What is Motion Capture Data?

Optical systems (Vicon and OptiTrack are the standard names), track reflective markers with cameras at 100 to 240 Hz and resolve positions to a fraction of a millimeter inside a calibrated volume. The characteristic failure is occlusion: a marker hidden by the body drops out, and the gap is filled by software and checked by hand. Cleanup routinely takes longer than the capture session itself.

Inertial suits are the portable alternative: no cameras, capture anywhere, but position drifts over time and the suit is calibrated to the wearer at the start of each session. Both types output skeleton data, not robot commands, and the retargeting step (mapping a human skeleton onto a robot with different proportions and fewer joints), is where a shoot becomes useful or does not.

A commercial optical stage rents for hundreds of dollars an hour in the US or EU, before cleanup, which is why the target robot decides the shoot: marker set, frame rate and capture volume all follow from the machine that will imitate the motion. If the target has few joints, VR teleoperation may capture what is needed for less. Human-motion archives exist (AMASS unifies 15 capture datasets into one skeleton format), and a custom shoot for a specific robot can be commissioned to order.

Types of motion capture systems

Optical systems track reflective markers with cameras and are the reference standard for accuracy. They are also the most constrained: the capture volume is finite, markers drop out when the body hides them, and the gaps are filled by software and checked by hand afterwards. Cleanup is routine, not exceptional.

Inertial suits trade accuracy for portability: no cameras and no fixed volume, so capture can happen anywhere, at the cost of drift that accumulates through the session and a calibration step every time the suit goes on. Video-based estimation is the newest and least precise option: no suit and no stage, at the price of noisier joint estimates and more post-processing before the motion is usable.

Capture scope is the third variation. Full body for locomotion and gross motion; upper body and hands when the target task is manipulation; one subject or several. Whether props and objects are tracked alongside the body also changes the shoot, and for robot learning that often matters as much as the skeleton, because the hand-object relationship is the behavior being copied.

What to specify when buying

Start from the target robot, not the motion. The joint count and proportions of the machine that will imitate the motion decide the marker set, the retargeting convention and whether an optical shoot is worth commissioning. If the target machine has a small joint set, a different capture route may be cheaper than a stage shoot; it is worth asking which one the supplier would use.

Describe the motion and the space it needs. Fine hand motion needs a different setup from walking, and a capture volume that fits a few steps cannot record a warehouse route. A task list with the movement style and the space it requires is more useful to a capture studio than a total duration, and it makes quotes comparable between suppliers.

Agree the cleanup and the delivery format. Ask what happens to frames where markers were hidden: how the gaps are filled, and whether anyone reviews the result before it ships. State the skeleton format and frame rate the pipeline expects, and ask whether the version retargeted to your robot ships with the data. Where a public archive holds similar motion, licensing it is worth pricing against a custom shoot.