Figure's billion-dollar bet, and the company that won the robot Olympics

Figure builds a gig economy for robot training

Figure thinks one of the biggest bottlenecks in humanoid robotics is simple: not enough data.

So it has officially launched Index, a platform that pays people to record themselves performing real-world physical tasks that can be used to train Figure's Helix robot AI. Figure says the app spent four months in stealth and has already reached more than 264,000 downloads across 108 countries, with over 44,000 weekly active users.

Those users — Figure calls them Creators — have uploaded more than 16 million videos, and Figure says it has already paid them $15 million.

So this is basically the gig economy meeting the robot-data flywheel.

And strategically, it's a huge bet.

Instead of waiting for thousands or eventually millions of expensive humanoid robots to gather enough real-world experience, Figure is trying to use humans everywhere to bootstrap the physical dataset those robots will eventually learn from. Figure says it will spend a billion dollars on this, mostly on the back-end data analysis.

That’s a massive bet that synthetic data is not good enough.

But it might just give Figure the biggest dataset in robot history.

Agibot dominates the World Humanoid Robot Games

Meanwhile, the second World Humanoid Robot Games are finished, and Agibot is walking away as the overall winner.

The company collected 46 medals: 18 gold, 16 silver and 12 bronze, putting it first both for gold medals and total medals.

But the more interesting part is where those medals came from.

Agibot wasn't just winning dance routines or sprint races. Its OmniHand reached the finals in all eight dexterous-hand events and won seven golds, including tasks like weighing powder and picking up beans with tweezers.

Its X2 won the 100-meter obstacle race. Its A3 won Tai Chi. And Agibot took five more gold medals in scenario-based competitions including hotel work, libraries and emergency response.

And Agibot says the robots it brought weren't one-off competition machines. Models including the X2, G2 and A3 are already mass-produced or deployed commercially.

That's probably the biggest significance of the result.

Robot competitions are becoming less about showing what can be built once, under perfect conditions, and more about testing hardware and software stacks that companies actually intend to ship.

What the Robot Games actually taught us

Zoom out from Agibot, and the results tell a more complicated story.

There was no single super-robot that dominated everything.

RoboPodium's verified results show Tiangong winning both large and small 400-meter events, plus the 4-by-100 relay and heavyweight tug-of-war. Another team won the 1,500 meters. Another won the long jump. A Peking University joint team beat the University of Hong Kong in the table-tennis final.

In other words, we're still seeing specialization.

Some teams are exceptional at locomotion. Others at balance, manipulation, coordinated sports or autonomous perception.

And that might be the most useful way to view these Games.

The athletic events show just how quickly low-level motion control is improving. But the industrial and service challenges are testing something much harder: can a robot perceive a messy real environment, make decisions and successfully finish useful work?

That's the gap the entire humanoid industry is trying to close.

Japan starts building its humanoid supply chain

In Japan, startup Atom and precision-component giant NSK are teaming up on something equally foundational: building more of the humanoid stack domestically.

The companies signed an MOU to test NSK-developed actuators inside Atom humanoids and use those results to improve performance, reliability and eventually manufacturability.

They'll also deploy Atom robots at NSK manufacturing sites to collect physical-AI training data from real production environments.

That's important because Japan certainly doesn't lack robotics expertise.

The question is whether it can combine that manufacturing base, component supply chain and modern embodied AI into a competitive domestic humanoid industry.

This agreement is still early — there's no commercial deployment timetable yet — but that's clearly the ambition.

Germany gets another humanoid challenger

And Germany has a new humanoid startup.

Darmstadt-based Pink Robotics has unveiled the P0: a 1.75-meter, 82-kilogram humanoid with 32 degrees of freedom and an eight-kilogram payload.

The target is logistics, and the company says it's preparing a pilot in a major supermarket distribution center where P0 will sort packaged goods.

Founder Christopher Helmke has been unusually outspoken about Germany falling behind the United States and China in humanoids.

But the next part is what matters.

Pink says it has started training vision-language-action models, but its current demonstrations are still teleoperated.

So the supermarket pilot becomes the real test: not whether Germany can build another impressive humanoid prototype, but whether a small startup can make one reliably autonomous and economically useful.

Number of the day: 45%

And our number of the day is 45%.

That's roughly how far Unitree shares fell from their debut peak after one of the most spectacular robotics IPOs we've seen.

Unitree finished its first trading day in Shanghai 460% above its IPO price, briefly reaching a valuation of about $66 billion. Then came three straight days of declines, erasing roughly $30 billion in market value and cutting the shares about 45% from that post-listing high.

Here's the nuance: even after that plunge, Unitree remains above its original IPO price.

So this isn't a verdict that humanoids don't matter ... it's a reminder that an extraordinary technology story and an extraordinary stock price are not necessarily the same thing.

Sources

1. Figure — Figure’s Billion-Dollar Bet: Gig Platform For Humans To Generate Robot Training Data. https://www.forbes.com/sites/johnkoetsier/2026/08/26/figure-launches-gig-platform-to-get-humans-to-do-work-to-train-robots-will-spend-1-billion/ 

2. Agibot press release supplied for this episode.

3. RoboPodium — World Humanoid Robot Games 2026 results. https://robopodium.com/results

4. Atom / PR Times — Atom and NSK strategic partnership announcement. https://prtimes.jp/main/html/rd/p/000000002.000172692.html

5. Humanoids Daily — Pink Robotics emerges in Germany. https://www.humanoidsdaily.com/news/pink-robotics-emerges-in-germany-with-a-mission-to-revive-industrial-engineering

6. Reuters / AOL — Unitree post-listing stock decline. https://www.aol.com/articles/china-robot-maker-unitrees-post-021417000.html