Physical Address
304 North Cardinal St.
Dorchester Center, MA 02124
In 2026, we see robotics advancing by leaps and bounds with a marked improvement in dexterity, the kind of progress long needed in the search for truly useful home assistants. Now a new AI model has arrived to operate robots through activities, including… Foldable laundryBuild chests, repair other robots, and even fill wallets with flimsy paper money.
Earlier this month, a California-based company Generalist AI released the first generationnew Physical artificial intelligence A model that makes robots capable of successfully performing all these tasks (and more). It’s a big step forward for robots designed for the real world based on intelligence born from the real world, Pete Florence, co-founder and CEO of Generalist AI, told me.
In most of the mock-up videos posted by the company, the Gen-1 is shown working on a pair of robotic arms, but that’s not all it’s designed for. “The Gen-1 is designed to be the brain of any robot, which means the same model could work on a human, a prosthetic arm or other robotic systems,” Florence said.
Indeed, this has been proven to be the case Breakthrough year for general purpose robotswith companies including Boston Dynamics and honor Uncovering advanced robots capable of performing supernatural movements. The robot market is expected to explode with one Estimates from Morgan Stanley Expect to grow into a $5 trillion market by 2050. Forecasts point to robots coming into industrial, retail, hospitality and care environments before eventually landing in our homes. To get there, we need to see more progress in AI.
Training robots to live alongside humans
Over the past few years we have seen major linguistic models, e.g ChatGPT, Gemini and Claudedeveloping at lightning speed. The same has not been true for the physical AI models required to operate robots, largely due to a lack of data to train those models on. Robots — especially humanoid robots — must learn how to navigate a world designed for humans just as humans do.
This data is often collected from robots performing tasks while being operated remotely by humans, but not by the first generation. Instead, the dataset used to train AGI models was collected by humans who completed millions of different tasks using wearable technology.
“We built our own lightweight ‘data hands’ and distributed them globally to learn how people actually interact with objects, with all the precise force feedback, haptic feel, glides, corrections and recoveries that define real-world human dexterity,” Florence said. “This type of data is critical for teaching robots common sense, intuitive understanding, and the ability to adapt in real time rather than executing strict instructions.”
Generalist AI has released a series of videos showing the robotics model repeatedly performing a range of different tasks, perhaps the most convincing of which involves the robot withdrawing cash from a wallet before reinserting it into the same pocket. This is a trivial task that many people struggle with. Obviously, it’s not easy for the robot either, given the fragility of paper money and the fabric of the wallet – yet it completes the task.
Another video shows a robot that sorts socks by color, folds them into neat piles, and counts the number of pairs using a touch screen. Other challenging tasks the model can complete include unzipping and filling a pencil case with pens, stacking oranges into a neat pyramid and connecting an Ethernet cable.
These videos show the breadth of the Gen-1’s abilities, but what’s most impressive is the success rate at which he can complete certain tasks. Generalist AI measured the success rate of the model against the previous version and found that the Gen-1 could successfully serve Robotic vacuum cleaner Fold boxes in 99% of cases (up from 50% for Gen 0), fold boxes in 99% of cases (up from 81% of Gen 0) and wrap phones in 99% of cases (up from 62% of Gen 0).
Robots improve
Most robots are programmed to complete a task in a specific and organized manner. But what happens when a curveball is thrown? “The smallest changes in the environment can cause failure,” Florence said.
An important skill that robots need, and that humans naturally possess, is the ability to think on their feet. That’s why Gen-1 was designed with improvisation in mind so he can come up with strategies to complete tasks. Florence gave me an example of a robot that uses two hands to reposition an inappropriate part for a car task, even though it has only been trained to use one.
“This kind of creativity has been largely absent from robots until now,” he said.
There is still significant work to be done when it comes to enhancing the capabilities of improvised robots, but early progress shows glimpses of a positive impact on both reliability and speed, Florence says. “We’re starting to see real progress and we’re excited to push the boundaries of embodied intelligence.”
After all, there may come a day when you need a robot in your home that can repair all the other little robots.