By 2026, we see robots advancing in leaps and bounds with dramatically improved intelligence, the kind of progress that is long overdue in the search for truly useful home assistants. Now a new model of AI has arrived to empower robots with tasks, including in the folding laundrybuilding boxes, repairing other robots and even filling wallets with loose paper money.
Earlier this month, California-based company Generalist AI released Gen-1, a new physically the AI a model that enables robots to perform all these tasks (and more) effectively. It’s a big step forward in terms of robots designed for the real world based on intelligence from the real world, Pete Florence, founder and CEO of Generalist AI told me.
In several prototype videos published by the company, the Gen-1 is seen working on robotic arms, but that’s not all it’s built for. “Gen-1 is designed to be the brain of any robot, meaning the same model can be used in a humanoid, an industrial arm or other robotic systems,” said Florence.
Already, this has proven that a breakthrough year for humanoid general purpose robotsand companies that include Boston Dynamics again Fame revealing sharp bots capable of human-like movements. The robotics market is expected to explode, with one estimate from Morgan Stanley predicting growth to a $5 trillion market by 2050. Forecasts see robots coming through the industrial, retail, hospitality and care sectors before eventually reaching our homes. To get there, we need to see more advances in AI.
Training robots to live alongside humans
In the last few years we have seen some big language varieties, such as ChatGPT, Gemini and Claudechange at lightning speed. The same has not happened with the physical AI models needed to make robots work, in large part due to a lack of data to train those models. Robots — and especially humanoid robots — must learn to navigate the human-made world as a human would.
Usually this data is collected from robots that perform tasks while being used by humans, but not Gen-1. Instead, the dataset used to train Generalist AI models is collected from people completing millions of different tasks using wearable technology.
“We built our own lightweight ‘data hands’ and distributed them around the world to learn how people interact with things, with all the subtle energy feedback, tactile sensation, smoothness, adjustment and recovery that defines human intelligence in the real world,” says Florence. “That kind of data is critical to teaching robots common sense, intuitive understanding and the ability to adapt in real time rather than issuing rigid instructions.”
Generalist AI released a series of videos showing the model working on robots that repeatedly perform a range of different tasks, the most compelling, perhaps, being a robot that takes money out of a wallet before returning it to the same pocket. This is a difficult task that many people struggle with. It’s obviously not easy with a robot, either, given the dullness of paper money and the fabric of a wallet — it gets the job done, though.
Another video shows a robot sorting socks by color, folding them into neat piles and counting the number of pairs using a touch screen. Other tricky tasks the model can complete include unzipping and filling a pencil case with pens, stacking oranges in a neat pyramid and connecting an Ethernet cable.
These videos show the breadth of Gen-1’s capabilities, but what’s most impressive is the level of success with which it can complete certain tasks. Generalist AI measured the model’s hit rate against the previous version and found that Gen-1 can effectively serve a robot vacuum cleaner in 99% of cases (up from 50% in Gen-0), wrap boxes in 99% of cases (up from 81% in Gen-0) and wrap phones in 99% of cases (up from 62% in Gen-0).
Robots make improvements
Most robots are designed to complete a task in a precise and orderly manner. But what happens when a curve ball is thrown? “Very small changes in the environment can cause failure,” says Florence.
An important skill that robots need, which humans naturally lack, is the ability to think on their feet. That’s why Gen-1 is designed with development in mind to come up with strategies to complete tasks. Florence gives me the example of a robot that uses two hands to fix a misplaced part in a car job, even though it’s only trained to use one.
“This kind of innovation has been largely absent from robotics until now,” he said.
A lot of work still needs to be done when it comes to building better robots, but early progress shows glimpses of the positive impact on both reliability and speed, Florence said. “We’re starting to see real progress and we’re excited to push the boundaries of synthetic intelligence.”
After all, there may come a day when you need a robot in your house that can fix all your other small robots.
