Why open source is the ‘missing brick’ for robot adoption

This article summarizes an episode of Sequoia Capital’s video series featuring Thomas Wolf, co-founder of Hugging Face.

Thomas Wolf, co-founder and Chief Scientist Officer of Hugging Face / Photo credit: WeCNLP

As co-founder of Hugging Face, a platform built on an open-source community for AI models, Thomas Wolf believes the only way to solve problems in robotics is to make the technology open for everyone. If anyone can become a creator, he argues, the field can move forward.

Why every coder could soon be a robot builder

Robotics is at a turning point. The challenge is making the software easy enough for a new wave of creators to use, rather than building better machines.

It’s happened before in AI
For Wolf, this moment feels a lot like the early days of AI language models.

He says, “I have a spidey-sense that we’re at the same moment for robotics today that we were for transformers and language models a handful of years ago. Hardware was already there… but the missing brick was really software that could adapt, that could be dynamic.”

The goal is to grow the community
Wolf says, “the question was could you move this tiny vertical [of robotics] to a full horizontal thing… there’s this very smooth transition where the 100-200 million software developer are becoming AI aware, and I think there is a potential future transition where all of these people also become roboticist.”

Step two: Make the robots cheaper

Once the software is easy to use, the next problem is the price. Wolf argues the machines need to be cheap enough for people to experiment and play with, not just for big companies to use in factories.

The “impulse buy” robot
He points to one of his company’s own products, a robot designed to be affordable enough for anyone to try.

Wolf says, “Reachy Mini for instance is priced at $300, that’s something that can definitely be like an impulsive buy. You buy it as a gift and you’re not sure if it’s going to work or not. But for this price, what we want to find is… a lot of potential in more like entertainment, fun, education.”

An “App Store” for robots
This low price could create a new market, much like the one on your phone.

He says his vision is to make a robot “almost like a smartphone. It comes, you have a couple of behaviors. But just because you can tweak it and people can build new behavior and share them… the possibility are kind of endless. It’s kind of a open door on rebuilding the app store of iPhone.”

The real-world gap in robotics

Getting more people involved also helps solve the main technical hurdle in robotics: data. While AI language models can learn from the internet, robots only learn from the specific places they are trained.

Why a robot fails in a new room
Wolf explains, “the main limitation is you can record a lot of task yourself but usually what you will lack a lot is the diversity. So you will basically be able to train a robot to do something very well in your room… but once you put it in the next door room where maybe the walls are green instead of red the robots has a lot of troubles to generalize.”

Why China is unexpectedly leading the way