Datasets:

XraiSyntheticInterpolations
/

100_FAST_Synths

pretty_name: 100 FAST Synths
annotations_creators:
  - machine-generated
tags:
  - synthetic-data
  - virtual-reality
  - xr
  - motion-tracking
  - time-series
  - tabular
  - user-identification
  - biometrics
  - pandas

100 FAST Synths

Dataset Description

100 FAST Synths is a synthetic extended reality (XR) motion dataset intended for machine-learning research on motion-based user identification. It contains motion sequences for 100 synthetic people derived from participant pairs in the Full-scale Assembly Simulation Testbed (FAST) Dataset.

The dataset follows the two assembly conditions in FAST:

FAB_A: motion during assembly condition/build A
FAB_B: motion during assembly condition/build B

Every synthetic person has one CSV file in FAB_A and one corresponding CSV file in FAB_B, for a total of 200 motion files. This supports experiments such as training on one assembly condition and evaluating identification on the other.

The associated paper, "Generating Synthetic Behavioral Populations from XR Motion" by Xiaozheng Wang and Ryan P. McMahan, has been accepted for publication at the 2026 Future of HCI Workshop (FutureHCI '26).

Dataset Contents

.
|-- pair_to_sy_map.csv
`-- Syn_Inspector180/
    |-- FAB_A/
    |   |-- Female/   # 46 files
    |   `-- Male/     # 54 files
    `-- FAB_B/
        |-- Female/   # 46 files
        `-- Male/     # 54 files

Summary:

Item	Count
Synthetic people	100
Motion files per synthetic person	2
Motion CSV files	200
`FAB_A` files	100
`FAB_B` files	100
Female synthetic IDs	46
Male synthetic IDs	54

Motion filenames follow the pattern:

Sy<synthetic-person-id>_<BuildA-or-BuildB>.csv

For example, the files Sy008369F_BuildA.csv and Sy008369F_BuildB.csv contain the two build conditions for the same synthetic person.

Motion File Schema

Each motion CSV contains 49 columns. Every row is a time sample with tracking data for the head, left hand, and right hand.

Field group	Columns	Description
Time	`Timestamp`	Sample timestamp
Position	`<Tracker>_position_{x,y,z}`	3D position
Quaternion	`<Tracker>_quat_{x,y,z,w}`	Quaternion orientation
6D rotation	`<Tracker>_sixD_{a,b,c,d,e,f}`	Six-dimensional rotation representation
Euler angles	`<Tracker>_euler_{x,y,z}`	Euler orientation in degrees, in the range [-180, 180]

<Tracker> is one of Head, LeftHand, or RightHand.

Important column-order note

The files use two column-order variants. In 66 files, the RightHand position/quaternion/6D fields precede the corresponding LeftHand fields; in 134 files, the left-hand fields precede the right-hand fields. Field names are consistent across both variants. Load columns by name rather than by numeric position.

The coordinate system, position units, timestamp units, Euler convention/order, and precise 6D rotation convention should be confirmed from the associated data-generation documentation before geometric reconstruction.

Synthetic-Person Mapping

pair_to_sy_map.csv contains 100 rows and three columns:

Column	Description
`personA_id`	First FAST participant used to synthesize the person
`personB_id`	Second FAST participant used to synthesize the person
`sy_id`	Resulting synthetic-person identifier used in motion filenames

Each sy_id maps to exactly one pair of source participant IDs.

Loading the Data

The CSV files can be loaded directly with pandas:

from pathlib import Path
import pandas as pd

root = Path("Syn_Inspector180")
motion_files = sorted(root.glob("FAB_*/*/*.csv"))

df = pd.read_csv(motion_files[0])

# Select by name so both column-order variants are handled correctly.
head_position = df[[
    "Head_position_x",
    "Head_position_y",
    "Head_position_z",
]]

mapping = pd.read_csv("pair_to_sy_map.csv")

Intended Uses

The dataset is intended for research on:

XR motion-based user identification and recognition
Cross-condition identification between FAB_A and FAB_B
Evaluation of motion representations, including position, quaternion, 6D rotation, and Euler angles
Synthetic-data methods for XR motion research
Robustness and generalization studies for motion-based machine-learning systems

Limitations and Responsible Use

The data are synthetic and derived from participant pairs in FAST. Results on this dataset may not generalize to motion from real individuals, other XR applications, other tracking systems, or other populations.
The dataset contains only the two FAST assembly conditions represented by FAB_A and FAB_B.
The Female and Male folders reflect the categories used when constructing this release. They should not be treated as a complete representation of gender.
Synthetic data can preserve characteristics of its source data. The mapping file exposes the pseudonymous FAST participant IDs used for each synthetic identity.
Motion-based identification is a biometric application. Researchers should consider privacy, consent, misuse, surveillance risk, demographic bias, and the consequences of false identification.
This dataset should not be used to identify real people, make high-stakes decisions, or infer sensitive attributes.

Source Dataset

This dataset is derived from participants in the Full-scale Assembly Simulation Testbed (FAST) Dataset, which contains VR tracking and interaction data from people learning to assemble two full-scale structures.

FAST paper:

Alec G. Moore, Tiffany D. Do, Nayan N. Chawla, Antonia Jimenez Iriarte, and Ryan P. McMahan. "The Full-scale Assembly Simulation Testbed (FAST) Dataset." 2024. https://arxiv.org/abs/2403.08969

@article{moore2024fast,
  title   = {The Full-scale Assembly Simulation Testbed (FAST) Dataset},
  author  = {Moore, Alec G. and Do, Tiffany D. and Chawla, Nayan N. and
             Iriarte, Antonia Jimenez and McMahan, Ryan P.},
  journal = {arXiv preprint arXiv:2403.08969},
  year    = {2024},
  doi     = {10.48550/arXiv.2403.08969}
}

Citation

If you use this dataset, please cite the associated FutureHCI publication and the FAST paper.

Temporary citation for the accepted paper (to be updated with pages and DOI after publication):

Xiaozheng Wang and Ryan P. McMahan. 2026. "Generating Synthetic Behavioral Populations from XR Motion." Accepted for publication in the Proceedings of the 2026 Future of HCI Workshop (FutureHCI '26). Forthcoming.

@inproceedings{wang2026synthetic,
  title     = {Generating Synthetic Behavioral Populations from {XR} Motion},
  author    = {Wang, Xiaozheng and McMahan, Ryan P.},
  booktitle = {Proceedings of the 2026 Future of HCI Workshop (FutureHCI '26)},
  year      = {2026},
  note      = {Accepted; forthcoming}
}

License

A separate license has not yet been specified for this synthetic release. Users are responsible for confirming the terms applicable to the source FAST Dataset and this derived dataset before redistribution or commercial use. Repository maintainers should add an explicit license before broad reuse.