Figure launches Project Go-Big to build massive humanoid dataset

Training an LLM is easy when you can just scrape the entire internet, but teaching a humanoid robot to navigate a living room is another story.

Figure just announced Project Go-Big, an ambitious initiative to build the world's largest humanoid pretraining dataset.

Moving into 100,000 real homes

According to Figure, physical robot data can't simply be scraped from the web. You have to capture it in the real world.

To scale this up, Figure teamed up with real estate giant Brookfield, which owns more than 100,000 residential units. That footprint gives the robotics company a massive testing ground inside real physical environments.

Why it matters

The biggest bottleneck for humanoid robotics right now isn't compute or model size, it's real-world embodied data. Robots trained only in sterile labs quickly break down when faced with the chaos of actual human homes.

Partnering with Brookfield is a clever shortcut. Instead of slowly negotiating access or building fake mockups, Figure gets a direct line to thousands of diverse living spaces. The catch, of course, will be handling safety and privacy inside people's personal homes while gathering all that physical interaction data.