Managing Editor’s Note: Fear and uncertainty have once again struck the markets, and as Jeff has been predicting, the short-term volatility is far from behind us.
Today it’s tariffs, and who’s to say what will shock the markets next? Fortunately, Jeff’s got a winning strategy to take advantage of this market turbulence.
He’ll get into all the details on Wednesday, May 7, at 1 p.m. ET. Make sure you go here to automatically add your name to the guest list.
The “ah-ha moment” isn’t far off now…
For many consumers, that “ah-ha moment” for artificial intelligence (AI) will come when they experience some form of manifested AI that seems to magically solve some kind of undesirable real-world problem for them.
Of course, anyone who has ridden in a Tesla running version 13 of its full self-driving (FSD) software has had that kind of moment already.
If anyone has a chance to experience it, I strongly recommend doing so. I promise, it will be like a lightbulb turned on in a dark room. Seeing is believing, after all. And witnessing the technology at work makes it click in a way that simply reading about it never could.
But many don’t have access to a Tesla running FSD, and some will never be comfortable letting the car drive itself.
But an intelligent autonomous robot designed to clean the house and perform undesirable cleaning tasks on a moment’s notice, with a single voice command?
Yeah, everyone will want one of those.
Of course, an intelligent autonomous humanoid robot like Tesla’s Optimus will perform these tasks, and a whole lot more, but it won’t be cheap. The initial target price is $25,000, making it at the very high end of luxury consumer electronics products.
But as with most markets, there will be a range of products available to consumers. At the lower end, there will be task-specific autonomous robots like the iRobot (IRBT), famous for its Roomba robotic vacuum cleaners.
In the last couple of weeks, we got a look at what that might look like from a manifested AI company, Physical Intelligence, and its latest vision-language-action (VLA) model π0.5.
π0.5 is embodied in an autonomous robot on wheels that provides a stable foundation for the robot’s two “arms” to perform a wide range of tasks that have great utility around a home.
Below is a short video of the robot responding to a request to clean up and place the dishes in the sink.
The video itself has been sped up 6X as the robot is not nearly as fast as a human…. yet. But the reality is that tasks like this are far less time sensitive. We’re happy to just give the command, walk away, and enjoy our newfound free time while the autonomous robot completes the work in its own time.
Source: Physical Intelligence
Physical Intelligence has taken an approach to its vision-language-action (VLA) model similar to what Tesla has done for its self-driving software and its Optimus robots. It is designing its AI for generalization, not for individual tasks in a controlled environment.
This approach is exactly what is necessary for robots to operate autonomously in the real world.
It doesn’t matter whether or not it is a self-driving robot or a robot designed for the home. To achieve scale and general utility, intelligent robots need to be able to adapt to whatever environment they are working in.
Every home is different. Most of them have the same kinds of appliances and furniture, but they are in different locations and of various types. Autonomous robots need to have general knowledge about the kinds of things that they’d experience in a home and the specific skills to be able to manipulate objects in the home to complete their assigned tasks.
Physical Intelligence collected some interesting and quite surprising data on its training process for generalization. Below we can see the skill level achieved by the π0.5 model in a controlled environment (i.e., just training in one home environment where all the objects are in the same place). This is shown as the green line and is used as a baseline for performance.
Source: Physical Intelligence
And shown in yellow above is a graph of how the π0.5 model’s performance improves with an increase in training in new home environments.
What the chart above shows is that at around 100 different home training environments, the π0.5 model outperforms the training from a single controlled environment.
That’s a lot less than I expected. Said another way, the π0.5 model is learning quickly, suggesting a clear path towards human-level skills in any household.
Think about it. If we just extrapolate a little bit, it’s easy to imagine that by 500 or 1,000 unique home environments to train on, the π0.5 model would approach human-level adaptability and skill (note: the learning curve isn’t a straight line, the last 5% is always the hardest).
The end goal is simple: for Physical Intelligence to deploy an out-of-the-box robot into any new home.
Need to have the robot put away all of the items on the kitchen counter? Done.
Source: Physical Intelligence
Need the bedroom to be cleaned? No problem.
Source: Physical Intelligence
Or would you like the counters in the kitchen to be cleaned? It might take a few minutes, but it will get done without you lifting a finger.
Source: Physical Intelligence
Just imagine being able to leave the house in the morning, give your home robot the commands to “clean up the kitchen, the bedrooms, and tidy up the family room before we get home around 5 this evening,” and to have the house in perfect condition when you get home.
Like magic.
Except it’s not. It’s software and hardware engineering on an accelerated technological path of improvement.
Consider this: Physical Intelligence was founded in the first quarter of last year with $70 million. It now has working prototypes of what has the potential to become an affordable home robot, likely priced around $5,000–7,000 by my estimation. Some households pay close to that in a year to have a person come clean just twice a month.
This has the potential to become a blockbuster product, and there is a major backer of Physical Intelligence that would make for an obvious acquirer.
Amazon (AMZN) invested in the seed round. Then Jeff Bezos himself invested in the Series A deal alongside his family office, Bezos Expeditions.
Amazon has long wanted to get into the home robotics space. Some might remember the Amazon Astro, which is designed for home monitoring, installed with Alexa, integrated with Ring (another Amazon acquisition, which it bought for $1 billion in 2018), and sells for $1,599.99.
Amazon’s Astro |Source: Amazon
Astro hasn’t gone very far, and you have to “request an invitation” to purchase one. It is expensive with limited utility, not exactly mass market potential.
And Amazon attempted to acquire iRobot in 2022 for $1.7 billion, but the deal fell through for both companies due to antitrust issues in the EU.
The deal was dropped in early 2024 for that reason, and it turned out to be disastrous for iRobot (IRBT). iRobot’s share price has declined by more than 98% from its 2021 all-time high.
But Amazon and Bezos clearly haven’t dropped their interest in a major play in home robotics.
And Physical Intelligence just might be the solution to what they’re looking for… and what every homeowner has dreamed about for hundreds of years.
Jeff
The Bleeding Edge is the only free newsletter that delivers daily insights and information from the high-tech world as well as topics and trends relevant to investments.
The Bleeding Edge is the only free newsletter that delivers daily insights and information from the high-tech world as well as topics and trends relevant to investments.
