Our Mission

Our mission is to enable every person on the planet to be able to program and scale their ideas in the physical world the same way programmers can in the digital world.

This will have two civilization level impacts:

it will drop the marginal cost of goods and physical labor to zero
It will enable us to domesticate space in the next century

To do this we will democratize autonomous robots as the primitives with which to assemble autonomous industries.

Problems

Robots are inaccessible. They cost millions for any serious useful application, installing them requires special knowledge, maintaining them requires special knowledges, updating the system take too long, and on and on.

The future is fat-tailed but robot developers are playing whack-a-mole. They build proprietary software, ML, hardware, and so on, to tackle one niche at a time. This drives up cost and results in inflexibility because businesses that need to automate in a bespoke way (the future of business) have trouble doing so.

The cost of goods and services are, in many cases, rising rather than falling as tech is over-allocated to the internet economy. As information work rises in value, physical labor becomes a weaker alternative over time thus deflating supply and driving up costs. An aging population will only make it increasingly difficult for companies to scale their physical workforces and will further reduce supply.

The Opportunity

The time is ripe for a new paradigm in robotics. A tide is rising at the confluence of foundational ML, cheaper hardware, and growing labor shortages. For the first time fundamental research in embodied AI through technologies like object reconstruction, simulation, and various deep learning techniques has made it possible to build robots that can teach themselves. At the same time hardware costs have declined even as edge-compute power has risen an order of magnitude.

The Company

We imagine a world where everyone can scale their idea in the same way that a programmer can today. Doing physical work will become information processing as the marginal cost of doing physical work drops to zero. Instead of a programmer being able to write a program and deploy it to millions of static robots in a server farm that creates value by processing bits, anyone will be able to record a program that will deploy to millions of dynamic robots. The physical data center will be all around us.

Robots are better than humans at physical labor by orders of magnitude but there are too many barriers to using them.

Our long term goal is to lower the barrier to entry by becoming the Microsoft of robotics. We will provide the operating system for training autonomy on arbitrary hardware. As our robotics fleet grows we gain an experiential flywheel advantage improving quality as we go.

Our near term goal is to enable DIY roboticists to make autonomous robots for X at home. They bring the robot body and we provide the methods to:

pre-train primitive skills e.g. locomotion and manipulation
program specific tasks with a point-and-tell interface

As we provide this new capability to the world the number of hardware and robotic services entrepreneurs and SMBs will explode enabling the fat tail of robotic use cases and driving costs to implement a robotics system down by an order of magnitude.

Today, manual labor compensation is the primary driver of goods and services prices, accounting for ~50% of global GDP (~$42 trillion/yr), but as these robots “join the workforce,” everywhere from factories to farmland, the cost of labor will decrease until it becomes equivalent to the price of renting a robot on-demand from a vendor, facilitating a long-term, holistic reduction in costs.

Over this century, humans will be required less and less in the loop altogether as robots become capable of designing and building other robots — driving prices down even more in a virtuous cycle. This will change our productivity in exciting ways. Higher production will bring an abundance of affordable goods and services, creating wealth for everyone.

Taking this to the limit we imagine a world where the cost of physical labor becomes 0 because it’s outsourced to robots that can build robots for free. In the next 50 years building (many) physical goods will be as easy as generating images or text is now — though it will take longer!

The autonomous industries that are born are chips, composed of millions of robots (smaller chips) that can mine, build, and make more robots. As the time and cost of doing physical work goes to zero — it effectively looks like compute does today — though on larger time scales.

How We Can Do It

A few key insights:

The missing link is the “robot nervous system”. 3.7 billion years of evolution shows us there is no general purpose form factor — but there is general purpose understanding, locomotion, and manipulation of 3D environments so the missing link that is not being served is the robotic nervous system.
The scaling thesis is wrong. Learning is based on diverse experiences that disprove the model — not volume alone.
The super power of robots is distribution and recursion. For a robot there is no reason they cannot be distributed. Unlike humans, a robot can have it’s “arms” on either side of a room. This will enable a recursive composition of robots i.e. where sets of robots (small circuits) can be composed into a super robot (bigger circuits).
A solvable definition of intelligence. Intelligence is the process of mapping a set of sensor inputs to a world model and inferring actuator outputs from simulated futures from that world model. This view will enable a data flywheel across seemingly different robots configurations and lead to increasing generalization.
Data flywheel. Because the hardware we will produce is minimal, and we sell a digital service, this empowers us to be a data company to start, with the option to expand into hardware verticals, similar to Amazon with their vendors.

The challenges:

Building the Autonomy Operating System — the ease of creating high-fidelity digital twins to enable pre-training for zero-shot real2sim transfer will be central
Creating Simple Point and Tell Interfaces — human-in-the-loop “programming” will be key to adoption. We will need experts in imitation learning and reinforcement learning with human feedback paradigms.
Safety & Alignment — collision avoidance, a safety assessment process, and failsafe hardware overrides will need to be baked into the product and required in all robots that we power.

The roadmap:

we will start by enabling DIY autonomous roboticists with simple zero-shot real2sim transfer for point-to-point locomotion piggy backing on systems like Mobile Aloha, Stretch 3, and similar

We believe the best way to do succeed is to ship fast and then keep shipping. There is no need for long timelines or capital investments.

We will do what all paradigm defining platforms have done by starting with the DIY community following in the footsteps of companies like Apple, Oculus, Microsoft, and other paradigm defining companies like them.

Conclusion

We have the potential to enable the world to drive abundance, fundamentally improve lives, and to help enable the next great era of human expansion. Here are the steps to move forward: