Datasets:
The dataset is currently empty. Upload or create new data files. Then, you will be able to explore them in the Dataset Viewer.
EgoGaming VR Gameplay Dataset
First-person VR gameplay video from commercial PC VR games, synchronized frame by frame with measured 6DoF head and controller poses, velocities, controller inputs and tracking flags, all on one capture clock. Recorded from paid adult players who consented in writing. Every session is reviewed by a person.
This repository has no data files. The recordings are licensed directly. A free sample of 100 action clips, in RLDS and WebDataset with the data card, is available at https://data.egogaming.ai/#request.
- Specification: https://data.egogaming.ai/vr-gameplay-dataset
- Live demo (video, pose graph and figure on one clock): https://data.egogaming.ai/#demo-section
- Licensing: https://data.egogaming.ai/licensing
- FAQ: https://data.egogaming.ai/faq
What each recording contains
| Stream | Details |
|---|---|
| Video | The game's rendered left-eye view at the game's own frame rate (about 70-90 fps), H.264, silent by default |
| Head pose | 6DoF, metres in the game's stage space (y up, floor at 0), unit quaternion (x, y, z, w), every video frame |
| Controller poses | Left and right grip pose, 6DoF, every video frame |
| Velocity | Linear (m/s) and angular (rad/s) per device |
| Tracking flags | pos_valid, ori_valid, pos_tracked, ori_tracked per device and frame |
| Inputs | Button, trigger and stick input where the capture route records it |
| Events | Tracking loss, recenters, state changes |
| Hand joints | 26 per hand, only where the headset tracked hands |
Captured by the EgoGaming recorder, an OpenXR API layer (or an OpenVR route for older SteamVR titles), on PC VR headsets: mostly Meta Quest 3S linked to a PC, plus Quest 2 and Valve Index. More than 15 commercial PC VR games so far.
Formats
- RLDS (TFDS): one episode per clip (up to 60 s), 15 Hz steps: 320×180 image, 21-float state, velocity, flags, hand joints, 18-float per-device
action, episode metadata. - WebDataset: TAR shards, one sample per clip at the full recorded rate: 720p MP4 with per-frame pose, velocity and flag arrays, plus poses, hands and inputs Parquet and an events file.
- Plain files: Parquet, CSV or JSON with MP4. LeRobot and HDF5 are planned.
import tensorflow_datasets as tfds
ds = tfds.builder_from_directory("rlds/egogaming_vr_sample/1.0.0").as_dataset(split="train")
for ep in ds.take(1):
for step in ep["steps"].take(3):
print(step["observation"]["state"], step["action"])
Collection and consent
Players are adults, mostly in the United States, who sign up at egogaming.ai, pass a review, sign a player contract and agree in writing, on the website and again in the recorder, before anything is captured. They are paid for every hour that passes review. Each session has a consent record. Players can withdraw; licensees then delete the files within 30 days.
Privacy
Pseudonymized: no names, contact details, dates of birth, locations or payment details. VR movement can identify people, so the license treats every file as personal data and forbids any attempt to identify players.
Limits
- The video is a rendered game view, not the real world.
- Only the head and two controllers are measured; hand joints are rare.
- PC VR only.
- No demographics are collected.
Licensing
Custom recording (exclusive for 24 months, pay only for accepted hours), ready-made packages with video or motion only, and academic packs for non-commercial research. Annotated or unannotated. Quoted per project: https://data.egogaming.ai/licensing
Contact: jim@egogaming.ai · EgoGaming, United States
- Downloads last month
- 49