Helix 2.5 30-Home Generalization

Summary
Brett Adcock (CEO of Figure) and Corey Lynch (Director of AI at Figure) announce the release of Helix 2.5, a neural network model powering Figure's humanoid robots. The video showcases the robot performing domestic tasks—tidying a living room, making a bed, and folding laundry—in unfamiliar home environments using zero-shot generalization powered by their "Index" human-data pretraining pipeline.

What is shown

• [00:07] Announcement of Helix 2.5.
• [00:39] Task 1: Figure 3 robot picking up scattered children's toys and placing them into a portable basket in an unfamiliar living room.
• [01:18] Task 2: Figure 3 autonomously making a bed, straightening sheets and arranging pillows end-to-end.
• [01:46] Task 3: Figure 3 folding towels on a kitchen/laundry counter and neatly stacking them into a basket.
• [02:24] Map and montage showing evaluations across 30 rented homes throughout the San Francisco Bay Area.
• [04:01] The "Index" data-collection system: workers wearing head-mounted capture rigs gathering first-person manipulation and task data in real-world settings.
• [04:31] Side-by-side comparison experiment demonstrating a failure to grasp an object without Index pretraining versus successful grasping with Index.
• [05:04] Scaling law chart showing a log-linear decrease in validation loss for humanoid robot action prediction as Index pretraining data is doubled (from 1x to 8x).

Claims & numbers

• Helix 2.5 is a single model capable of tidying entire rooms, making beds, and folding laundry in unseen homes without environment-specific training (Corey Lynch).
• Figure tested Helix 2.5 across 30 rented homes across the Bay Area with zero prior data collection in those spaces, reporting success in every home (Brett Adcock and Corey Lynch).
• Over 90,000 people contribute weekly to Figure's Index project (Corey Lynch).
• 35 new minutes of first-person human experience data are uploaded to Index every second (Corey Lynch).
• Pretraining on Index enables "zero-shot whole-body generalization" and establishes a human-to-humanoid-robot transfer scaling law, where validation loss scales predictably down to four decimal points before training runs begin (Corey Lynch).
• Figure is committing $3.5 billion of compute toward training Helix (Corey Lynch).

Notable quotes

• [00:00] "The holy grail for robotics is being able to generalize. This means doing work in unseen places." — Brett Adcock
• [03:30] "In robotics we call this zero-shot whole-body generalization, and it's the first result of its kind." — Corey Lynch
• [05:40] "We're committing to $3.5 billion of compute for Helix." — Corey Lynch

Assessment
This is an official promotional launch video and technical demonstration from Figure. While the video displays smooth autonomous physical manipulation across varied settings, the footage contains rapid jump-cuts, speed-ups, and curated montage clips rather than uninterrupted single-take runs of full task cycles.

Described by gemini-3.8-flash on 2026-09-29 from the video's audio and frames.