Skip to content
IFAI

Robotics

Figure 02 tackles laundry duty autonomously

Published Reach 38

Figure just showed off its Figure 02 humanoid robot autonomously folding and stacking towels with zero teleoperation or pre-scripted code. Running on its Helix neural network, the robot's multi-fingered hands can trace edges, smooth wrinkles, and adapt to fabric slips. Why it matters: The same model that worked in BMW's plant made the jump from warehouse logistics to household chores with just 500 hours of human demo data.

Figure 02 tackles laundry duty autonomously
Image: therundown.ai

Folding laundry is one of those deceptively brutal challenges in robotics. Fabric is soft, unpredictable, and rarely falls into the same shape twice.

Figure just posted new footage showing its Figure 02 humanoid robot folding and stacking towels entirely on its own. There is no teleoperation behind the curtain, no hardcoded trajectory scripts, and no staged lab setup. The entire demo was filmed right inside the founder’s home.

The Helix leap: from car factory to living room

Using its proprietary Helix visual-language-action (VLA) architecture, Figure 02 leverages multi-fingered hands to trace edges, locate fabric corners, and smooth out creases. If the towel slips or wrinkles mid-fold, the robot adjusts its grip dynamically in real time.

According to Figure, this marks a world first for an autonomous multi-fingered humanoid.

What makes this truly wild is the model provenance. The exact same Helix model running this chore recently finished a grueling 20-hour shift moving components at BMW’s Spartanburg manufacturing plant.

Figure did not rewrite its underlying software or run a massive retraining cycle. Instead, engineers simply fed the existing Helix network 500 hours of human demonstration data. With that modest dataset, the humanoid leaped from industrial heavy lifting to delicate domestic chore handling.

Why it matters

We think this demo matters far more than just having freshly folded towels.

Historically, teaching a robot to transition from rigid auto parts to deformable soft fabrics meant redesigning control pipelines from scratch. Figure just showed that a unified physical AI model can generalize across radically different domains with minimal fine-tuning data.

For enterprise operators and future consumers, this changes the timeline on general-purpose utility. You don't need a custom machine for every chore; you need a capable base model and roughly 500 hours of quality demonstration. While you won't be buying a Figure 02 for your laundry room anytime soon given current costs, the path toward a single humanoid platform that works the factory floor by day and tidies up by night suddenly looks much more realistic.

Sources (1)

  1. [1]
    x.com(opens in a new tab)Originalx.com

Best of AI, daily