Ropedia introduced Xperience 2 in a preprint submitted on 12 November 2026, pairing 3D scans of real environments with first-person recordings of human tasks for physical AI research. The paper describes a larger collection than the initial public download: the dataset card identifies a first batch of 126 scenes, with further batches planned.
The paper’s collection and the public release differ
The preprint reports 1,589 scans in its released scan component. Its 4,230 recordings, 74 tasks and 131 hours describe the broader collection inventory. Those totals should not be read as a count of files currently available from the public repository.
The dataset card describes a staged publication process. The distinction extends beyond scene counts to whether recordings already share coordinates with the scans:
| Component | Paper or project scope | Public release description |
|---|---|---|
| Scene scans | The paper reports 1,589 scans. | The dataset card identifies 126 scenes in the completed first batch. |
| Human recordings | The paper reports 4,230 recordings in the collection inventory. | The inventory total does not establish that all recordings are downloadable. |
| Scene registration | The project provides software to align recordings with scanned scenes. | The card says recording-to-scan registration is not yet part of the first-batch files. |
| Reuse terms | The code repository specifies Apache 2.0. | The dataset card specifies CC BY-NC 4.0 for data. |
Connecting environments with human behavior
Ropedia’s project page describes a two-stage capture process: an operator scans a space with handheld LiDAR, then records tasks with the head-mounted HOMIE Gen 2 device. Multiple task recordings can therefore relate to one scanned environment, rather than treating each human demonstration as a video detached from its surroundings.
The accompanying code repository provides scene registration software that aligns a processed HOMIE recording with a KOSMO scene and exports trajectories and estimated hand and body motion in the scene frame. That software is a separate resource from a download of already aligned data.
The authors also describe simulation, human-to-robot motion retargeting, virtual teleoperation and generated robot trajectories as applications. These are demonstrations reported in a preprint; they do not establish general improvements in deployed robots.
Why release scope matters for university research
For robotics laboratories, the practical significance is a resource that connects human demonstrations with measurable scene geometry. For a reproducible AI experiment, however, the relevant unit is the batch and its available components. A published collection total cannot substitute for the specific files and coordinate alignment used in an experiment.
The authors state that hand and body motion are estimated and activity captions are model-generated. Their three benchmarks use the same 20 scanned scenes. These qualifications matter for training data provenance and for interpreting results beyond the evaluated subset.
For research offices supporting university-industry collaborations, the release also separates access from reuse permissions. The dataset card describes attribution and noncommercial use under CC BY-NC 4.0, while the code has a different licence. The research implication is that training data composition, release scope and data permissions remain distinct from the availability of the processing software.
Sources
- Li and colleagues, Ropedia Xperience 2, arXiv preprint and submission history. Checked 2026-11-15.
- Ropedia, Xperience 2 research project. Checked 2026-11-15.
- Ropedia, Xperience 2 dataset card and release terms. Checked 2026-11-15.
- Ropedia, Xperience 2 code repository. Checked 2026-11-15.







