HIW-500 — Rerun RRD conversion

Rerun recordings converted from the full BitRobot/HIW-500 humanoid-teleoperation dataset. Each episode is written into multiple .rrd files that share one recording_id, one file per layer, ready to view in the Rerun Viewer or to register as a catalog dataset.

Layers live in their own directories, so a single episode is spread across the bucket like this:

base/<task>__<session>__<episode_NNNN>.rrd   # every MCAP message as recorded, plus the split head stereo ('base' layer)
derived_archetypes/…                         # the '/wbc_lerobot' payload as a struct, plus its end-effector poses
urdf/…                                       # forward kinematics
odom/…                                       # the pelvis pose in the odometry frame
cameras/…                                    # head stereo intrinsics and poses
ir/…                                         # wrist infrared streams
properties/…                                 # per-episode static info (task, scene, duration, …)
assets/urdf-model.rrd                        # the Unitree G1 meshes, once for the whole dataset
blueprints/default.rbl                       # default viewer layout

Provenance and attribution

This dataset is a converted derivative of BitRobot/HIW-500, released by its original authors under the Creative Commons Attribution 4.0 International license (source revision 2ca7ffcd, dated 2026-06-29).

All credit for the underlying data collection belongs to the original authors. The original authors do not endorse this conversion.

Changes from the source

• Each episode's ROS 2 MCAP, its info.json, and its calibration/ sidecars become Rerun recordings.
• The stereo head camera, originally stacked into one, has an extra copy split into /camera/head/left and /camera/head/right.
• Calibration sidecars become named fields, for example ir1.intrinsics.fx and camera_matrix_left.

Notes on file size

The MCAP-to-RRD conversion itself does not increase the file size, but the extra layers and the duplicated head split increase the total size.

The table below shows examples of file sizes: the "base − split + ir" column is the full converted image and signal payload, corresponding to the MCAP source.

Sizes are in MB (10⁶ bytes), sub-MB layers in KB. "3D view" combines the urdf, cameras, and odom layers.

Measured on these source episodes:

Clothes-Washing : episode_2026-05-25_16-41-06/episode_0002/episode_0002.mcap
Move-The-Pillow-To-The-Sofa-From-Floor : episode_2026-02-24_14-31-06/episode_0001/episode_0001.mcap
Picking-Trash-To-Rubbish-Bin : episode_2026-04-13_09-30-35/episode_0002/episode_0002.mcap
Sweep-Floor : episode_2026-05-19_14-35-04/episode_0006/episode_0006.mcap

Scope

Conversion code is published at rerun-io/rrd-datasets. The code and these recordings demonstrate the conversion; they are not a versioned dataset release. Files may be regenerated or replaced, and converting the same episode twice does not give identical bytes.

How to use

Fetch one episode from the bucket and view it locally:

pip install "huggingface_hub[cli]" rerun-sdk

# Every layer of one episode, plus the shared robot model and the default blueprint
hf buckets sync hf://buckets/rerun/hiw-500 . \
  --include "*/<task>__<session>__<episode_NNNN>.rrd" \
  --include "assets/urdf-model.rrd" \
  --include "blueprints/default.rbl"

rerun */<task>__<session>__<episode_NNNN>.rrd blueprints/default.rbl

Keep the * so every layer loads.

Opened as above, the 3D view shows the end-effector markers and the head camera but no robot. To see the posed robot in 3D, register the episodes on a catalog with assets/urdf-model.rrd as the dataset's asset. The Local Catalog section in the code repo shows how.

These recordings were written with Rerun 0.37.0. A newer viewer reads them as they are in most cases; rerun rrd migrate updates them if it cannot.

To copy the whole bucket, mirror it bucket-to-bucket rather than to a local disk. The bucket is reachable over an S3-compatible API, so rclone can stream it straight into your own S3-compatible storage.

License

Creative Commons Attribution 4.0 International, the same license as the source dataset — the full text ships as the LICENSE file at the root of this bucket. If you redistribute or build on this data, keep the attribution above and comply with the license terms.

The shared robot model assets/urdf-model.rrd is the one file not derived from the HIW-500 dataset. It carries geometry from Unitree's g1_29dof_mode_15_with_dex1_1.urdf, taken from unitree_ros. Unitree distributes that model under the BSD-3-Clause license, which governs that file.