Figure AI CEO on Building General Purpose Robots

In this talk

1. 0:38

   Building robots for work and home

   Adcock sees humanoid robots as a major business and product opportunity. Figure designs its hardware for humanlike physical tasks, with plans for commercial work and household chores such as laundry, dishes, and tidying. Teleoperation supports data collection and testing, while the aim for deployed robots is autonomous operation.

2. 5:59

   Manufacturing and commercial deployments

   Adcock says BotQ can produce 12,000 robots per line and considers shipping 100,000 units within four years possible. Figure’s BMW deployment moves sheet metal in a body shop, giving the company daily experience with site networks, safety, production systems, and reliability. He says the team uses those operating metrics to improve the software and process.

3. 10:21

   Autonomy across different tasks

   Adcock describes a logistics robot that uses one neural network to turn camera images into actions, sorting packages at near-human speed and accuracy. Figure has signed a second logistics customer and is preparing an initial deployment. He says the system could manipulate an unfamiliar strawberry, illustrating the goal of generalizing beyond specific training examples.

4. 13:38

   A full-stack approach powered by data

   Figure aims to sell completed work, rather than hardware or models alone, and is developing an integrated system spanning robot design, software, training, manufacturing, and service. Its Helix system uses an open-source backbone for semantic understanding, while Figure handles data collection, model development, training, and testing. Adcock says the company needs more robots generating useful data more than additional GPUs.

5. 16:54

   Scaling the fleet and lowering costs

   Adcock argues that a growing fleet can improve through shared learning while manufacturing scale lowers the cost of each unit of work. He describes a potential advantage for companies that deploy many useful robots, since the fleet could become both cheaper and more capable. He also claims Figure 03 costs roughly 90 to 93 percent less than Figure 02 and says the hardware is reliable enough for software and data to expand its abilities.