This Robot-Training Startup Is Now Worth $1.2 Billion

XDOF is a young company that helps teach robots how to do real-world tasks. It doesn't build the robots themselves. Instead, it gathers data — for example, videos of people folding clothes or flattening boxes — that robot makers use to train their machines. Less than three months after coming out of stealth, XDOF is already in talks to raise money at a $1.2 billion valuation, a sign of how urgently the industry needs this kind of data.

Why the rush? The artificial intelligence boom was powered by feeding AI programs text and images from the internet. But robots don't have an "internet of real life" to learn from. They need someone to show them how to pick up a cup, open a door, or wipe a table. XDOF does this using remote-controlled robot arms and humans wearing body sensors to record everyday motions. It compares itself to Scale AI, the data-labeling company that helped make ChatGPT possible, but for physical robots.

The company was co-founded by UC Berkeley researchers and has grown fast — its yearly revenue is approaching $50 million. It already works with 20 customers, including major AI labs. The plan is to hire teams of data collectors worldwide, including people who steer robots from a distance and others who record their own movements with sensors.

The catch: robots that learn from human demonstrations still struggle with new situations. And the deal isn't final — terms could change. But the huge valuation shows investors believe the bottleneck to useful home and factory robots isn't the machines themselves. It's the data to teach them.