Where everyday human behavior becomes model-ready training data.

No items found.

Download Dataset

‍

We train humanoids to be trusted around children, pets, and soft materials.

Our training datasets teach humanoids gentle interaction, spatial awareness, adaptive movement, and reliable decision-making in human spaces. Get Hailing.

First Impression

Step 1

Step 2

Step 4

Download the App

Research

Can robots today apply eyeliner on a human, safely, consistently, and with perfect symmetry?

Robotic learning systems are increasingly bottlenecked not by model capacity, but by the availability of diverse, real-world interaction data that complements controlled teleoperation and simulation pipelines. While these approaches provide high-fidelity, robot-aligned trajectories, they are inherently limited in scale and often underrepresent the variability, correction behavior, and micro-adjustments observed in unconstrained human interaction.

Ocean™ introduces a distributed human demonstration layer designed to augment existing robotics data pipelines. The system leverages mobile devices to capture structured, constraint-driven task executions from non-expert users, producing repeatable demonstrations with explicit temporal structure, multi-attempt variation, and observable correction dynamics. Each task enforces controlled capture conditions (fixed framing, sequencing, and execution constraints), enabling extraction of consistent visual trajectories and behaviorally meaningful signals across diverse environments.

Rather than serving as a replacement for teleoperation, Ocean™ is designed as an upstream data source for representation learning and task prior formation. The resulting datasets encode alignment behaviors, error correction patterns, and fine motor intent that can be used to pre-train perception modules, initialize imitation learning policies, and improve robustness in downstream policy optimization.

We turn messy human behavior into model-ready data for Physical AI.

About

We built the first consumer-facing app to generate structured datasets for physical AI. We’re entering a world where robots will outnumber humans, and the next generation of platforms won’t just capture attention, but continuously generate the data that trains frontier models.

We started Ocean™ with a simple belief: Universal Basic Income is achievable when human behavior itself becomes a source of value. Real-world intelligence cannot be learned in simulation alone, it must be grounded in how humans actually move, adapt, and correct in the physical world. Every day, billions of people perform complex, precise actions, from applying makeup to manipulating objects with subtle, continuous adjustments. These moments are rich with signal, yet largely invisible to existing training pipelines.

Ocean™ transforms this latent human intelligence into structured, model-ready data through guided, repeatable tasks captured on mobile. By turning everyday behavior into a scalable data layer, we accelerate the development of physical AI while redefining how people participate in its creation. In the future, contributing to AI won’t be limited to those with specialized training, it will be embedded in daily life.

Download the App

Our training datasets teach humanoids gentle interaction, spatial awareness, adaptive movement, and reliable decision-making in human spaces. Get Hailing.

‍

‍