In thousands of kitchens, living rooms, and garages around the world, a quiet revolution is underway. People in Nigeria, Argentina, India, and more than 50 other countries are strapping iPhones to their heads, pressing record, and then doing the dishes. Or folding laundry. Or cracking eggs. Not for social media — but to teach the next generation of humanoid robots how to be human.

A new report from MIT Technology Review pulls back the curtain on one of the most fascinating — and overlooked — supply chains in the AI industry: the gig economy that’s powering humanoid robot training data.

Why Real-World Data Matters

Building a robot that can navigate a simulated warehouse is one thing. Building one that can reliably pick up a wet sponge, open a cabinet, or fold a fitted sheet in someone’s actual home is another challenge entirely. Virtual simulations can teach robots to do athletic tricks, but they fall short when it comes to the messy, unpredictable physics of everyday objects.

“For robots to work in factories and serve as housekeepers, real-world data is what we need,” one researcher told MIT Technology Review. The problem is that real-world data is expensive and slow to collect — unless you crowdsource it.

That’s exactly what companies like Micro1, a Palo Alto-based data firm, are doing. Micro1 has hired thousands of contract workers across more than 50 countries to wear head-mounted cameras and record themselves performing ordinary household tasks. The footage is then processed, labeled, and sold to robotics companies racing to teach their machines how to operate in unstructured human environments.

The Economics of Robot Training

The pay is around $15 per hour — modest by U.S. standards, but often significantly above local wages in countries like India, Argentina, and Nigeria. For many workers, it’s an appealing gig: flexible hours, no commute, and the novelty of contributing to cutting-edge AI research from your own home.

The demand is being driven by some of the biggest names in humanoid robotics. Tesla’s Optimus, Figure AI’s Figure 03, and Agility Robotics’ Digit are all systems that will eventually need to operate in real homes and factories — environments full of variables no simulation can fully replicate. Investors poured over $6 billion into humanoid robotics in 2025 alone, and a significant chunk of that capital is flowing toward data collection.

The scale is striking. Micro1’s network spans 50+ countries. Similar operations are reportedly running through other data brokers. The volume of video footage being generated and processed is enormous — all of it aimed at giving robots the embodied intelligence to handle the physical world.

Ethical Questions on the Horizon

The gig model isn’t without complications. MIT Technology Review’s reporting highlights thorny questions around privacy and informed consent. Workers are recording their homes and daily routines — spaces they share with family members, roommates, and pets who may not have agreed to be captured on camera. How that footage is stored, shared, and used raises real questions that the industry hasn’t fully answered yet.

There’s also the broader question of transparency. The workers recording footage may not fully understand how their data will be used, which companies will access it, or how long it will be retained. As humanoid robotics scales from lab experiments to mass-market products, these ethical frameworks will need to keep pace.

The Hidden Labor Behind the Robots You’ll See Tomorrow

There’s something poetic — and a little uncanny — about the fact that the most advanced robots in the world are learning from anonymous gig workers in Mumbai and Buenos Aires. Every time a humanoid robot smoothly picks up a coffee mug or wipes down a counter, there’s a good chance it learned from someone who did the same thing on camera for $15 an hour.

This is the hidden infrastructure of the robotics revolution: not just chips and actuators and reinforcement learning algorithms, but hundreds of thousands of ordinary people performing ordinary tasks so that machines can one day do the same. It’s a reminder that even the most futuristic technology rests on very human foundations.

As humanoid robots move from factory floors to our homes and hospitals, the question of who trains them — and under what conditions — will become increasingly important. The gig workers making this possible deserve recognition, and the industry building on their labor owes them transparency.

On April 19, Beijing’s Yizhuang district will host one of the most remarkable spectacles in the brief history of humanoid robotics: more than 300 robots from 26 brands will line up alongside human runners for the second annual E-Town Humanoid Robot Half-Marathon. It’s not just a photo opportunity — it’s a carefully measured test of just how far bipedal AI has come in a single year.

From 60 Robots to 300: A Sector Sprinting Forward

Last year’s inaugural Beijing Humanoid Robot Half-Marathon was groundbreaking by any standard, but 2026 dwarfs it in scale. The number of participating teams has grown nearly fivefold, with more than 80 corporate teams and over 20 university and research camp teams registered across 76 institutions spanning 13 provincial-level regions of China. Any robot taller than 75 centimeters must complete the full 21.0975-kilometer distance in a single continuous run — no shortcuts, no handoffs.

That kind of participation growth in a single year isn’t marketing noise. It’s a direct reflection of how quickly Chinese humanoid robotics companies are producing functional, field-deployable units. According to China’s Ministry of Industry and Information Technology, the country had more than 140 humanoid robot manufacturers and over 330 distinct models at the end of 2025, with the sector expanding at more than 50 percent annually. The half-marathon has become an unofficial national benchmark — if your robot can finish, it’s real.

The Real Milestone: 38% Running Autonomously

What separates 2026 from the debut event isn’t just the scale — it’s the addition of a dedicated autonomous navigation group. Approximately 38 percent of competing robots will field units capable of navigating the course without human remote control. Last year’s runners were largely tele-operated or used pre-programmed gait patterns. This year, a significant portion of the field must read the environment and adapt in real time, outdoors, on a shared course with human athletes separated only by barriers.

That’s a fundamentally different technical challenge than a controlled factory demo or a lab stress test. Sidewalks aren’t factory floors. Lighting changes. Surfaces vary. Spectators create unpredictable noise and movement. Getting 38 percent of over 300 competing robots to navigate that environment without human intervention is genuinely meaningful progress — and the kind of data point that tells researchers and investors far more than any polished product reveal.

New Obstacle Challenge: The Robot “Baturu” Course

The 2026 event also introduces a Robot “Baturu” Challenge on April 18, the day before the marathon itself. Featuring 17 obstacle courses designed to simulate real-world scenarios — including disaster recovery environments and industrial floor layouts — the Baturu challenge tests agility, manipulation, and balance in ways a flat road course cannot. For robotics companies, it’s a public showcase of capabilities that would otherwise stay locked in private test facilities. For spectators, it’s a preview of where these machines are headed next: not just factories, but emergency response, construction, and complex unstructured environments.

What 300 Competing Robots Tells Us About China’s Ecosystem

For observers tracking the global humanoid robotics race, the sheer number of participants at this event underscores something important about China’s industrial strategy. Companies like Unitree — whose robots competed prominently in last year’s inaugural race — alongside dozens of other domestic firms are competing not just for market position inside China but for international credibility. Producing enough functional, race-capable robots to field a 300-unit event isn’t a stunt. It reflects real manufacturing momentum backed by China’s new national standard system for humanoid robots, which the MIIT published in early 2026 to unify technical specifications and reduce production costs across the supply chain.

The race also serves as a de facto product benchmark that no controlled demo can replicate. Finishing times, fall rates, and autonomous versus remote-controlled performance ratios will all be public. That data will matter to procurement teams, investors, and international partners evaluating which Chinese robotics firms are ready for serious commercial deployment.

InteliDroid Perspective

For those of us tracking the long arc toward widely-owned, versatile humanoid AI, the Beijing Half-Marathon captures something the spec sheets and funding announcements often miss: robots operating in the messy, unpredictable real world. The question isn’t just whether a humanoid can run 21 kilometers. It’s whether it can do so while perceiving, reacting, and adapting the way any field-deployed robot will eventually need to. Thirty-eight percent autonomous navigation in a live public race environment is a credible proof point — and with the event just weeks away, it’s a number worth watching closely.

What to Watch Next

With the race set for April 19, pay attention to which brands post the fastest autonomous completion times — those results will carry far more weight than any press release. Beyond the finish line, expect Chinese robotics firms to leverage strong performances here to accelerate international partnerships and export strategies through the rest of 2026. The half-marathon is becoming the industry’s clearest signal of who’s ready to move from prototype to product.

On March 30, 2026, Shanghai-based Agibot rolled its 10,000th humanoid robot off the production line — becoming one of the first companies in the industry to reach this threshold at genuine commercial scale. What makes this number remarkable isn’t just its size; it’s the speed. Agibot went from 5,000 to 10,000 units in only three months, a pace more than four times faster than its previous production phase. That kind of acceleration, in an industry that has spent years promising mass deployment, is worth paying close attention to.

From Prototype Dreams to Production Reality

Agibot was founded in February 2023 by former Huawei engineers Deng Taihua (CEO) and Peng Zhihui (CTO) — the latter having built a cult following in China for personal robotics projects, including a self-driving bicycle and an Iron Man-style robotic arm. The company moved fast: it secured six funding rounds from major backers including HongShan (Sequoia China), Baidu, and SAIC Motor, reaching a valuation north of $960 million USD before its third birthday.

The production numbers are telling. It took Agibot nearly two years to hit its first 1,000 units, about a year to scale to 5,000, and just three months to double that to 10,000. According to analyst firm Omdia, Agibot shipped 5,168 humanoid robots in 2025 alone — more than any other company globally, in a market that collectively shipped roughly 13,000 units industry-wide that year. That means Agibot commanded roughly 40% of global humanoid robot shipments in 2025.

Where the Robots Are Going

These aren’t warehouse curiosities sitting in labs. Agibot’s robots are deployed across logistics, retail showrooms, hospitality services, education, and increasingly, industrial manufacturing lines. Their flagship A2 bipedal humanoid stands 1.75 meters tall and is capable of tasks as precise as threading a needle. The industrial-grade G2 wheeled model achieves sub-millimeter assembly precision with force control suited for automotive-grade tasks. The company’s embodied AI operating system, Lingqu OS, runs across their product family — providing a common software foundation that speeds up real-world deployment.

This isn’t a single product story. Agibot’s portfolio now spans six product lines ranging from the ~$20,000 X2 education robot to the full-size A2 priced between $100,000–$190,000. That product breadth — from classroom to factory floor — positions Agibot as something genuinely different from the narrowly focused Western competitors still figuring out their first commercial deployment.

The China Advantage — and Its Implications

Agibot’s milestone doesn’t exist in a vacuum. It’s part of a broader Chinese push to dominate humanoid robotics at scale. In late February 2026, China’s Ministry of Industry and Information Technology published its first national standard system for humanoid robots and embodied intelligence — a formal framework covering brain-like computing, limb components, full-machine systems, and safety ethics. More than 120 researchers and policymakers convened in Beijing to establish these standards, signaling that China is treating humanoid robotics as a strategic national industry, not just a startup story.

The contrast with Western competitors is sharp. Tesla’s Optimus program, despite years of headlines, is still in R&D phase with no robots confirmed doing productive work in factories as of Q1 2026. Figure AI carries a $39 billion Series C valuation that many analysts view as running ahead of commercial reality. Boston Dynamics is shipping its electric Atlas — but all 2026 units are already committed to Hyundai’s facilities and Google DeepMind. Meanwhile, Agibot has already shipped 10,000 robots across real-world paying customers.

Physical AI Investment Is Accelerating Globally

Agibot’s milestone lands at a moment of unprecedented investor conviction in physical AI. Q1 2026 saw $300 billion flow into global venture deals, with AI commanding 80% of that total. Physical Intelligence — the two-year-old San Francisco startup building foundation models for robots — is reportedly in talks to raise $1 billion at an $11 billion valuation, effectively doubling its valuation in just four months. The robotics funding boom is real, and it’s not confined to any one geography.

But there’s an important distinction between capital and output. Agibot has translated investment into actual units, actual deployments, and actual production velocity. That’s the gap that matters — and it’s the gap that the rest of the industry is racing to close.

InteliDroid Perspective

The vision at InteliDroid has always been a future where capable, versatile humanoid robots are widely owned and genuinely useful — not just enterprise curiosities locked behind million-dollar contracts. Agibot’s 10,000-unit milestone moves that vision measurably closer to reality. When production curves bend this sharply upward, prices tend to follow — downward. The X2 at $20,000 and the A2 at $100,000–$190,000 are still out of reach for most consumers, but the economics of scale are doing their work.

China reaching this production threshold first matters not just competitively, but for the whole industry. It proves that humanoid robots can be manufactured at commercial scale. What happens in Agibot’s Shanghai facility today shapes what robots cost — and who can afford them — tomorrow.

What to Watch Next

Keep an eye on whether Western humanoid robot makers respond to Agibot’s production momentum with accelerated timelines of their own — and watch for Agibot’s next pricing moves as it eyes global market expansion beyond China.