Figure

Helix is the brain behind the robot. It runs two systems in parallel: a slower one for understanding language and scenes, and a fast one that turns those plans into continuous motion at high rates. Figure says the slower system runs at 7 to 9 Hz while the fast one hits 200 Hz, which lets the robot think about a goal and react to a slip at the same time. The company reports that one set of model weights handles picking, placing, drawers, and multi-robot handoffs without task-specific retraining.

The Figure 03 is built to move the way we do. Its upper body runs a large action space covering individual fingers, wrists, head, and torso, all coordinated so the robot can track its own hands with its head while shifting its stance for reach. That coordination is what separates a robot that waves from one that folds a stack of towels.

The robot's camera setup is designed for high-rate visual-motor control. Figure says the newer architecture doubles the frame rate, cuts latency to a quarter, and widens each camera's field of view by 60 percent. There's also a wide-angle camera in each palm, so the robot still senses what it's grabbing when its main view is blocked inside a cabinet.

Figure built its own tactile sensors for the fingertips. The hands can detect pressure as light as three grams, about the weight of a paperclip, which matters for grabbing eggs, tomatoes, or a single card without crushing them. The company says the sensors help the robot tell a firm grip from one that's about to slip.

The robot charges by induction through coils in its feet, stepping onto a base to top up at up to 2 kW without a cable. Figure also softened the exterior with fabric and foam, cut weight compared with the previous generation, and made the outer shell washable and swappable, all with an eye toward living spaces rather than factories.

Figure designed Figure 03 around manufacturing from the start. It uses casting and injection processes instead of slow machining, runs its own BotQ factory, and says the first line can build up to 12,000 robots a year. The company frames this as the step that turns a prototype into a product you can actually buy at scale.

Index is Figure's crowdsourced effort to gather human motion data at a scale no single lab can match. Contributors film everyday tasks through an app and get paid, and Figure uses the footage to pretrain Helix so the robot arrives in a new home with knowledge it never had to learn first-hand.