Egocentric Video Datasets with Synchronized IMU

Quick facts

Resolution

1080p @ 30fps baseline; 2160p for close-range dexterity work

Field of view

≥120° horizontal so the moving hands and workspace stay in frame through head turns

Mount

Head-mounted rig with a calibrated IMU rigidly coupled to the camera; optional wrist or waist IMU for whole-body motion

Sensors

RGB (baseline), Head IMU: 3-axis accelerometer + 3-axis gyroscope (baseline), Optional wrist/body IMU, Optional gaze and depth

Labels

Frame-to-IMU sync markers with a stated tolerance in milliseconds; Per-device IMU-camera calibration parameters (extrinsics + time offset); Frame-aligned action segments

Volume

40–160 accepted hours per program

What IMU-synced data contains

An IMU-synced egocentric dataset pairs head-mounted video with time-aligned inertial (accelerometer + gyroscope) streams, so a model gets the ego-motion and fine action dynamics the camera alone misses. This is head-mounted video+IMU, not the bare wrist- or waist-worn inertial logs that dominate human-activity-recognition datasets. The open corpora that carry synchronized IMU — Ego4D, Aria Everyday Activities, Nymeria — are signed-license or non-commercial, so calibrated custom capture is the commercial route.

The capture settings this covers:

• A worker pushing a parts cart across a shop floor — the gait, the stops, and the weight shifts register on the accelerometer trace before the object of interest ever enters frame.
• Reaching high onto a shelf and back down: the head pitches, the torso leans, and the gyroscope captures the arc the RGB feed only implies.
• Rapid head turns while scanning a workspace — the exact motion that smears a single video frame but reads cleanly as an angular-velocity spike on the IMU.
• Whole-body activity: crouching to a low bin, standing, then pivoting to a bench — locomotion and posture changes a head-mounted camera alone can't separate from a static scene.
• Fine manipulation where the hand and tool micro-jitter at contact — high-rate inertial samples resolve the grasp and release moments that fall between video frames.
• Stair climbs and uneven-terrain walking, where per-step vertical acceleration is the signal a locomotion or ego-motion policy actually learns from.

Why robotics and AI labs need IMU-synced data

Apple's EgoDex pretrains dexterous manipulation on large-scale egocentric human video, and the head-and-wrist IMU streams synchronized to that video are what make the motion machine-readable for a policy. ^[1]

EgoScale found dexterous-manipulation performance scaling with the volume and diversity of egocentric human data, which turns 'capture more synchronized IMU clips' into a measurable ROI lever rather than a hunch. ^[2]

EgoLive shows large-scale egocentric human demonstrations of real-world tasks lifting manipulation policies, so IMU-synced first-person capture directly improves the robot skill it targets. ^[3]

AoE frames scalable, low-cost collection of egocentric human video of manual tasks as an answer to the data scarcity an IMU-synchronized capture program is built to close. ^[4]

Capture and delivery spec

Every IMU-synced capture program runs to an explicit spec so the footage is training-ready on delivery rather than after a re-shoot. The baseline below is tuned per program; sensors, labels, and volume scale with the buyer's model.

Open IMU-synced datasets

The 4 open corpora most relevant to IMU-synced are compared below on scale, sensors, license, commercial use, and the gap each leaves for a buyer. None of them are cleanly licensed for commercial model training — which is the whole reason custom capture exists.

Open datasets vs Truelabel custom capture

Every corpus with clean, synchronized IMU is legally locked. Aria Everyday Activities and Nymeria are non-commercial research licenses, and Ego4D is a signed agreement where only a share of the hours carry IMU at all. You can benchmark ego-motion on them; you cannot ship a policy trained on them.

Open sets hand you whatever rate and placement the original device used. Custom capture guarantees the IMU sample rate, the frame-sync tolerance, and the sensor placement your model consumes — head-only for gaze-and-scan work, or head plus wrist and waist for whole-body locomotion — instead of reverse-engineering someone else's rig.

Sync and calibration are the whole game, and open corpora rarely document either per device. We validate IMU-camera calibration on every collector device before a batch is accepted and ship the extrinsics and time offset as a per-device calibration JSON, so the inertial stream and the pixels actually line up.

Inertial traces are motion signatures of a specific person and place, so provenance is not optional. Custom capture gives you a per-clip consent chain and the option of exclusivity, so the same motion data isn't also sold to the lab training against you.

IMU-synced: by the numbers

The figures below are specific to IMU-synced egocentric data and anchor the comparisons above.

• Project Aria glasses carry dual IMUs running at roughly 800 Hz and 1 kHz.
• Aria Everyday Activities: ~7.3 hours of dual-IMU everyday-activity recordings, non-commercial.
• Nymeria pairs Aria inertial data with in-the-wild full-body motion capture ground truth (non-commercial).
• Only a share of Ego4D's 3,670 hours carries synchronized IMU.
• "imu dataset" 10/mo and "imu sensor data" 20/mo (DataForSEO, 2026-07-06).

How Truelabel captures IMU-synced data

Truelabel runs IMU-synced programs on a network of 20,000+ consented collectors across nine countries, capturing to your brief on a head-mounted rig. Every clip passes per-clip machine QA — head-mount stability, field of view, and hands-in-frame coverage — and ships with a signed wearer consent artifact and provenance manifest. A calibration pilot returns its first batch in days, then accepted batches scale to 40–160 accepted hours per program, delivered as H.265 video + synchronized IMU streams + a per-device calibration JSON, Hugging Face-streamable, with a consent artifact on every clip. Go deeper via what egocentric data is, egocentric data licensing, VLA training data, teleoperation data, how datasets are delivered, and physical AI data marketplace.

Related pages

Use these to move from category-level context into specific task, dataset, format, and comparison detail.

External references and source context

1. EgoDex: Learning Dexterous Manipulation from Large-Scale Egocentric Video

   EgoDex is a large-scale egocentric video corpus built to pretrain dexterous manipulation from first-person human video.

   arXiv ↩
2. EgoScale: Scaling Dexterous Manipulation with Diverse Egocentric Human Data

   EgoScale reports dexterous-manipulation performance scaling with the volume and diversity of egocentric human data.

   arXiv ↩
3. EgoLive: A Large-Scale Egocentric Dataset from Real-World Human Tasks

   EgoLive is a large-scale egocentric dataset of real-world human tasks used to lift manipulation policies.

   arXiv ↩
4. AoE: Always-on Egocentric Human Video Collection for Embodied AI

   AoE frames scalable, low-cost collection of egocentric human video of manual tasks as an answer to embodied-AI data scarcity.

   arXiv ↩
5. Egocentric video remains useful but incomplete for robot data buyers

   Ego4D is a large egocentric corpus under a signed license in which only a subset of hours carries synchronized IMU.

   ego4d-data.org
6. Aria Everyday Activities (AEA)

   Aria Everyday Activities provides ~7.3 hours of everyday-activity egocentric recordings on dual-IMU Aria glasses under a non-commercial research license.

   Meta / Project Aria
7. Nymeria: egocentric full-body motion dataset

   Nymeria pairs Aria egocentric video and inertial data with in-the-wild full-body motion capture under a non-commercial research license.

   Meta / Project Aria
8. EgoVid-5M

   EgoVid-5M layers kinematic action annotations over Ego4D-sourced video rather than adding measured inertial data.

   EgoVid-5M project

FAQ