Figure AI Tested Humanoid Robot in Unfamiliar Homes

The company's Helix 2.5 humanoid achieved a 56% success rate during zero-shot testing in 30 residential environments.

Updated on Sept. 18, 2026 in Robotics

Isometric editorial illustration of a robotic arm manipulating unbranded wooden blocks on a clean wooden surface. — Figure AI successfully completed zero-shot testing of its Helix 2.5 humanoid robot, which performed household tasks in 30 residential environments. AI Illustration. Upload story photo >

Live Poll

Would you trust a humanoid robot to perform household chores like folding laundry in your home?

Figure AI recently completed tests of its Helix 2.5 humanoid robot across 30 unfamiliar homes in the Bay Area. The robot, which utilizes an Index-pretrained model, successfully completed various household chores without specific prior fine-tuning.

Why it matters

This testing cycle aims to advance how robots navigate and manipulate the physical world, bringing the industry closer to deploying machines capable of handling human-level labor. The results demonstrate a significant performance gap between pre-trained models and those developed from scratch.

Figure AI has committed $3.5 billion in compute resources to train the Helix robot while gathering 35 minutes of human experience data every second. The Helix 2.5 model successfully completed chores like folding towels and making beds in its zero-shot trials.

The players

Figure AI

Figure AI is a robotics company focused on developing general-purpose humanoid robots designed for physical work.

The details

The Helix 2.5 was tested in environments where it had no previous experience, demonstrating the efficacy of its training model on unseen household tasks. By comparing its performance to a control model trained from scratch, Figure AI highlighted the importance of pre-trained data in achieving task proficiency.

September 2026
2031

The Tech Race

This development represents a shift toward more capable, general-purpose autonomous machines that can operate in unstructured human environments. It follows the industry trend of scaling compute and training data to bridge the gap between niche robotic automation and versatile humanoids.

As these robots move closer to commercial viability, users may eventually see autonomous assistance for household chores in their own homes. The technology could redefine how physical tasks are managed by incorporating advanced AI into home-care appliances.

The takeaway

The performance gap between pretrained and scratch-trained models emphasizes the critical role of massive data collection in modern robotics. If current progress holds, humanoids may achieve human-level physical capability within five years.