The Humanoid Takeover: $50T Market, Figure's Full Body Autonomy, and Robots in Dorms #229

In this talk

1. 2:20

   Helix 2 moves to full-body autonomy

   Adcock says Helix 2 replaces the remaining 109,000 lines of C++ control code with neural networks. Figure 3 uses cameras and tactile sensors to plan and adjust whole-body movements, including room-scale tasks such as unloading a dishwasher. Adcock says the system runs onboard and responds to sensor input in real time.

2. 12:22

   Data and deployment define the challenge

   Adcock says language models provide semantic understanding, but robots also need to learn contact, movement, and physics from experience. He argues that closed-loop autonomy in unfamiliar places is a much harder benchmark than teleoperation or replayed motions. Figure aims to advance from short tasks toward days of useful work with little human intervention.

3. 29:48

   Figure plans to expand production and commercial use

   Adcock says Figure is scaling Figure 3 production and expects to put robots on its Baku manufacturing lines in 2026. The company plans commercial deployments with customers, favoring leasing as a way to learn about fleet operations, repairs, and maintenance. He says the current facility could eventually support just under 50,000 robots a year.

4. 33:11

   Home robots depend on capability and safety

   Adcock describes a future robot combining physical skills with conversation, memory, and personality in one model. For home use, he wants robots to handle long tasks in unfamiliar houses and reduce the need for human intervention over time. He says safety, privacy, cybersecurity, and a proven track record are essential before broad household deployment.

5. 47:39

   A large labor market, with major hurdles to scale

   Adcock characterizes human labor as a roughly $50 trillion market and predicts humanoids could eventually work across homes and commercial settings. He says reaching that scale depends on general-purpose neural networks, manufacturing capacity, and robots helping build other robots. He also discusses onboard inference, four to five hours of battery life per charge, and inductive charging through the feet.