What Is Motion Blur in Robot Training Video?

Reach, grasp and release are the moments a manipulation policy most needs, and they are when the head and hands move fastest. The site states that RGB alone leaves a model to infer 3D hand structure it cannot see reliably through occlusion and motion blur. Firsthand records 3D hand pose at capture time rather than estimating it afterwards.

Firsthand’s head camera is global-shutter, 3840 × 2160 at 60 fps, so fast motion is not skewed by row-by-row exposure. Blur inside one exposure is a separate matter and depends on how long the exposure lasts. Exposure time, shutter speed and any blur metric are Not stated; they are set per program in the skill spec and checked on the Spec Pilot batch.

Motion blur is the smear that appears when the scene or the camera moves while one frame is being exposed. In first-person video both move at once: the head turns, the hand reaches and a held object swings. Blur softens edges and fingertips within a single frame. It does not bend the frame the way a rolling shutter does.

No. A rolling shutter reads a frame row by row and skews fast motion; a global shutter exposes every pixel at the same instant. Firsthand states a global-shutter head camera, so it avoids that skew. The site does not claim global shutter removes blur. The shutter type of the wrist and depth cameras is Not stated.

Frame rate is how often a frame is recorded, not how long each one is exposed. The site states 60 fps for the head and both wrist cameras and 30 fps for depth. A higher rate samples fast motion more finely, but the site also states it does nothing for a frame skewed by a rolling shutter. Whether 60 fps footage is sharp depends on exposure, which is Not stated.

The site states no blur flag and no blur reject rule. Fast frames ship uncut, including exposure hunting in dim light, and a low-light task is captured as found under the lighting condition named in the skill spec. The per-recording camera_dump.txt carries per-frame exposure and shutter data, so a buyer can check how frames were exposed.

Rejection happens on other stated criteria: cross-stream sync beyond a 2.0 ms ceiling, calibration at or above 0.28 px RMS reprojection, missing streams, labels that fail the spec, missing consent, the manipulated object out of frame for more than 15 percent of an episode, and more than 35 percent invalid depth pixels in the manipulation region. Rejected episodes go to the reject log and are recaptured at Firsthand’s cost.

The HANDS23 benchmark paper (arXiv 2403.16428) names motion blur from head movement, alongside heavy occlusion, as a reason egocentric 3D hand reconstruction is hard, and finds fast hand motion still challenging. EgoXtreme (arXiv 2603.25135) reports that applying deblurring gave no improvement in extreme conditions, while a tracking-based approach did gain. Neither evaluates Firsthand data, so this page attributes no result to it.

Custom egocentric data collection starts from a written skill spec, usually proved by a Spec Pilot before volume scales. Where coverage is deep, a signed spec can reach a validated first batch in a median of 48 hours; where contributors must be recruited first, the site gives three to four weeks. Send the task, lighting conditions, motion speed, streams and your blur bar, and ask for exposure targets and a check of per-frame exposure data on the pilot.

What is motion blur in egocentric video?

It is smearing in a frame when the camera or a hand moves during one exposure, common in fast reach and grasp.

Is global shutter video free of motion blur?

Not on its own. Global shutter removes row-by-row skew; blur inside an exposure depends on exposure time, which the site does not state.

What shutter and frame rate does Firsthand state?

A global-shutter head camera at 3840 × 2160 and 60 fps, two wrist cameras at 1920 × 1080 and 60 fps, and depth at 30 fps. Wrist shutter type is Not stated.

Does Firsthand reject blurred frames?

The site states no blur reject rule. Episodes are rejected on listed criteria, such as sync beyond 2.0 ms or the object out of frame for more than 15 percent.

Is motion blur the same as face blurring?

No. Face and plate blurring is a privacy step applied at ingest. Motion blur is an optical effect of movement during exposure.

Can I ask for low-blur footage in a custom dataset?

Yes. Put your blur bar and lighting conditions in the skill spec; exposure targets are set per program and checked on the Spec Pilot.

What does Firsthand not state about motion blur?

No exposure time, shutter speed, blur metric, blur reject threshold or wrist-camera shutter type.