README.md

metadata

license: other
license_name: worldrover-research
pretty_name: WorldRover (lite)
task_categories:
  - depth-estimation
  - robotics
tags:
  - video
  - panoramic
  - equirectangular
  - depth
  - camera-pose
  - embodied-ai
  - world-model
  - synthetic
size_categories:
  - 100B<n<1T

WorldRover

Paired 360-panoramic and first-person video of photoreal 3D environments, with per-frame depth, camera pose and action labels. Four scenes, ~30 min of each view per scene — 129 clips per view, ~4.1 h of video at 30 fps.

This repository is the lite subset: RGB + camera pose + actions + metadata, no depth (~103 GB instead of ~800 GB). Lossless 16-bit depth is 87% of the bytes, so it lives in the per-scene repositories — take it only for the scenes you need.

Scene	Clips per view	Per view	Full repo (with depth)
med_village	13	32.1 min	WorldRover-med_village — 270 GB
paris	23	30.7 min	WorldRover-paris — 152 GB
venice	46	30.4 min	WorldRover-venice — 154 GB
art_nouveau	47	30.2 min	WorldRover-art_nouveau — 164 GB

What makes it paired

pano/<clip_id> and fp/<clip_id> share the same camera path frame for frame: the first-person clip was rendered from the panoramic clip's per-frame trajectory, so the two camera_trajectory.csv files match exactly. That gives the same world state under two very different projections without relying on an interpolated alignment.

Clips are 35 s to 3.5 min of continuous motion — no cuts, no teleports.

Tools

git clone https://github.com/AlayaLab/WorldRover
pip install -r WorldRover/requirements.txt

from worldrover import Clip
clip = Clip("venice/fp/venice_000003")
rgb, depth_m = clip.rgb_frame(100), clip.depth_frame(100)   # sRGB uint8; planar metres
pts = clip.points_world(100)                                # world points, centimetres

Three conventions the tools handle for you, and that are easy to get wrong by hand: depth codes are log-quantized and store radial distance (convert before unprojecting); camera_trajectory.csv has n_frames + 1 rows (the last is the closing keyframe, the authoritative count is depth/depth.meta.json); poses are Unreal-style — left-handed, centimetres, X-forward / Y-right / Z-up, camera looking down its own +X.

Related

• Collection (all parts in one place): https://huggingface.co/collections/xjxu21/worldrover-6a851193b19350ca6de9f424
• Lite subset (no depth, ~103 GB): https://huggingface.co/datasets/xjxu21/WorldRover
• Full per-scene: med_village · paris · venice · art_nouveau
• Tools: https://github.com/AlayaLab/WorldRover

License

Rendered video, depth, camera pose and action labels are released for research use. The underlying 3D environments are third-party commercial assets, are not redistributed here, and each clip's description.json records its asset pack and license. Tools are MIT.