with physical AI is entirely, is sort of the next step beyond that.

It's AI that's ultimately designed to operate sort of in the physical world, and ultimately interact back with us.