Figure, a company dedicated to building advanced humanoid robots, leverages its proprietary neural network, Helix, to enable general-purpose autonomy. Moving away from traditional code-based controllers, the team utilizes vision-language-action models to process camera input and execute complex physical tasks in real-time. The development process prioritizes rapid iteration, with the company transitioning from expensive, CNC-manufactured prototypes to more cost-effective, mass-producible designs like the Figure 3. Key engineering milestones include the "Never Fall" initiative, which ensures stability even during hardware failures, and the use of inductive charging through the robot's feet to support 24/7 operations. By integrating these robots into commercial environments like automotive assembly lines and testing them for domestic chores, Figure aims to achieve artificial general intelligence by grounding AI in physical, real-world interactions.
Sign in to continue reading, translating and more.
Open full episode in Podwise
