[the innate way]

[the innate way]

0:00/1:34

What I listen to when I think about our future

Simplicity always wins


TL;DR: We are destroying the barriers to entry in robotics by making the first robot everyone can actually teach easily. A massive data collection will ensue to enable an AGI for robotics that thrives on data diversity the most importantly.



It has never been that simple to build robots


During the past 2 years, the barriers to entry to building in robotics have drastically lowered. This is in the major part due to the AI wave of course. It has made learning robotics much simpler, and allowed for much simpler solutions using large models that understand the world much better.


It has never been that simple to build robots. It's a good thing, and the impact of it is that the amount of builders in the field is growing. One can look at the number of stars on LeRobot's github repository, or the amount of users in K-Scale Labs' discord. Every week on X, someone reveals their new surprisingly smart robot and brings other people in. New robotics hackathons pop more frequently in Silicon Valley, and they are filled with young people and software engineers who are discovering the field and learning fast.


It has never been that simple to build robots. And yet, it's still not simple enough. It still takes an enormous amount of knowledge to work on the complete stack: on localization that works without spending hours on configuration, on manipulation that needs fast and working inference, on easy data collection, on a software architecture that does not break every 2 days, and on a working, affordable hardware.


The secret to bringing folks like myself (a year ago) in robotics is NOT to make bigger robots with 36,000 degrees of freedom, it is to just make it simpler to build something smart that works. That's it.

The Apple I, the first PC that was accessible and functional


The Apple I, the first PC that was accessible and functional


The winners have always made things simple


In the early days of personal computers, everyone thought mainframes were going to be a thing. PCs were this fun thing that hobbyists built in their garage that didn't really provide business value. Until some folks in Palo Alto made it simpler to develop on a platform that just works. The year was 1976, and the Apple I was born. During that year, Steve and Steve sold 500 of these computers to mostly hobbyists, and the reason many were impressed in that population was because it simply worked and was so expandable.


From this success, they made the Apple II, a more widely accessible computer built on everything they learned from the Apple I. It sold to millions of units, and Apple went public 3 years after.


We're now 50 years later, and once again a new hardware x software revolution is happening in Silicon Valley: Robotics. And again, everyone is racing towards adding more complexity, not simplifying. Why? Because making complex things simple take dedication and time. It is much easier to strap together pieces of hardware to make a humanoid form factor, than to make a small robot that learns fast, but the first one seems more impressive.


At the end of the day, the platform that wins will be, like history has many times proven, the most accessible one. The one people will be able to look at and feel "this was the first robot that really, innately understood the world".

To make robot models that work, what matters is data diversity

To make robot models that work,

what matters is data diversity

Crazy upsides: Should we win, we will also win Physical AGI


The reward for success will be even greater than Apple. We are not talking only spreading a product to millions of units and revolutionizing the way we interact with technology once again. We are talking unlocking a massive data collection method for society.


In a world were robots are easy to train and teach everywhere, data is the new app. Everyone will be able to share and get rewarded for their data, especially if it gets used in AGI models that power all robots of the future.


This is our final goal: When everyone can collect all kinds of data on effortlessly teachable platform, we collect exactly the kind of diversity required to generalize AI models for robotics, not just for manipulation, but also interaction, world modeling, everything.


This is how we get to Physical AGI.

Maurice, our first robot, a physical AI agent that works

Maurice, our first robot,

a physical AI agent that works

Robots of today and tomorrow


Maurice, our first robot, is but the first step to mass-democratizing our technology. Bigger robots are already roaming our house in Palo Alto, and we're taking great care to their design.


There is a lot to invent, and we will do it by staying close to people and staying opinionated. We make accessible, innately intelligent systems that you can use today. We stand for every builder that wants to join the robotics revolution.


It will only take a year for our first systems to autonomously interact with humans in homes like ChatGPT does today, but with full physical capability.


By 2027, our robots will have full capability in your home, leveraging training from all humans, doing tasks the way you want. From doing dishes to laundry, from securing your place to being a companion for your kids and alders, all the while being a natural extension of your computer and phone.


By 2030, this knowledge will be used in the real world. Innate robots will be helping wherever humans are too few or too busy. They will start helping repairing bridges before they falter, replanting trees where we need them, delivering aid in disaster-stricken areas. They will help us increase energy generation safely, streamline our manufacturing lines, and explore the universe.


All of it will be enabled because everyone will be a part of it. Because, in our future, robots will have been taught by the people, for the people, and the benefit of all.


– Axel & Vig

innate inc.

Made with 💙 in Palo Alto