Figure humanoid tackles chores in 30 random houses with Helix 2.5

The robotics lab unleashed its humanoid into 30 completely unfamiliar homes powered by a new neural system dubbed Helix 2.5. Engineers did not feed the bot custom 3D scans or spatial layouts of these living spaces beforehand, forcing the machine to figure out the domestic chaos on the fly using raw generalized intuition.

The test protocol threw three classic domestic headaches at the bipedal worker: collecting scattered toys into a basket, folding towels, and making unmade beds. Handling squishy, deformable fabric usually turns robotic grippers into confused spaghetti, yet the machine had to manipulate blankets and towels while constantly recalibrating its own physical stance across unpredictable room geometries.

Under the hood, the foundation model was prepped on Index, a dataset of organic human movements, before getting fine-tuned on chores across completely unrelated sites. When dropped into the test homes, pre-training on human habits rocketed the robot's zero-shot success rate from an abysmal 9% straight to 56%, with researchers giving zero participation trophies for half-made beds. During missteps, the machine autonomously took steps back, circled mattresses, and straightened out sheets without needing an engineer with a remote control to save face.

A coin-flip chore completion rate is hardly enough to fire the local housekeeper, but skipping the need to 3D-map every suburban hallway rewrites the timeline of domestic automation. The boundary between sci-fi fantasy and clunky real-world utility just shifted from lab stagecraft directly into messy living rooms.