Figure Launches Humanoid Robot Upgrade Helix 02 for Full-Body Autonomy

Figure has introduced Helix 02, its most advanced humanoid AI yet, designed to control an entire robot’s body as a single coordinated system. 

The new model allows a humanoid to walk, handle objects, balance, and adapt continuously without human input, marking a significant step toward practical robot autonomy. The company says the system moves beyond short demonstrations and scripted motions, enabling robots to perform real-world tasks from start to finish.

To showcase the upgrade, Helix 02 completed a four-minute autonomous task in a full-sized kitchen. The robot walked to a dishwasher, unloaded dishes, placed items around the room, reloaded the machine, and started it again, executing 61 separate locomotion and manipulation actions without resets or teleoperation.

Figure described the demonstration as the “longest horizon, most complex task completed autonomously by a humanoid robot to date.”

The test highlights a key capability: the robot can walk while holding delicate items, use both hands in coordination, and even use its hip or foot to interact with doors and drawers when its hands are occupied.

‘System 0’ replaces over 109K lines of code

Helix 02 runs on what Figure calls a System 0, System 1, and System 2 architecture, with each layer operating at different speeds.

• System 2 handles high-level thinking, understanding language commands such as “walk to the dishwasher and open it” or “carry the bowls to the counter.” It processes scenes and sequences behaviors but doesn’t worry about the mechanics of walking or gripping.
• System 1 translates those goals into full-body movements 200 times per second (200Hz), pulling data from head cameras, palm cameras, fingertip sensors, and the robot’s sense of its own position in space. According to Figure, this is the first time they’ve demonstrated neural networks using input from palm cameras and tactile sensors.
• System 0 executes everything at 1,000 times per second (1 kHz), handling balance, contact forces, and coordination across the entire body.

The foundation comes from System 0, which Figure trained on over 1,000 hours of recorded human motion data. Rather than programming separate rules for walking, turning, or reaching, the system learned how people naturally move while staying balanced.

“System 0 replaces 109,504 lines of hand‑engineered C++ with a single neural prior for stable, natural motion,” the company stated.

Beyond the dishwasher task, Figure demonstrated four manipulation challenges that require the robot’s new tactile sensing and palm-mounted cameras.

The robot unscrewed a bottle cap while stabilizing the container without crushing it. It located and extracted a single pill from a medicine organizer when the pill was hidden from its head camera. It pushed exactly 5 milliliters from a syringe despite variable resistance. And it picked small metal pieces from a cluttered box where objects overlapped and shifted during handling.

Why this matters

Robots have struggled with “loco-manipulation” for decades. Usually, a robot has to walk, stop, stabilize itself, and then reach. Helix 02 kills that stop-and-go style. By treating the robot as one unified system that connects every sensor to every motor, Figure is moving away from choreographed stunts toward robots that can actually think and work on their feet.

“The results are early – but they already show what continuous, whole-body autonomy makes possible,” Figure wrote.

Meanwhile, China is taking humanoid robotics in a very different direction. Read how the URKL Humanoid Robot Combat League is turning bots into prizefighters

The post Figure Launches Humanoid Robot Upgrade Helix 02 for Full-Body Autonomy appeared first on eWEEK.