🤖 THE FUTURE WAS CURRENTLY STREAMING LIVE. Right now, Figure AI is broadcasting their humanoid robots working a 24/7 autonomous warehouse shift on X. Meet Bob, Frank, and Gary Rose, and Jim the Figure 03 humanoids processing thousands of packages live on camera. Why this is a massive milestone: 100% Autonomous: Powered by the Helix-02 AI platform, they translate camera data directly into physical actions with zero human intervention. Tag-Team Operations: When a robot's battery runs low after 5 hours, it automatically walks to a charging dock while another steps in. Insane Scale: The team has already successfully sorted over 149,000 packages completely on their own. Mundane warehouse labor just became the internet’s most fascinating viral reality show. 👉 Watch the operations or check daily recaps directly on the official account: @Figure_robot#AI #Robotics #TechNews #FutureOfWork #FigureAI

The Indian workers training AI robots to take their jobs hry.yt/2IS

🤖🇮🇳 THE PEOPLE TRAINING THE ROBOTS OF THE FUTURE Thousands of workers across India are reportedly being paid to wear head-mounted cameras and record everyday tasks—from folding towels to factory workto help train AI-powered robots. Workers capture first-person (“egocentric”) video of real-world activities for AI models Tasks reportedly include household chores, textile work, and industrial processes The data helps robots learn how humans interact with unpredictable environments India is emerging as a major global hub for AI data collection and annotation. Supporters say the industry is creating new income opportunities, while critics warn that workers may be helping automate the very jobs they perform.

The humanoid robot boom has a hidden supply chain. It runs through Tamil Nadu. Thousands of Indian workers strap GoPros to their heads and record themselves folding towels, making sandwiches, slicing mangoes. This footage trains AI robots to move like humans. It is called egocentric data. Without it, robots cannot navigate the real world. $MS Morgan Stanley projects over a billion humanoids in use by 2050. The picks-and-shovels play is not the robot makers. It is the data layer underneath them. Objectways, an AI data company working with Amazon $AMZN SageMaker, sources this footage from Tamil Nadu and supplies Fortune 500 clients. The spatial AI data market is scaling right now. Before the robots even arrive at scale. Service as Software is being hand-labelled in Chennai. The next industrial revolution is being trained by the people it will eventually replace. (Credits: @AFP) #HumanoidRobots #SpatialAI #ServiceAsSoftware

A housewife in India earns 250 rupees an hour about $3 just for folding laundry at home. And with that exact work she is training the robots that will one day fold that laundry instead of her. On her head is a camera on a strap. On the clothes and around them sit black-and-white markers that let the system understand where an object is and how the hands move. She does ordinary household chores, and the camera records everything in first person: what she sees, how she picks up the fabric, in what order she folds it. Under recording everything goes slower. What takes 30 minutes at home stretches to 35 to 40 minutes on camera. All it takes is patience. And she is not the only one: thousands of people across India are already doing this, putting on a camera and filming their ordinary day so there is something to train tomorrow's robots on. Her line hits harder than any chart: "Who else will give you 250 rupees an hour just for doing housework?" And here is the surprise. The housework that nobody at home ever pays for, the data industry buys by the hour. What was invisible for centuries has suddenly become raw material. Because robots and AI do not come out of thin air. For a machine to one day fold a t-shirt on its own, a human first has to fold it on camera: by hand, movement by movement, step by step. The woman herself says it plainly: "I see a future full of robots and AI. But there is no going on without people: without human data, all this robotics simply will not run." Out of everything I have seen this year about automation, this is the most honest shot of all: before it can replace a person, the machine first has to be led by that very person's hand. The future really will belong to robots. But right now it is being assembled by hand, at 250 rupees an hour.

Talking to other researchers in the "learning from human video" space early this year, a common observation was that it's hard to show transfer from true in-the-wild Internet video, compared to curated research datasets. In most of these datasets, the humans deliberately move like robots, and hands are tracked with clean 3D labels. The usual starting point for Internet video — run a monocular hand pose estimator on YouTube videos, cotrain with robot data — often doesn't work. In our recent work, we study this "YouTube-type" video setting and try to understand what it takes to absorb egocentric Internet videos into a VLA training pipeline. 1/n

🦔Workers in India are wearing head-mounted cameras for 12 cents an hour to collect training data for humanoid robots. The footage of them doing everyday tasks like cooking, cleaning, sorting, and walking through public spaces gets sold to robotics companies building the models meant to replace those same kinds of jobs in higher-wage countries. The arrangement has been running for roughly two years. Workers do not own the data, do not get residuals, and in many cases are not told what their footage is being used to train. My Take The workers wearing the cameras live in a country where robotics automation will hit decades later, so they are training their own future replacements at a delay that hides the consequence from them personally. The companies buying the data are mostly US and Chinese, building humanoid robots aimed at warehouses, retail, and service jobs in countries paying $15 to $25 an hour rather than 12 cents. Robotics companies need motion data that mimics how humans actually move through real environments, and synthetic data has not been good enough yet. Paying 12 cents an hour in Bengaluru is cheaper than running motion capture studios in Boston, and it works at scale because the worker absorbs the cost of the camera, the discomfort of wearing it, and the long-term loss of any rights to their own movement data. The robotics labor market that eventually emerges from this footage will displace far more wages than the data collection cost to gather. That is the trade investors funding humanoid robotics startups are betting will pay off, and the workers in the videos are the ones paying the tab up front. Hedgie🤗

People in India are getting paid around 250 rupees (~$2.60) an hour to record themselves doing mundane tasks from a first-person perspective. The footage is for training humanoid robots that will eventually replace the human need to do chores.

🇮🇳 Indian workers are now strapping cameras to their foreheads for a living, training AI robots to do their own jobs later. Thousands are recording every movement: folding towels, factory work, basic human tasks... all so machines can replace them. They’re building the noose that will hang their jobs. The AI revolution is already eating the developing world from the inside. Is this peak dystopia? Or do we actually want those jobs be taken from us? Source: AFP / Writer: Oliver

🔥 What is SEAR? SEAR turns the meals you already cook into the data that teaches tomorrow's home robots — and pays you for it. THE PROBLEM 👇 Home robots still can't reliably cook. The blocker isn't hardware - it's data. Robots learn manipulation from real human demonstrations, but high-quality first-person ("egocentric") footage of real hands cooking in real kitchens barely exists. Landmark datasets like EPIC-KITCHENS (Univ. of Bristol, 2018) and vision-language-action models like Google DeepMind's RT-2 (2023, trained on ~130K robot demonstrations) proved the method works. But the volume of real-world kitchen data needed to generalize across millions of homes is orders of magnitude beyond what's publicly available — and most home-robot demos today are still teleoperated, not autonomous. That gap is the bottleneck. THE BUSINESS MODEL 👇 SEAR is a data network. We supply vetted robotics teams with the scarce, high-value resource they can't easily get: real, consented, first-person cooking demonstrations. Instead of paying lab operators, we pay the people already cooking every day. Cooks earn per accepted session; robotics partners license the data under confidentiality. Both sides win — contributors get paid for meals they'd make anyway, builders get data that doesn't exist anywhere else. HOW IT WORKS — 4 steps, one tap 👇 1. Apply → when your city opens, we ship smart glasses gloves, free. 2. Cook → make any dish you like. No scripts, no new chores. 3. Upload → glasses capture the clip, gloves capture hand/finger motion; you review and approve with one tap. 4. Get paid → each accepted cook is graded automatically and credited in-app. Harder dishes teach robots more, so they're worth more. PRIVACY & SECURITY — by design, not as a setting 👇 • Glasses frame your hands and food — never your face. • Personal details are blurred automatically. • Nothing leaves your phone until you approve it. • Footage is never made public, never sold on an open marketplace, never used for ads — shared only with vetted partners under confidentiality. • One-time KYC with government ID; strict 18 only. • Hand-motion data is used to teach skills, not to biometrically identify anyone. WHERE WE ARE Early stage. Equipment, app and contributor onboarding are in active rollout, city by city. Operated by Orlov Innovations LLP. 🔗 sear.world

Welcome to SEAR. Your kitchen is training tomorrow's robots.