Figure AI Tested Humanoid Robot in Unfamiliar Homes

The company demonstrated that pretraining methods significantly increase a humanoid's ability to navigate and perform domestic tasks.

Updated on Sept. 18, 2026 in Robotics

A close-up of a sleek, metallic robotic arm reaching toward stacked white fabric in a clean, modern laboratory setting. — Figure AI is conducting research-stage trials of its Helix 2.5 humanoid robot, testing its ability to perform domestic tasks in diverse home 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 has tested its Helix 2.5 humanoid robot across 30 unfamiliar Bay Area homes. The research-stage trial evaluated the robot's success in performing zero-shot household tasks without environment-specific fine-tuning.

Why it matters

The company aims to develop systems that can learn how humans manipulate the physical world, a critical hurdle for general-purpose automation. Figure AI has committed $3.5 billion in compute resources to training the Helix model to overcome existing limitations in environmental adaptability.

The Index-pretrained Helix 2.5 model achieved a 56% success rate on blind zero-shot trials in new environments. This significantly outperformed the 9% success rate achieved by a model trained from scratch.

The players

Figure AI

A robotics company focused on building autonomous humanoid robots capable of human-level physical tasks.

The details

The Helix 2.5 was tested on household tasks including tidying living rooms, folding towels, and making beds. To drive these capabilities, Figure AI collects 35 minutes of new human experience data every second to inform the robot's navigation and manipulation processes. The Index-pretrained model represents a shift toward using massive pre-existing datasets to teach humanoids how to interact with the physical world without requiring environment-specific training.

September 2026: Figure AI reported on recent humanoid robotics testing.
2031: Projections indicate robots could match human capabilities in physical work.

The Tech Race

This development sits within the broader, high-stakes competition to achieve general-purpose physical labor via robotics. The field is currently transitioning from highly constrained industrial automation toward systems capable of adapting to diverse, unstructured physical environments.

While these tests took place in 30 Bay Area homes, the technology remains in the research-stage and is not currently available for consumer use. The company suggests that robots may match human capabilities for most physical work within five years.

The takeaway

The gap between models trained from scratch and those using large-scale pretraining underscores the necessity of massive, diverse physical datasets for robotics. Future industry progress will be marked by whether these success rates hold as Figure AI scales trials beyond the initial 30 homes.