By robot type

Matched to your embodiment — not over-specified.

Most Open X-Embodiment / DROID robots use 2-finger parallel grippers, so full 25-joint finger tracking is over-specified for the majority. We sell the stack your robot actually needs — and offer dexterous, force and legged tiers where they earn their place.

Humanoid / whole-body VLA

Multi-view RGB-D · proprioception · 7-DoF action · language · calibration

Environmental & demographic diversity a single lab rig can't produce.

Parallel-gripper manipulation

Multi-view RGB · 6-DoF wrist pose · gripper state · action label

The mainstream embodiment — matched, not over-specified with 25-joint tracking.

Dexterous 5-finger hand

+ per-frame 25-joint hand pose · optional tactile

Precise recovered hand action, occlusion-robust across viewpoints.

Contact-rich assembly

+ 6-axis force / torque · tactile · contact timestamps

The friction, shear and compression physics simulation breaks on.

Legged locomotion

IMU · joint encoders · optional terrain vision

Proprioception-dense trajectories for blind and perceptive policies.

Drone / UAV & specialized

IMU · depth/RGB · state → control commands

For flying platforms, IMU + depth/RGB paired to control commands matters more than manipulation modalities. Have a specialized embodiment or sensor need? We can co-design the capture — that's what the Custom tier is for.

Out of our lane, honestly: autonomous-vehicle perception (LiDAR + radar + CAN driving logs), surgical robotics (PHI-gated), and aerial multispectral agriculture are different data problems we don't pretend to serve. We stay where human-egocentric and teleoperation manipulation capture is the right tool.

Scope for your robot