datraAI
A human operator and robot organizing synchronized physical-world data

Turning physical work intotraining-ready trajectories

ACTUATE / MULTIMODAL PROCESSINGEGOCENTRIC · DEPTH · UMI · TELEOPERATIONLEROBOT · RLDS · HDF5

Backed by

Founders, Inc.The ResidencyNVIDIAInception program

We turn raw data from humans and robots into synchronized trajectories models can learn from.

See the pipeline →

What we build

A hardware-agnostic pipeline that aligns clocks, recovers task structure, extracts perception signals, and exports model-ready behavior.

/01
Explore Actuate →

Continuous human demonstrations captured through owned egocentric wearables and adaptable stereo, depth, UMI-style, and teleoperation rigs.

/02
Plan a capture →
ACTUATE

Different sensors.
One trajectory.

Every rig produces different timestamps, coordinate frames, sample rates, and schemas. Actuate resolves that fragmentation before training begins.

01Ingestion
02Temporal sync
03Quality control
04Segmentation
05Perception
06Trajectory construction
07Retargeting
08Export
OUTPUT

observationt→statet→actiont→observationt+1

Supported now

We are deliberately focused on four data families rather than every possible robotics sensor.

EGO

Egocentric

First-person RGB and IMU from continuous human task demonstrations.

3D

Stereo / depth

Calibrated camera and depth streams for geometry and spatial understanding.

UMI

UMI-style

Portable manipulation capture with camera, gripper, and end-effector trajectories.

TEL

Teleoperation

Robot observations, state, and actions captured during human-controlled execution.

Primary capture today: first-person RGB, accelerometer, and gyroscope from continuous human demonstrations using Panoculon Labs wearable devices.

Data in motion

Public research demonstrations representative of the input families Actuate is built to process. Videos remain attributed to their original teams.

CAREERS

Build the data layer for physical intelligence.

Work at DatraAI