How Humanoid Robot Neo is Learning to do Laundry (and More)

Humanoid Robot NEO has a new AI model that's unlocking new capabilities as it's being tested

in the homes of ONE X employees. The robot's new abilities include picking up unfamiliar objects in unfamiliar settings

whole body control, improved coordination, running, multi-directional walking, stairs, and more. NEO's new Stair Mode uses the robot's onboard cameras to estimate the height of the surrounding

floor without the use of LiDAR sensors which are commonly used by competing robotics

companies for the same purpose. Video from One X also shows Neo stepping sideways downstairs

stumbling a bit over the last step, and catching itself. To perform tasks that require Neo to get

lower to the ground, One X showed the robot sitting, kneeling, and standing up from those positions

including an example of handing someone a tool, working under a car, and crouching down to put a

piece of clothing in the washing machine. The robot has also shown pivoting and walking backwards

as well as sidestepping, which could be useful in navigating tight spaces often found in homes

These demos are part of 1X's rollout of its new Redwood AI, which the company says runs completely on Neo's onboard GPU

The company also shared updates on its 1X world model, which is designed to help bridge the gap between the robot's training simulations and real-world action

The model is trained on video inputs observations from the robot and input trajectory allowing the robot to simulate various potential future actions from a given starting point Examples on One X website show this in action

where a slider moves between generated imagery and real-world imagery from the robot's POV

doing laundry, opening a door, and cleaning. One X also showed off new details about how

its robots learn to do new tasks. In the example of opening an air fryer, the company shows how

the AI model begins by hallucinating that the air fryer basket is inseparable from the whole of the

air fryer. After the robot has had some experience with the air fryer, the data gathered is then added

to the mix, and the simulated air fryer interaction becomes much closer to the real thing. Like humans

1X's robots also learn from their failures, some of which 1X shared. The company also acknowledged

that the 1X world model struggles with simulating unfamiliar held-out objects like this plant

and predicting locomotion such as walking or shuffling. While a more general AI model that can power robots to accomplish a wide variety of tasks with minimal training

is the holy grail that most robotics companies are aiming for, OneX says it will take time to get there

and focusing on a variety of task-specific training goals will help make it happen

We met OneX's Neo Gamma robot in person back at NVIDIA GTC

and interviewed the company's co-founder and CEO. You can find a link to that video right here

What would you like to see us cover next on What the Future? Let us know down in the comments

Subscribe to CNET for the latest and greatest tech news, and please give this video a like if you learned something new