All your robots are belong to us
For the last months, I've been freelancing to save up enough money to freely ideate for a year. April marks the end of this period. I've saved up approximately 40.000€, which should last me for approximately a year.
In recent months, one particular topic has been capturing my attention: robotics. I've been watching a lot of demos from Physical Intelligence, DeepRobotics, and Unitree in particular. It is clear that it is only a matter of when, not if autonomous robotics will arrive. And like most technical innovations, it will happen very slowly and then all at once.
Watching the demos, I noticed how some are American, most are Chinese, and virtually none are European. This is alarming. Drones are inflicting 70% of all casualties in the Ukraine war, and it's not far-fetched to picture a machine gun being screwed onto Unitree's quadrupeds, which sell at 2,700$. But the more immediate effects will be the economic impacts of widespread robotic automation. Just like LLMs have created a huge productivity boost for those using them, so will robots. Most people still aren't aware of what you can do with chatGPT because they're not digital native and aren't in the social bubble where they can learn from friends. Just like we have huge gaps between individuals in their use of LLMs, we will have large differences in the use of robotics, but across nations. The Chinese are already living in the future. Europe is still many years away from it.
What's changing is the software, not the hardware
What particularly sparked my interest was reading about foundation models for robotics. ML is my area of expertise and my research in tabular foundation models meant that I have a reduced reverence for these kind of trainings. I also spent last year building a fleet of raspberry pis to supervise pizza ghost kitches and this had broken down the initial barrier to working with hardware.
I wondered how hard it would be to build in robotics. The single biggest question was whether robotics is more like computational genomics where you need a profound domain knowledge to ask the interesting questions or whether its more like quant finance where mathematicians without domain knowledge are dominating the field.
On the 11th of April I flew to the Hugging Face robotics hackathon in Paris to learn more. I was positively surprised by how the problem we were trying to solve (building scribble.io
Other thoughts not tied into this essay yet:
When I think about what kind of company I want to build, I think about what I want to be doing for the next 10+ years. I want to build something exciting, something that lights up the eyes of others when they see it. I want to be technically challenged every week and be required to read papers to stay ahead. I also want to work with other smart people. Robotics naturally attracts these people and for another, it's also easier to convince smart people to join you if you're building robots over a SaaS for accounting.
Main hypotheses:
- There will be an application layer for robotics (elaborate on what this means)