Add news
March 2010 April 2010 May 2010 June 2010 July 2010
August 2010
September 2010 October 2010 November 2010 December 2010 January 2011 February 2011 March 2011 April 2011 May 2011 June 2011 July 2011 August 2011 September 2011 October 2011 November 2011 December 2011 January 2012 February 2012 March 2012 April 2012 May 2012 June 2012 July 2012 August 2012 September 2012 October 2012 November 2012 December 2012 January 2013 February 2013 March 2013 April 2013 May 2013 June 2013 July 2013 August 2013 September 2013 October 2013 November 2013 December 2013 January 2014 February 2014 March 2014 April 2014 May 2014 June 2014 July 2014 August 2014 September 2014 October 2014 November 2014 December 2014 January 2015 February 2015 March 2015 April 2015 May 2015 June 2015 July 2015 August 2015 September 2015 October 2015 November 2015 December 2015 January 2016 February 2016 March 2016 April 2016 May 2016 June 2016 July 2016 August 2016 September 2016 October 2016 November 2016 December 2016 January 2017 February 2017 March 2017 April 2017 May 2017 June 2017 July 2017 August 2017 September 2017 October 2017 November 2017 December 2017 January 2018 February 2018 March 2018 April 2018 May 2018 June 2018 July 2018 August 2018 September 2018 October 2018 November 2018 December 2018 January 2019 February 2019 March 2019 April 2019 May 2019 June 2019 July 2019 August 2019 September 2019 October 2019 November 2019 December 2019 January 2020 February 2020 March 2020 April 2020 May 2020 June 2020 July 2020 August 2020 September 2020 October 2020 November 2020 December 2020 January 2021 February 2021 March 2021 April 2021 May 2021 June 2021 July 2021 August 2021 September 2021 October 2021 November 2021 December 2021 January 2022 February 2022 March 2022 April 2022 May 2022 June 2022 July 2022 August 2022 September 2022 October 2022 November 2022 December 2022 January 2023 February 2023 March 2023 April 2023 May 2023 June 2023 July 2023 August 2023 September 2023 October 2023 November 2023 December 2023 January 2024 February 2024 March 2024 April 2024 May 2024 June 2024 July 2024 August 2024 September 2024 October 2024 November 2024 December 2024 January 2025 February 2025 March 2025 April 2025 May 2025 June 2025 July 2025 August 2025 September 2025 October 2025 November 2025 December 2025 January 2026
1 2 3 4 5 6 7 8 9 10 11 12 13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
News Every Day |

How 1X Is Teaching Its Humanoid Robot to Learn on Its Own

1

The home robot from 1X is taking a step that could change how machines learn everyday skills.

This week, 1X unveiled what it calls the 1X World Model (1XWM), a new AI system designed to help NEO understand the physical world and act inside it more naturally. Instead of relying mainly on hand-coded rules or thousands of hours of human-operated robot training, NEO now learns from internet-scale video, much like humans do.

According to 1X, this marks a shift away from traditional robot AI systems that directly map images and text to actions. Instead, NEO first imagines what should happen next… then acts on it.

The company says this approach enables the robot to handle objects, motions, and tasks it has never encountered before, using knowledge already embedded in video footage of people interacting with the world.

From seeing to doing: The video brain

Traditionally, training a robot for simple chores like picking up a cup required thousands of hours of costly, real-world data from human operators. 

The new 1XWM approach takes a different path. It uses a video-generation model at its core — similar to advanced AI that creates video from text — but crucially grounds it in the physics and viewpoint of NEO’s own body.

When given a prompt like “pack the lunchbox,” NEO’s AI doesn’t just guess the next arm movement. Instead, it first imagines a short video of the future, visualizing itself completing the task. A second AI component, called an Inverse Dynamics Model, then translates the imagined video frame by frame into the exact motor commands needed to make it a reality.

This method works specifically because NEO is built like a person. The company argues that the robot’s human-like shape and movement are key. Since the AI is trained on vast amounts of video of people interacting with the world, that knowledge transfers more directly to a robot that moves similarly.

“After years of developing our World Model and making NEO’s design as close to human as possible, NEO can now learn from internet-scale video and apply that knowledge directly to the physical world,” said Bernt Børnich, CEO and founder of 1X, in the company’s announcement. “With the ability to transform any prompt into new actions — even without prior examples — this marks the starting point of NEO’s ability to teach itself.”

From known tasks to novel challenges

In demonstrations, this has allowed NEO to handle both familiar and brand-new situations.

It can perform learned tasks, such as grabbing objects, even when they’re placed in new, cluttered settings. More strikingly, the company has shown that it’s attempting entirely novel tasks, such as pulling a tissue from a box, ironing clothes, or brushing a person’s hair, without specific prior training for those actions.

Daniel Ho, an AI Researcher at 1X, stated: “With the 1X World Model, you can turn any prompt into a fully autonomous robot action — even with tasks and objects NEO’s never seen before.”

Perhaps the most profound implication is how NEO could improve.

Instead of being bottlenecked by the speed at which engineers can collect new demonstration data, the robot can now generate its own experience. A successful attempt reinforces its knowledge; a failure provides new data for the model to learn from. This creates a potential “flywheel” where the robot’s capabilities can expand autonomously.

Pricing and availability

The technology is still evolving. 1X openly notes in its technical blog that the system can sometimes generate videos that look correct but subtly violate physics or depth, causing the real robot to miss its grasp. Dexterous tasks such as pouring and drawing remain challenging.

However, the team has found that generating multiple possible future videos and selecting the best one can already improve real-world success rates.

For those looking to bring this evolving intelligence home, NEO is currently available through 1X’s online store. Early Access units are priced at $20,000, with priority delivery slated for 2026. A subscription model is also available for $499 per month.

Industrial humanoids are moving quickly, with HMND 01 Alpha demoing live factory work by picking parts at CES 2026.

The post How 1X Is Teaching Its Humanoid Robot to Learn on Its Own appeared first on eWEEK.

Ria.city






Read also

Galatasaray Vice-President lands in Italy to hold talks for Juventus, Inter and Milan midfielders

Watch: Blue Jackets introduce coach Rick Bowness

How realistic is Mattel’s new autistic Barbie?

News, articles, comments, with a minute-by-minute update, now on Today24.pro

Today24.pro — latest news 24/7. You can add your news instantly now — here




Sports today


Новости тенниса


Спорт в России и мире


All sports news today





Sports in Russia today


Новости России


Russian.city



Губернаторы России









Путин в России и мире







Персональные новости
Russian.city





Friends of Today24

Музыкальные новости

Персональные новости