惯性聚合 高效追踪和阅读你感兴趣的博客、新闻、科技资讯
阅读原文 在惯性聚合中打开

推荐订阅源

J
Java Code Geeks
Stack Overflow Blog
Stack Overflow Blog
B
Blog RSS Feed
C
Check Point Blog
D
Docker
Y
Y Combinator Blog
Recent Announcements
Recent Announcements
Google DeepMind News
Google DeepMind News
MongoDB | Blog
MongoDB | Blog
博客园_首页
Apple Machine Learning Research
Apple Machine Learning Research
量子位
有赞技术团队
有赞技术团队
IT之家
IT之家
大猫的无限游戏
大猫的无限游戏
D
DataBreaches.Net
M
MIT News - Artificial intelligence
B
Blog
阮一峰的网络日志
阮一峰的网络日志
freeCodeCamp Programming Tutorials: Python, JavaScript, Git & More
腾讯CDC
让小产品的独立变现更简单 - ezindie.com
让小产品的独立变现更简单 - ezindie.com
V
V2EX
月光博客
月光博客

Forbes - Consumer Tech

This Unhackable Quantum Navigation System Is The Size Of A Loaf Of Bread Apple At 50 — A Leadership Shift And An AR Future We Are Under-Investing In Robotics ... 90% Of Humanoid Robots Are Made In China Ditch The Apple White: Beats Expands Colorful Cable Line-Up With New 10-Foot Option Satechi’s New ChargeView 140W Desktop GaN Charger With Real-Time Display The Hasselblad In Your Pocket: Oppo’s Find X9 Ultra Challenges The Galaxy S26 Ultra There's No Such Thing As Brain Honey How AI Agents Could Rebuild Fashion’s Visual Production Layer Sennheiser’s New Closed-Back Headphones Are Made For The Studio QClaw Goes Global. The Agent Built Itself In 5 Days Apple’s Tim Cook Exit Hides A $4 Trillion Agentic AI Power Move EZQuest Reveals A New Line Of Pro Series USB-C Hubs For MacBook Neo Samsung Galaxy Z TriFold 2 Already In The Works, Report Claims Apple Revealed New Siri Release Date For iPhone, Latest Report Claims How Arcani’s HARK Is Designed For Modern Battlefield Acoustics The Newest Trend In Tech Embraces Femininity And Fun Samsung’s 75R95H Ushers In A New World Of LCD TVs New Apple iPhone Fold Design Pushes Smartphone Rivals To Go Wider And Taller iPhone 18 Pro Report: Four New Colors Leak As Apple Cancels Popular Shade Nothing’s Design-Led Strategy: Carl Pei Reveals The Tech Brand’s Philosophy iOS 26.5 Release Date: When To Expect Your iPhone Messaging Upgrade Google Pixel And Highsnobiety Build A Talent Pipeline For Fashion Android Circuit: Samsung Raises Galaxy Prices, Oppo Pad Mini Teased, Microsoft Closing Outlook App Apple Loop: iPhone Fold Launch Dates, iPad Air Upgrade, iPhone 18 Pro Specs Comcast $117.5 Million Breach Settlement — Are You Eligible? Amazfit Cheetah 2 Pro Takes Aim At The Garmin Audience Disney’s Launches ‘Infinity Vision’ Certification For Premium Theaters SoundPeats Reveals New Air6 HS Semi-Open Wireless Earbuds Amazon’s $11.57 Billion Leap Into Space: A Challenge To Starlink Meta Quest 3 Hit With $100 Price Increase
Humanoid Robots’ 88% Fail Rate: Completing Home Tasks
John Koetsier · 2026-04-14 · via Forbes - Consumer Tech
Humanoid robots are shockingly bad at safely completing common household tasks, says a new report from Stanford.

Humanoid robots are shockingly bad at safely completing common household tasks, says a new report from Stanford.

VCG via Getty Images

Humanoid robots are shockingly bad at completing common household tasks safely, according to the 2026 AI Index Report from Stanford University’s Institute for Human-Centered Artificial Intelligence. So while you can now buy a humanoid robot for around $5,000, or order the 1X Neo for $20,000, or get the new M1 from AiMoga for just over $40,000, you should not expect the robot to function as a mechanical Jeeves.

At least not yet.

AI can probably win a gold medal at the International Mathematical Olympiad. It can outperform human chemists, write pretty good code and even find zero-day exploits in the world’s most secure software. But it may not reliably pick up your dirty socks. In fact, the robots we have today, with the AI we have today, fully succeed safely at only about 12% of real household tasks.

In controlled software simulations, Stanford’s report says, robots can achieve an impressive 89.4% success rate, up from roughly 48% back in 2022. But in the messy and unpredictable real world, that success rate drops like a stone.

"The benchmarks that prove hardest for AI are the ones that require acting in the
real world, where environments are unpredictable and mistakes have physical consequences," the report states. "Even the top model failed to complete
more than a third of tasks safely, with frequent failures when both task completion and safety must be satisfied simultaneously."

Reality is tough. Floors get slippery, cups are angled away from a robot’s hand, drawers might stick a bit when you try to open them, and a child might leave Legos on the floor.

A core problem: our top AI models are trained on words on the internet. That’s fine for understanding how to put words together meaningfully, one after the other.
They’re much more challenged, however, when trying to build a plan for executing physical actions in the real world: a task for physical AI and world foundation models.

And those just aren’t very mature yet.

“The gap between what these models can do in a controlled setting and what they can handle in the real world is still wide,” the report says.

One of the toughest tests for humanoid robots that are going to function in our homes is the Behavior-1K, built on 1,000 real-world tasks sourced from actual humans reporting what they want robots to do in their homes. The best teams in a recent challenge achieved a 25% success rate on these tasks at an “acceptable” rate of quality, but full task success rates were much lower.

That means we need more work to understand how to drive robot actions and behavior safely and successfully.

The good news is that leading robot companies like Figure AI are training their robots on home environments and showing them complete real-world tasks like emptying a dishwasher, or putting away the groceries. The latest videos show that the robots aren’t fast, but they are fairly intelligent about what to put in the fridge versus what to put in the cupboard.

When I interviewed futurist and engineer Peter Diamandis about humanoid robots in January of last year, he predicted that we’d have at least some humanoid robots in homes within calendar 2026.

That prediction has proved to be true.

The next step is to make them truly useful at an affordable price. According to the Stanford report, there’s some room to improve before we can conclusively call mission accomplished on that goal.