The Robots are coming: Genesis
2025 might open up a new paradigm for RL learning in robotics. Are we near an inflection point for humanoid general purpose robots? 🤖
AI Supremacy is giving out 3-month free complimentary paid subscriptions to our top sharers and restackers every two weeks. The cost of a paid subscription is less than $2 a week but to make it more affordable for my 🌎 international readers, consider sharing: 🙏 🇮🇳, 🇳🇬, 🇵🇰, 🇵🇭, 🇧🇷, 🇮🇩, 🇲🇽, 🇹🇼, 🇮🇷, 🇧🇷, 🇪🇬, 🇲🇽, 🇨🇴 + 🤗
Hey Everyone,
Let’s talk about robots and the humanoid robots today.
Reinforcement learning in 2025 might make Robots more maneuverable than ever.
“2024 is the year of prompt engineering.
2025 is the year of reward engineering.”- Jim Fan, December 24th, 2024.
Robot maneuverability is improving - 1:46
The Unitree stunt videos are fairly entertaining lately. They display robots with more advanced maneuverability in the spirit of Boston Dynamics.
What is Genesis?
In December, 2024 an important project was announced. Genesis project — after a 24-month large-scale research collaboration involving over 20 research labs — a generative physics engine able to generate 4D dynamical worlds powered by a physics simulation platform designed for general-purpose robotics and physical AI applications.
This is a new framework for training robots.
Put simply, this is an open-source physics engine that allows fast creation of 4D environments for robotics and AI.
It will accelerate robotics innovation, including in military scenarios. The massive research collab is highly likely to turn into several exciting startups.
I cannot overstate enough how important Genesis is as the result of a team of 20 researchers unveiled Genesis (mostly Chinese researchers trained in top tier Universities), the result is a framework that is useful.
If an AI can control 1,000 robots to perform 1 million skills in 1 billion different simulations, then it may "just work" in our real world. - Jim Fan (Nvidia)
Genesis Video: 3:26
Back in August, Morgan Stanley projected that the number of U.S. humanoid robots could reach 63 million units by 2050 with a $3 trillion impact to wages. Put another way, they estimates the humanoid robot population will be 40,000 by 2030 and swell to 63 million by 2050. Clearly the 2030s are going to be a pivotal decade for the 21st century’s moving into the Machine economy.
It’s not yet clear to me if humanoid (general purpose robots) are a bigger market than the Autonomous vehicle sector (or when that shift occurs) although obviously they have more potential utility in the smart cities of the future including eventually consumer home robotics.
I asked the super talented Diana Wolf Torres (one of my favorite voices in robotics) if she could summarize a Robotics report from Peter Diamandis and full disclosure, he’s an investor in Figure AI. Diana runs a Newsletter her too called Deep Learning With The Wolf.
Deep Learning With The Wolf
Figure AI and Tesla are the likely duopoly from the U.S. in humanoid robotics at least so far as it looks in 2024. Although eventually I expect Nvidia, Amazon, Apple and Google to have their own offerings as well.
Advances in Generative AI are mostly credited with a faster acceleration in robotics innovation that will likely have a nascent stage in the 2024 to 2028 period.
“Breakthroughs in generative AI are bringing 3D perception, control, skill planning and intelligence to robots,” said Rev Lebaredian, Nvidia’s vice president of omniverse and simulation technology.
I’ve written dozens of articles in my time blogging about the future of Robots in society, but this time it definately feels different. And I want you as a reader to think seriously about the future market for robots.
Source: CNBC.
Prediction: Nvidia will position itself for Robotics more in 2025
Audio breakdown: 5 minutes 34 seconds. Read “Nvidia bets on robotics to drive future growth”1 (FT).
Selections from Deep Learning With The Wolf Newsletter
Diana’s work can also be found on LinkedIn in the Newsletter Deep Learning Daily. Here are articles that I found highly worth reading:
The Impossible Task of Trying To Sum Up 2024- The Year of AI
Tesla's Autopilot: A Glitch in the Matrix?
OpenAI's O3 Models: A Leap Toward AGI or a Cautious Step?
From 1-800-ChatGPT to Google's Deep Research: AI Goes Big on Accessibility and Innovation
I consider Diana Wolf a subject matter expert when it comes to the Robotics ecosystem and future of humanoid robotics, the topic of her deep dive today.
As per my opinion, I estimate that 2025 has around a 65% probability to be an inflection year for the future of robotics. OpenAI has had internal discussions about building a humanoid robot. It (actually Sam Altman) has invested in Figure, 1X and Physical Intelligence. OpenAI acquiring Figure remains a distinct possibility in my opinion in the 2020s.
I came across a report on Robotics by Peter Diamondis who is the cofounder and executive chairman of Singularity University, among other things and who is also an investor in Figure 2. I asked Diane if she would be willing to summarize the key findings of the report as she’s very familiar with the topic.
Jim Fan is so entertaining:
“We are the generation of “robot immigrants”, en route to a new world of ubiquitous Physical AI, much like our parents are “digital immigrants”, learning to realign their lives on 6 inches of touch screen.” - Source.
In this article Diana analyzes the report and gives you one of the most thorough overviews of the context of Humanoid robotics you’ll ever read.
Including: what is the value proposition, comparing projections on number of humanoid robots in the next 15 years, who are the first-movers (U.S. and China), why battery tech is so important, and the real-world use cases in some detail, manufacturing costs and spects, etc…
This one is truly for the robot nerds out there and one of the best articles on the topic I’ve ever read.
The Rise of Humanoid Robots: How 16 Companies Are Racing to Transform Our World by 2040
By
, December, 2024.Keep reading with a 7-day free trial
Subscribe to AI Supremacy to keep reading this post and get 7 days of free access to the full post archives.
 | A guest post by
|