Geographies
17
Countries of active data collection across 4 continents
Environments mapped
130+
Distinct real-world settings from studio apartments to commercial kitchens
Data ingested
4M+ hrs
Multi-modal real-world activity across modalities
Task taxonomy
12,400+
Labeled task vectors and subtask annotations across our activity ontology
What we're building

Physical AI built for the
environments people live in.

Most robotics research optimizes for controlled settings. We build for the opposite — the clutter, ambiguity, and variability of real homes and workplaces. That demands an AI foundation trained on how the world actually behaves. We have assembled one of the broadest real-world training data programs in the industry to make this possible.

Scene understanding
Core modality

Robots that parse a room — objects, relationships, and likely next events — before taking any action.

Human activity modeling
Core modality

Trained on large-scale video of people engaged in everyday tasks across diverse environments and demographics.

Dexterous manipulation
Core modality

Fine-grained interaction with objects — picking, placing, operating — generalized across form factors.

Spatial navigation
Core modality

Movement through real, changing spaces — not mapped environments, not simulation. Ground-truth real.

Instruction following
Core modality

Understanding and executing natural-language task requests across novel contexts without re-training.

We are actively
building our data ecosystem.

We partner with organizations across the data supply chain — collectors, annotators, platform operators, and technology providers. Tell us about what you bring. We move quickly with the right partners.

Start partnership inquiry  → Takes about 5 minutes  ·  Strictly confidential