Gilfoyle727/vr-ray-pointer-landing-pose

---
pretty_name: VR Ray Pointer Landing Pose Dataset
task_categories:
- time-series-forecasting
- other
language:
- en
tags:
- virtual-reality
- vr
- raycasting
- multimodal
- eye-tracking
- motion-capture
- time-series
- human-computer-interaction
size_categories:
- 1M<n<10M
configs:
- config_name: raw_archives
data_files:
- split: study1
path: Study1_Raw.zip
- split: study2
path: Study2_Raw.zip
license: other
---

# VR Ray Pointer Landing Pose Dataset

This dataset accompanies the paper **"Predicting Ray Pointer Landing Poses in VR Using Multimodal LSTM-Based Neural Networks."** It contains the raw trajectory archives used for the paper's two user studies, plus the original data processing code used to prepare model inputs.

Paper link: [IEEE Xplore](https://ieeexplore.ieee.org/abstract/document/10937427)

The data captures bare-hand raycasting selection behavior in VR with multimodal time-series signals from hand, head-mounted display (HMD), and gaze channels. The paper reports that the full dataset covers **72,096 trials** across two empirical studies:

- Study 1: 55,296 trials
- Study 2: 16,800 trials

## Paper Summary

The paper studies target-agnostic prediction of the final ray landing pose during VR pointing and selection. The proposed model is an LSTM-based predictor trained on time-series features derived from three modalities:

- hand movement
- HMD movement
- eye gaze movement

According to the paper:

- Study 1 recruited **16 participants**
- Study 2 recruited **8 new participants**
- Data was recorded at **90 Hz**
- Hardware used a **Meta Quest Pro**
- The model achieved an average prediction error of **4.6 degrees at 50% movement progress**

## Included Files

- `Study1_Raw.zip`
Raw CSV trajectories for Study 1.
- `Study2_Raw.zip`
Raw CSV trajectories for Study 2.
- `Dataprocessing_code.zip`
Original preprocessing scripts provided by the authors.
- `data_processing_code/`
Extracted copy of the preprocessing scripts for easier browsing on Hugging Face.

## Data Format

Each raw archive contains per-participant CSV files with frame-level trajectories. Typical columns include:

- participant / block / trial identifiers
- error flag
- target geometry variables such as depth, theta, phi, width, and position
- task progress and distance traveled percentage
- timestamp
- HMD position and forward vector
- hand position and forward vector
- left-eye position and forward vector
- right-eye position and forward vector
- target location and target scale

The data is sampled over time during reciprocal pointing selections.

## Study Design From The Paper

### Study 1

The paper describes Study 1 as a within-subjects design over:

- target depth combinations: `De` and `Ds` in `{3m, 6m, 9m}`
- theta values: `10, 15, 20, 25, 50, 75` degrees
- phi values: `0` to `315` degrees in `45` degree steps
- target widths: `4.5` and `9` degrees

The paper reports:

- `55,296` total trials
- `16` participants
- reciprocal 3D pointing with no distractors

### Study 2

The paper describes Study 2 as a validation study with:

- `8` new participants
- theta varying continuously across all integer values from `15` to `84` degrees
- `350` trial combinations
- `50` blocks
- `6` reciprocal selections per trial combination
- `2,100` trials per participant

The paper reports `16,800` total trials for Study 2.

## Important Notes About The Raw Archives

This repository preserves the raw files exactly as provided by the dataset owner. A few practical details matter when using the archives:

- `Study1_Raw.zip` currently contains **19 CSV files**
- `Study2_Raw.zip` currently contains **8 CSV files**
- the observed raw trial counts are **64,308** trials in `Study1_Raw.zip` and **16,800** trials in `Study2_Raw.zip`
- some Study 1 CSV files do **not** include a `ParticipantID` column in the header
- some Study 1 and Study 2 files share participant-like file IDs such as `72`
- raw archive contents therefore do not map one-to-one to the participant counts reported in the paper without additional curation context
- specifically, `Study1_Raw.zip` includes a `72_Trajectory.csv` file with **2,100** trials, which matches the Study 2 per-participant protocol rather than the Study 1 per-participant total of **3,456** trials reported in the paper

For reproducibility, this repository keeps the original archives unchanged. When reconstructing participant identity for Study 1, you may need to use the filename as the participant identifier when `ParticipantID` is absent from the CSV header.

## Recommended Usage

- Use `Study1_Raw.zip` and `Study2_Raw.zip` as the authoritative raw data sources.
- Use the scripts in `data_processing_code/` to reproduce feature engineering and preprocessing.
- If you build a Hugging Face `datasets` loader on top of this repository, treat the raw zip files as the source of truth rather than assuming fully standardized CSV schemas.

## Citation

If you use this dataset, please cite the paper:

```bibtex
@inproceedings{xu2025predictingray,
title={Predicting Ray Pointer Landing Poses in VR Using Multimodal LSTM-Based Neural Networks},
author={Xu, Wenxuan and Wei, Yushi and Hu, Xuning and Stuerzlinger, Wolfgang and Wang, Yuntao and Liang, Hai-Ning},
booktitle={IEEE Conference on Virtual Reality and 3D User Interfaces},
year={2025}
}
```

## Acknowledgements

This dataset was collected for the paper above and uploaded to Hugging Face by the dataset owner.

---
pretty_name: VR射线指针着陆位姿数据集（VR Ray Pointer Landing Pose Dataset）
task_categories:
- 时间序列预测（time-series-forecasting）
- 其他
language:
- 英语（en）
tags:
- 虚拟现实（Virtual Reality）
- VR
- 光线投射（raycasting）
- 多模态（multimodal）
- 眼动追踪（eye-tracking）
- 动作捕捉（motion-capture）
- 时间序列（time-series）
- 人机交互（Human-Computer Interaction）
size_categories:
- 100万<样本数<1000万
configs:
- config_name: 原始归档文件
data_files:
- split: 研究1
path: Study1_Raw.zip
- split: 研究2
path: Study2_Raw.zip
license: 其他（other）
---

# VR射线指针着陆位姿数据集（VR Ray Pointer Landing Pose Dataset）

本数据集配套于论文**《基于多模态LSTM神经网络的VR射线指针着陆位姿预测》**（原标题：*Predicting Ray Pointer Landing Poses in VR Using Multimodal LSTM-Based Neural Networks*），包含支撑该论文两项用户研究的原始轨迹归档文件，以及用于预处理模型输入的原始数据处理代码。

论文链接：[IEEE Xplore](https://ieeexplore.ieee.org/abstract/document/10937427)

该数据集采集了虚拟现实环境下徒手光线投射选择行为的数据，包含来自手部、头戴式显示器（Head-Mounted Display, HMD）以及视线通道的多模态时间序列信号。据论文所述，完整数据集涵盖两项实证研究的**72096次试验**：
- 研究1：55296次试验
- 研究2：16800次试验

## 论文概述

该论文研究VR指向与选择过程中，目标无关的最终射线着陆位姿预测问题。所提出的模型为基于长短期记忆网络（Long Short-Term Memory, LSTM）的预测器，以从三种模态提取的时间序列特征为训练输入：
- 手部运动
- 头戴式显示器运动
- 眼动追踪运动

据论文披露：
- 研究1招募了**16名参与者**
- 研究2招募了**8名新参与者**
- 数据采集帧率为**90Hz**
- 实验硬件采用**Meta Quest Pro**
- 该模型在运动进度达50%时的平均预测误差为**4.6度**

## 包含文件

- `Study1_Raw.zip`：研究1的原始CSV轨迹文件
- `Study2_Raw.zip`：研究2的原始CSV轨迹文件
- `Dataprocessing_code.zip`：作者提供的原始预处理脚本
- `data_processing_code/`：Hugging Face平台上便于浏览的预处理脚本解压副本

## 数据格式

每个原始归档文件包含按参与者划分的、带帧级轨迹信息的CSV文件。典型列字段包括：
- 参与者、分组、试验标识符
- 错误标记
- 目标几何变量，如深度、极角θ、方位角φ、宽度与位置
- 任务进度与行进距离百分比
- 时间戳
- 头戴式显示器的位置与前向向量
- 手部位置与前向向量
- 左眼位置与前向向量
- 右眼位置与前向向量
- 目标位置与目标缩放比例

该数据在往复指向选择任务中按时间采样采集。

## 论文中的实验设计

### 研究1

论文将研究1描述为被试内设计，实验变量包括：
- 目标深度组合：`De`与`Ds`，取值为`{3m, 6m, 9m}`
- 极角θ值：`10、15、20、25、50、75`度
- 方位角φ值：以`45`度为步长，取值范围`0`至`315`度
- 目标宽度：`4.5`与`9`度

据论文报告：
- 总试验次数：`55296`次
- 参与人数：`16`名
- 无干扰物的三维往复指向任务

### 研究2

论文将研究2描述为验证性研究，实验设置包括：
- `8`名新参与者
- 极角θ在`15`至`84`度的所有整数值间连续变化
- `350`种试验组合
- `50`个分组
- 每种试验组合对应`6`次往复选择
- 每名参与者完成`2100`次试验

据论文报告，研究2总试验次数为`16800`次。

## 关于原始归档文件的重要说明

本仓库完整保留了数据集所有者提供的原始文件。使用归档文件时需注意以下实操细节：
- `Study1_Raw.zip`当前包含**19个CSV文件**
- `Study2_Raw.zip`当前包含**8个CSV文件**
- 经统计，`Study1_Raw.zip`中实际包含**64308**次试验，`Study2_Raw.zip`中包含**16800**次试验
- 部分研究1的CSV文件的表头中**不含`ParticipantID`列**
- 部分研究1与研究2的文件共享类似参与者的文件标识符，例如`72`
- 若无额外的整理上下文，原始归档文件的内容无法与论文中报告的参与者数量一一对应
- 具体而言，`Study1_Raw.zip`中包含一个`72_Trajectory.csv`文件，内含**2100**次试验，这与研究2的单参与者试验协议相符，而非论文中报告的研究1单参与者**3456**次试验总量

为保证可复现性，本仓库保留原始归档文件的原貌。若需为研究1重构参与者身份，当CSV表头中缺失`ParticipantID`时，可将文件名作为参与者标识符。

## 推荐使用方式

- 以`Study1_Raw.zip`与`Study2_Raw.zip`作为权威原始数据源
- 使用`data_processing_code/`中的脚本复现特征工程与预处理流程
- 若基于本仓库构建Hugging Face `datasets`加载器，请将原始压缩文件作为事实来源，而非假设CSV格式完全标准化

## 引用方式

若使用本数据集，请引用以下论文：

bibtex
@inproceedings{xu2025predictingray,
title={"Predicting Ray Pointer Landing Poses in VR Using Multimodal LSTM-Based Neural Networks"},
author={Xu, Wenxuan and Wei, Yushi and Hu, Xuning and Stuerzlinger, Wolfgang and Wang, Yuntao and Liang, Hai-Ning},
booktitle={IEEE Conference on Virtual Reality and 3D User Interfaces},
year={2025}
}

## 致谢

本数据集为上述论文所采集，并由数据集所有者上传至Hugging Face平台。