Brian Gerkey

That has been just revolutionized by AI.

And we at Intrinsic have built what we call the Intrinsic Vision Model, which is a foundation model based on neural networks that we pre-trained it on a huge volume of data, some real, some synthetic.

And what we found is that it's producing really, really good results for answering the particular question of if I give it a camera image and I give it a CAD model for the object I want to find in that image, even if we never trained on that CAD model, we never saw it before.

We can find it in the scene with incredibly high precision, and then we can go pick it up.

And that turns out to be, that's a capability that's really, really important.

If you go back to that, you know, what I was saying earlier about these small manufacturing operations, if you're a business and you're, what you're doing is in the morning, you get a, you decide to run a job and you're going to manufacture a hundred of something, some kind of widget.

You've got the CAD for it because you're manufacturing it, but you don't have time to, you know, custom train any, any part of your automation system.

Yeah, that's a great question.

Honestly, what happens right now is you just don't use automation.

There's some threshold where...

It's, it's just not, it's not worth your time to, to train it, to customize it.

And so you just, you just don't, you're not able to use automation, which then means, well, okay, now I've got to, I've got to find people to do it, which can be tricky depending on the, you know, where you are.

It might be difficult to find people with the right training or people who are willing to work a third shift.

They're, they're constrained on that side.

And so what that might mean for a small business is you might just not be able to take that job.

It might mean that you've got customer demand, but you don't have the bandwidth to fulfill it.

And so you can, there's a productivity gap that we think we can close there.

So what the vision model does then is that it allows you to basically learn almost any object that you want to manipulate, give it a camera edge, we see where it is, now we can pick it up.

And what that means is that you can provide, put in front of the robot, a collection of objects, and it can reliably pick them up and then do whatever you need to do, whether that's

pack them or put them into a machine.