| Dear Reader,
I have a confession to make…
I like playing video games.
Of course, with a busy schedule and two young daughters, I haven’t had the time to play them in years.
But if you’ve ever played one, you understand how powerful a simulated world can be.
The physics in video games often mimics the physics of our own world. You press a button, your character jumps and it's pulled back down by gravity.
But sometimes these simulated environments don’t exactly match real-world physics, and you can end up with some pretty hilarious results.  So what happens when artificial intelligence learns to play in those same simulated worlds, then uses that experience to move real robots?
We know the answer because it’s happening right now…
And it's accelerating the development of humanoid robots in a big way. | 
A secret meeting just took place in Las Vegas.
Inside exclusive sky suites at the Aria Resort, Wall Street executives gathered behind closed doors.
They aren't warning the public. They're positioning themselves for what's coming.
By the time you see the headlines, it will be too late. But there's still time — if you act now.
Watch this controversial video now, and send it to everyone you know. | What’s Real Anymore? One of the biggest challenges in robotics is the disconnect between what works in simulation and what works in the real world.
Naturally, simulations are faster and cheaper than building hardware.
But they can often be too perfect.
Real life is messier. Objects aren’t always where they should be and flat surfaces aren’t necessarily flat.
That’s why there used to be a strongly held belief in the tech community that training robots in simulated environments would never be as effective as training them in the real world.
But AI is turning that belief on its head.
Recent advances in something called “sim-to-real” learning are closing the gap between simulated environments and the real world.
One approach, known as ASAP (Aligning Simulation and Real Physics), trains robots in simulation first, then fine-tunes them using real-world data.
It even uses something called a delta action model to account for all the little differences between the digital and physical worlds.
Tracking those differences is crucial because it helps robots learn faster and more efficiently than ever before.
Instead of spending months tweaking a robot's behavior in the lab, researchers can now simulate thousands of test scenarios using software in mere minutes.
Then they can refine the robot’s behavior in the real world with just a few adjustments.
A great example of how effective this process can be is the Unitree G1 humanoid robot.  Source: Unitree Researchers using the ASAP system taught this robot to perform agile, whole-body motions like the one pictured above.
This kind of movement used to either be clunky, slow or flat-out impossible with traditional training methods.
But the ASAP system is helping Unitree robots start to move more like people.
They can balance, jump, walk and adjust their movements on the fly... just like us.
And human-like agility will become especially important as we develop commercial humanoid robots for the home, and for things like elder care and companionship.
But movement isn’t the only thing being improved by AI-powered simulations.
These simulated worlds are also helping robots understand things like weight, momentum and energy conservation.
I believe that the ability to learn the laws of physics from data is one of the most exciting trends in robotics.
AI-powered simulations are being used to improve robotic arms and legged mobility.
It lets robots make real-time decisions about how much force to use or how to carry objects without dropping them.
And it’s improving self-balancing systems to help keep humanoid robots from tipping over.
Ultimately, we’re finding out that simulated environments can do a shockingly good job of teaching a machine how to understand the real world.
And the effectiveness of this approach has surprised many researchers.
But the benefits go beyond robotics.
Researchers are using AI-enhanced simulations in fields like materials science, drug development and fluid dynamics to model complex systems with more accuracy and less computational cost.
So with all this progress, does it mean that humanoid robots are about to become as common as the Roomba?
Well, there’s a catch… Are Robots Screwed? You see, there’s a surprisingly real-world problem that’s slowing down the development of robots today…
A screw.
More specifically, a planetary roller screw.
Planetary roller screws convert rotational motion into linear motion. They’re used in robot joints — like knees or elbows — to move limbs in and out.
And this tiny, unassuming mechanical part is becoming a linchpin for the future of robotics.
Older ball screws have been used in robot joints for decades. But robotic engineers have found that planetary roller screws are stronger, more precise and are better able to handle the wear-and-tear that comes with humanoid robots doing real work.
Simply put, these screws help robots move more like humans without breaking down.
Tesla’s Optimus robot uses four of them in its calves alone. Other leading robot makers, like Figure AI and Agility, rely heavily on them too.
And Morgan Stanley predicts these screws will eventually replace traditional ball screws in most humanoid robots.
They will be essential for robots to work in warehouses, factories and hospitals without constantly breaking down. And if you ever welcome a humanoid robot into your own home, you’ll want your robot to be built with them too.
But there are a couple of major problems with these planetary roller screws.
The first is that they are expensive.
Each one can cost between $1,350 and $2,700. And a single humanoid robot might need 40 or more of them. That adds up fast.
Even worse? Only a handful of manufacturers in the world can make them at scale.
Unfortunately for us, most of them are in China.
The U.S. supply chain for these screws doesn’t really exist yet.
Experts like Scott Walter, chief technical advisor at Visual Components, warn that even when actuators are assembled domestically, the core components still often come from China.
He told Fast Company: “"Planetary roller screws are precision equipment. Few companies have the expertise and equipment to produce vital components outside of China.”
That means China has a big advantage when it comes to humanoid robot production.
In fact, Shanghai Beite Technology just committed $260 million to build a new plant dedicated to planetary roller screws.
But here in the U.S., demand is rapidly growing, yet we don’t have the designs locked down to build these essential components.
And with the recent escalation of a trade war with China, this could soon become a huge problem. Here’s My Take I believe most people are going to be shocked at how fast the Botcom Revolution is coming.
Morgan Stanley estimates the humanoid robot population will reach 40,000 by 2030, and it will grow to 63 million by 2050.  But if the robotics boom arrives sooner than expected — which seems more likely every month — the U.S. could find itself behind the eight ball.
Because AI is doing an incredible job teaching robots how to move and interact with the world using simulated environments.
It’s accelerating development in ways few expected just a few years ago.
But until we solve the very real, very physical problem of sourcing planetary roller screws, our Botcom Revolution might happen years behind China’s.
Regards, 
Ian King
Chief Strategist, Banyan Hill Publishing
Editor’s Note: We'd love to hear from you!
If you want to share your thoughts or suggestions about the Daily Disruptor, or if there are any specific topics you’d like us to cover, just send an email to dailydisruptor@banyanhill.com.
Don’t worry, we won’t reveal your full name in the event we publish a response. So feel free to comment away! |
Post a Comment
Post a Comment