Project Go-Big Home Video Dataset

This partnership is really a data distribution deal disguised as a real estate deal. Figure is using Brookfield’s access to more than 100,000 homes to watch how people move through kitchens, living rooms, hallways, cabinets, and dishwashers at massive scale, so Helix can learn the messy edge cases of home life before robots are widely deployed there. That matters because home robotics is mostly a data problem, not just a hardware problem.1 2 3 4

• Figure has been explicit that Project Go-Big is meant to build the world’s largest real world humanoid pretraining dataset, using egocentric human video from Brookfield environments. The strategic value is diversity, many floorplans, object placements, lighting conditions, and chore routines, not just raw volume.1 2 4
• This gives Figure a different home learning path from peers. Sunday is collecting demonstrations through glove wearables across 1,000 plus households and has produced about 10 million trajectories, while The Bot Company is avoiding teleoperation in homes and leaning on autonomy first learning. Figure is betting that broad human video can teach navigation and task priors before a large home robot fleet exists.2 5 6
• Brookfield also points to a future go to market path, not just a training pipe. Managed apartments, rentals, and other property portfolios can become early B2B2C deployment channels where robots generate revenue and more training data in repeated, semi standardized home environments.1 4 5

The next step is a shift from passive observation to active deployment. If Figure can turn this video corpus into reliable home navigation and manipulation, Brookfield style property networks become the bridge from factory pilots to real residential operations, giving Figure a faster route into the home than consumer by consumer rollout.1 2 4