Domestic Dexterity: The Robot Butler's Laundry Problem

1X Technologies opened preorders for its NEO humanoid in late October 2025. The robot costs $20,000 outright, or $499 a month, and 1X says the first units reach American homes this year¹. On camera, NEO folds a shirt and empties a dishwasher. When a Wall Street Journal reporter tested it, Fortune reported, a remote operator handled every task NEO performed². Figure AI's Brett Adcock wants a robot that skips the headset before this year ends. Rodney Brooks, who cofounded iRobot and Rethink Robotics, puts dexterous humanoids more than a decade out.

The Twenty-Thousand-Dollar Roommate

Bernt Børnich, 1X's founder and chief executive, made the case when NEO's preorders opened: "Humanoids were long a thing of sci-fi … then they were a thing of research, but today — with the launch of NEO — humanoid robots become a product"¹. NEO sometimes meets a task past its own training, such as an odd dishwasher load or a specific batch of laundry. The owner then schedules a session, and a trained 1X operator drives the robot's limbs by VR headset, seeing exactly what its cameras see². It carries about 55 pounds and runs four hours on a charge, Fortune reported².

Figure AI is building for the same homes with different hardware. Its Figure 03, unveiled Oct. 9, 2025, wears a washable fabric cover and charges itself from a floor plate³. That design suits a living room more than a warehouse floor. Adcock told TIME what the company is chasing: "We want the robot to be able to do most things in your home, autonomously, all day." He attached a year to it: "We think we can get there in 2026, but it's a big push"³.

Sunday Robotics is starting smaller still. The company is recruiting a limited group of households for a "Founding Family" beta that launches in late 2026, in place of a public sale⁵. Sunday, cofounded by Tony Zhao, gathers training data a different way: more than 500 data collectors across the United States wear a glove built to match Memo's hands⁶. "In robotics, if the only thing we can rely on is teleoperation, to gather the amount of training data it would take like decades for sure," Zhao said on the TBPN podcast, as Euronews reported⁶. Sunday's own site draws the contrast directly, describing Memo as "an autonomous robot that does not need to learn through human tele-operation in your home"⁵.

The Middle-Class Wait

When does a home robot reach a household on a median income? Elon Musk has repeatedly said Optimus will cost $20,000 to $30,000 once Tesla builds it at scale⁷. At the World Economic Forum in January 2026, he put a year on the sale: "probably sometime next year — I would say that by the end of next year — I think we will be selling humanoid robots to the public"⁸. Tesla has ordered parts for about 15,000 Optimus units in 2026, according to supply-chain reports The Motley Fool cited, and Musk says Tesla could build up to a million in 2027⁷.

Unitree's Wang Xingxing took on a harder question at Beijing's World Robot Conference in August 2026. When will a robot placed in an unfamiliar home handle about 80 percent of its tasks from voice or text commands alone? Fortune reported Wang's answer: two to three years at the fastest, and five or even ten at the slowest⁹. He said it a day after Unitree's stock jumped 460 percent⁹. A year earlier, at the same conference, he had put the milestone within five years at the latest⁹. Goldman Sachs now expects 6.5 million humanoid robots shipped worldwide by 2035, worth $138 billion that year, and it raised its 2030 forecast from 256,000 units to 890,000¹⁰. Bank of America projects shipments climbing from 90,000 units in 2026 to 1.2 million by 2030, and it expects 62 percent of the world's humanoids to work inside homes by 2060¹¹. Both forecasts count robots shipped, a different measure from chores finished unattended in a kitchen.

Training on Video, Short on Touch

Every company here needs training data, and it comes from three places. Teleoperation is the most direct. A human drives the robot's actual limbs, so every action carries an exact label, though collecting it runs slow and costly. Physical Intelligence cofounder Sergey Levine describes his own company's start this way: "We initially controlled our robots with teleoperation in a variety of different settings"¹². Simulation turns a small set of real recordings into many more. Nvidia's Cosmos models generate synthetic video for training robots, and the company says its data pipeline can "process, curate and label 20 million hours of videos in 14 days using the NVIDIA Blackwell platform, instead of over three years using a CPU-only pipeline"¹³. Jensen Huang, Nvidia's founder and chief executive, frames the stakes in one line: "The ChatGPT moment for robotics is coming"¹³.

Figure AI has bet hardest on the third source: video of people going about their lives at home. A partnership with Brookfield Asset Management gives the company access to more than 100,000 residential units¹⁴. Figure trained its Helix model on what it calls "100% egocentric human video data, collected passively as people do behaviors in real Brookfield homes," and reports the approach "required no robot demonstrations whatsoever" for one task: walking through an unfamiliar house on command¹⁴. The company calls it a first for humanoid robots. That result covers navigation, a coarser skill than the fine motor control that laundry and dishes demand.

Meta has reached further, toward footage already on the internet, filmed long before anyone planned to train a robot on it. Its V-JEPA 2 model learned from more than a million hours of that video, and after 62 hours of robot data on top, it planned a robot arm's pick-and-place moves on new objects in new settings, succeeding 65 to 80 percent of the time¹⁹. The paper behind that model names its sources: YouTube, through both YT-Temporal-1B and the instructional clips of HowTo100M²². Nvidia's Cosmos models drew on YouTube too. In August 2024, a 404 Media investigation reported that Nvidia had been scraping YouTube video to train them, and Nvidia told the outlet its work was "in full compliance with the letter and the spirit of copyright law"²³. Tesla now films its own. Sherwood News reported in August 2025 that Tesla had moved Optimus training from motion-capture suits and operator headsets to video recordings²⁰. Video scales faster than any teleoperation crew. Rodney Brooks names what it leaves out. A human hand carries about 17,000 low-threshold mechanoreceptors, sensors that fire as pressure is applied and released, information a video frame skips entirely⁴. "We do not have such a tradition for touch data," he writes⁴.

The Laundry Problem

Two companies already tried selling a laundry-folding machine, and both shut down. Seven Dreamers Laboratories filed for bankruptcy in Tokyo in April 2019, roughly $22 million in debt¹⁵. It had built Laundroid, a wardrobe-sized cabinet that TechCrunch found too costly and too bulky for an ordinary home, calling it "little more than a fun show demo" and the wider category "an impossible nut to crack"¹⁵. FoldiMate, a countertop folder, needed each garment clipped in by hand, and it handled shirts, tops, trousers, and dresses while skipping small items such as underwear and large ones such as sheets²¹. In August 2021 it told customers it had ceased operations and sent them a form to reclaim their $85 deposits²¹.

The physics explain the pattern better than the balance sheets do. Wai-keung Fung, a senior lecturer in electronics, robotics and control engineering at Cardiff Metropolitan University, describes what a person handles on instinct: "A shirt may be inside out. One sleeve may be hidden underneath another garment. A pair of trousers may be twisted around a towel"¹⁶. He names the harder problem underneath that one: "Fabric is also soft and constantly changing shape. Pick up one corner of a T-shirt and the rest of it moves"¹⁶. Reading a care label, separating a silk blouse from a load of cottons, and sending a dry-clean-only coat out of the house all demand judgment on top of that. "No robot system can yet reliably handle the whole laundry process, from carrying a basket to the washing machine and sorting clothes for different programmes to taking them out, hanging them to dry, folding and ironing them," Fung writes¹⁶. He frames the chore as a sharper test than it looks: "This is why an everyday chore can reveal more about robot intelligence than a spectacular athletic feat"¹⁶.

Dishes and the Demo Gap

Loading a dishwasher seems the easier chore, and several labs have video to show it. Figure's Helix system learned to singulate stacked plates and load them in order¹⁷. It also picks up a glass with one hand, reorients it, and places it with the other. Sunday Robotics' Memo clears a dinner table into a dishwasher, and according to cofounder Tony Zhao, it kept every wine glass intact across more than 20 live demonstration sessions⁶. Each clip shows the good take and keeps the retry count, the success rate, and any messier setting off camera.

James O'Donnell of MIT Technology Review, after talking with Ken Goldberg, a UC Berkeley robotics professor, offers a rule of thumb for this kind of video: "The most important thing to know is whether a robot is being tele-operated by a human off screen rather than executing the tasks autonomously"¹⁸. A second question follows for any viral clip: how many takes? "If a humanoid shows off an impressive ability to sort objects, but it took 200 tries to do the task successfully, that matters," O'Donnell writes¹⁸. 1X answers the first question inside the product it sells: a 1X employee wearing a VR headset remotely operates NEO for any dishwasher load or laundry basket its training has yet to cover².

The Year Question

Ask the people building these robots for a year, and the answers span more than a decade. Adcock says 2026³, and Musk says Tesla will sell to the public by the end of 2027⁸. Sergey Levine gives Physical Intelligence a shot at a robot that runs a household on its own within about five years¹². Wang Xingxing's range runs from two years to ten⁹. Rodney Brooks draws the line furthest out. In his September 2025 essay on dexterity, he wrote: "By my third law of robotics that says that we are more than ten years away from the first profitable deployment of humanoid robots even with minimal dexterity"⁴.

Each camp rests on a real bet. Adcock's side needs more video and more robot hours to make a demo skill work in an unfamiliar house, and Nvidia, Meta, Tesla, and Figure are funding that bet at scale. Brooks's side holds that touch matters more than footage, since force and pressure carry information a camera skips entirely. The evidence that would settle it is simple to name: a robot that sorts and folds a stranger's mixed load of cottons and silks with the headset switched off. Until someone publishes that result, the year stays the reader's call.

By the numbers

• $20,000: 1X's price for NEO, or $499 a month on a six-month commitment².
• 55 pounds: NEO's carrying capacity, with four hours of runtime per charge².
• 6.5 million: humanoid robots Goldman Sachs now expects shipped worldwide by 2035, a market it prices at $138 billion that year¹⁰.
• 500-plus: the data collectors wearing Sunday Robotics' training glove across the United States⁶.
• Two to ten: the span of years Unitree's chief executive gives before a robot in an unfamiliar home handles most tasks by voice or text alone⁹.
• 20 million: hours of video Nvidia says its Cosmos pipeline can process, curate, and label in 14 days on Blackwell chips¹³.

What to watch

Watch whether a single Figure 03 reaches a paying household in 2026, the year Adcock called "a big push"³. 1X could settle the autonomy question directly by publishing what share of NEO owners' requests the robot finishes alone and what share goes to a remote pilot. A published success rate for sorting fabrics or reading care labels, from any manufacturer, would say more than another polished demo. The two laundry-folding ventures in this piece, Seven Dreamers Laboratories and FoldiMate, shut down before publishing one.

Sources

1. Mike Oitzman, "NEO humanoid designed for household use, available for preorder," The Robot Report, Oct. 30, 2025, https://www.therobotreport.com/1x-announces-pre-order-launch-neo-humanoid-robot/
2. Matty Merritt, "For $20,000, a humanoid robot will do your household chores for you like unloading the dishwasher and watering plants—but it still needs help," Fortune, Feb. 26, 2026, https://fortune.com/2026/02/26/humanoid-robot-that-will-do-chores-for-you-robotics-company-1x/
3. Billy Perrigo, "Figure 03 Is The Robot in Your Kitchen," TIME, Oct. 9, 2025, https://time.com/7324233/figure-03-robot-humanoid-reveal/
4. Rodney Brooks, "Why Today's Humanoids Won't Learn Dexterity," rodneybrooks.com, Sept. 26, 2025, https://rodneybrooks.com/why-todays-humanoids-wont-learn-dexterity/
5. Sunday Robotics, "Say hello to Memo," sunday.ai, accessed Sept. 27, 2026, https://www.sunday.ai/
6. Roselyne Min, "Meet Memo, a home robot that can grab wine glasses and load the dishwasher," Euronews Next, Nov. 25, 2025, https://euronews.com/next/2025/11/25/meet-memo-a-home-robot-that-can-grab-wine-glasses-and-load-the-dishwasher
7. Leo Sun, "Elon Musk Says Optimus Will Ship to Customers in 2027. Here's How Much Revenue That Could Actually Generate.," The Motley Fool, Sept. 14, 2026, https://www.fool.com/investing/2026/09/14/elon-musk-says-optimus-will-ship-to-customers-in-2/
8. "Transcript: Elon Musk at Davos World Economic Forum, Jan. 2026," whatsuptesla.com, Jan. 27, 2026, https://whatsuptesla.com/2026/01/27/transcript-elon-musk-at-davos-world-economic-forum-jan-2026/
9. Marco Quiroz-Gutierrez, "CEO of viral backflipping robot company says robots may be years from their 'ChatGPT moment,' even after investors sent his stock soaring 460%," Fortune, Aug. 21, 2026, https://fortune.com/2026/08/21/unitree-ceo-wang-xingxing-humanoid-robots-chatgpt-moment-years-away-stock-soars-460-percent-ipo/
10. Rich Duprey, "Robots Everywhere: Goldman Sachs Now Sees 6.5 Million Humanoid Robots by 2035," 24/7 Wall St, Sept. 13, 2026, https://247wallst.com/investing/2026/09/13/robots-everywhere-goldman-sachs-now-sees-6-5-million-humanoid-robots-by-2035/
11. Nick Lichtenberg, "Bank of America Sees a World With 3 Billion Humanoid Robots by 2060," Fortune, March 13, 2026, https://fortune.com/2026/03/13/bank-of-america-humanoid-robot-forecast-3-billion-2060
12. Dwarkesh Patel, "Fully autonomous robots are much closer than you think – Sergey Levine," Dwarkesh Podcast, Sept. 12, 2025, https://www.dwarkesh.com/p/sergey-levine
13. Nvidia, "NVIDIA Launches Cosmos World Foundation Model Platform to Accelerate Physical AI Development," Nvidia Newsroom, Jan. 6, 2025, https://nvidianews.nvidia.com/news/nvidia-launches-cosmos-world-foundation-model-platform-to-accelerate-physical-ai-development
14. Figure AI, "Project Go-Big: Internet-Scale Humanoid Pretraining and Direct Human-to-Robot Transfer," figure.ai, Sept. 18, 2025, https://www.figure.ai/news/project-go-big
15. Brian Heater, "The company behind Laundroid files for bankruptcy," TechCrunch, April 23, 2019, https://techcrunch.com/2019/04/23/the-company-behind-laundroid-files-for-bankruptcy/
16. Wai-keung Fung, "Can your robot do the laundry?," RTÉ Brainstorm, Sept. 10, 2026, https://www.rte.ie/brainstorm/2026/0910/1591014-robots-laundry-everyday-chores-robotic-intelligence-systems/
17. Figure AI, "Helix Loads the Dishwasher," figure.ai, Sept. 3, 2025, https://www.figure.ai/news/helix-loads-the-dishwasher
18. James O'Donnell, "A Skeptic's Guide to Humanoid Robot Videos," MIT Technology Review, Aug. 27, 2024, https://www.technologyreview.com/2024/08/27/1103035/a-skeptics-guide-to-humanoid-robot-videos/
19. Meta AI, "Introducing the V-JEPA 2 World Model and New Benchmarks for Physical Reasoning," ai.meta.com, June 11, 2025, https://ai.meta.com/blog/v-jepa-2-world-model-benchmarks/
20. Rani Molla, "Tesla's robots get the same treatment as its cars: Video training," Sherwood News, Aug. 25, 2025, https://sherwood.news/tech/teslas-robots-get-the-same-treatment-as-its-cars-video-training/
21. Wikipedia contributors, "FoldiMate," Wikipedia, accessed Sept. 27, 2026, https://en.wikipedia.org/wiki/FoldiMate
22. Mido Assran et al., "V-JEPA 2: Self-Supervised Video Models Enable Understanding, Prediction and Planning," arXiv, June 11, 2025, https://arxiv.org/abs/2506.09985
23. Samantha Cole, "Leaked Documents Show Nvidia Scraping 'A Human Lifetime' of Videos Per Day to Train AI," 404 Media, Aug. 5, 2024, https://www.404media.co/nvidia-ai-scraping-foundational-model-cosmos-project/