For Robotics Labs

Training data fromreal humans, at work.

AR glasses on warehouse workers. Every shift generates labeled human demonstrations — the kind of data that doesn't exist anywhere else.

Get in touch
Scroll
Egocentric
Video stream — live
3 more
Modalities in development
Piloting
Auto-annotation pipeline
<24 h
From capture to labeled dataset

What we capture.

And what's coming next.

60 fps · 4K · wide-angle

Egocentric Video

First-person perspective from AR headsets. Every grasp, reach, and manipulation captured with surgical clarity.

In developmentIMU · 200 Hz · 6-DOF

Inertial Motion

Full 6-axis body kinematics synchronized to the frame. Accelerations, rotations, and micro-corrections during physical tasks.

In development120 fps eye-tracking · spatial mapping

Gaze & Attention

Where humans look before they act. Fixation points, saccades, and attention shifts tied to every manipulation event.

Real-time · object + action + state

Semantic Labels

Every object tagged with class, pose, and state. Every action labeled — pick, place, inspect, scan, pack — at millisecond resolution.

idrak-data-stream · live
14:23:40.002[VIDEO]frame_6841 · grasp_event · SKU-1047
14:23:40.007[IMU]wrist_accel: [1.2, -0.4, 9.8] m/s²
14:23:40.010[GAZE]fixation → box_surface · 180ms
14:23:40.015[LABEL]action: pick · obj: cardboard_box · state: full
14:23:40.020[VIDEO]frame_6842 · lift_trajectory_start
14:23:40.025[IMU]shoulder_torque: 2.1 Nm · elbow_flex: 42°
14:23:40.030[LABEL]action: carry · trajectory: aisle_D → station_4
14:23:40.035[GAZE]saccade → destination_marker · 90ms
Streaming · 80,412 events today
The pipeline

Every shift is a

labeled dataset.

Workers do their jobs. We capture what they do — structured, timestamped, ready for policy learning. No special setup. No staged environments.

Timestamped cross-modal alignment to 5ms precision
Auto-annotated with human-verified spot checks
Delivered in HDF5, JSON, or custom format
No PII — workers are anonymized by default

Who we work with.

Robotics labs

Real demonstrations, not simulated ones. Train manipulation policies on data that reflects how humans actually move.

Foundation model teams

Need scale? Our warehouses run every day. The data doesn't stop.

Automation companies

Skip the bootstrapping phase. Start with humans doing the exact task your robot needs to learn.

Interested in a dataset or pilot? Reach out directly.

Get in touch