We introduce a multi-modal physiological dataset named "Industrial VR Activity Self-Efficacy Dataset" (IVRASED) to study Self-efficacy of a learner during learning task in a virtual reality learning environment (VRLE), especially in an industrial use-case. (assembly task).

This dataset should allow learner analysis and modeling, in order to persue the goal to make an Intelligent Tutoring System (ITS) through the study of learner Self-efficacy.

The dataset can be downloaded from this link : https://zenodo.org/record/8136155

IVRASED dataset consists of two parts:

1. Physiological recording (EEG, ECG, GSR, Eye-tracking) of 15participants done in an industrial VRLE during assembly task for around 1hour long, including the VR recorded video.
2. The perceived self-efficacy of the current task self-assessed before and after the assemblage on a discrete 10-point scale (range of 1 to 10)

Assembly task	SEF answer

iMotions platform has been used to organize the study and aggregate all the sensors data.

The B-Alert-X10t provides 3 types of datas. Raw and decontaminated signals sampled at 256Hz and brain metric through iMotions platform at 1Hz. ECG signal is delivered along 3 leads at 512Hz , supplied with heartrate and interbeat interval at 1Hz. GSR sensor provide a raw signal sampled at 128Hz, both extracted GSR conductance and resistance, and a realtime analysis of peak properties. iMotions include a real-time ET analysis alongside pupillometry, giving clue about fixations and saccades properties.

We provide both raw and preprocessed data.

raw_data/
├── csv/
│   ├── (respondent_id)_(sequence_id).csv
│   └── ...
├── parquet/
│   ├── (respondent_id)_(sequence_id).parquet
│   └── ...

pre_processed_data/
├── iMotionsToPython/ #list of dataframe with data sampled to 128Hz
│   ├── BySequence/
│   │   ├── (respondent_id)_(sequence_id)_full_features_data.array
│   │   ├── (respondent_id)_(sequence_id)_annotation_data
│   │   └── (respondent_id)_(sequence_id)_dict
│   │   └── ...
│   └── AllSquence/ #Aggregated file
│   │   └── all_data_full_features_10_classes.array
├── processed_sensor_combination/(fold_number)/
│   ├── Sensor n x Sensor m x Sensor k/
│   │   ├── X_train.npy
│   │   ├── Y_train.npy
│   │   ├── X_test.npy
│   │   └── Y_test.npy
│   ├── Sensor n x Sensor m/
│   │   └── ...
│   └── ...

Original raw data recordings are exported through iMotions platform and saved into CSV format. We also provide the data under parquet format which is faster in read and takes less storage space.

There are 34 files. Each participant have consecutively done two sequence, therefore there is two files for each subject, except for the participants :

• ee2d3 who realised only Sequence1 due to cybersickness.
• 2f672e and 317ed5 which have both sequence 1 and sequence 2 in two part and 3754b5 which have sequence 2 in two part, due to hardware disconnection.

Sequence organisation is presented below.

Filenames follow this specific pattern : "(respondent_id)_(sequence_number).csv", e.g. : 02e52_sequence1.csv. Participant id list can be found below.

Raw data files respect this specific pattern of 3 parts:

• #INFO: general information about the recorded data
• #METADATA: explanatory information for each of the data-columns
• #DATA: sensor data

Feature set (from different source signal) can be identified by the "#Device" row:

• "B-Alert EEG" for EEG data (EEG)
• "B-Alert Decontaminated EEG" for decontamined EEG data (EEG_dec)
• "B-Alert BrainState" for brain metrics (EEG_brainstate)
• "Eyetracker HTC VIVE Pro Eye" for raw Eye-Tracking data (Eye-Tracking)
• "R Analysis GazeAnalysis I-VT filter" for Eye-Tracking analysis (GazeAnalysis)
• "Shimmer shim exg 02 5F2F ECG" for ECG data (ECG)
• "Shimmer GSR 4B59" for GSR data (GSR)

A total of 15 (10 males, 5 females) distinct participants were invited to perform industrial assembly task, age range is from 20 to 50. All the participants have completed higher education studies. Participants have been anonymized with an unique id number.

The self-assessed self-efficacy of the current task rated before and after the assemblage on a discrete 10-point scale (range of 1 to 10)

Self-efficacy is reported in the "External events API" columns:

• MarkerName="SOC_DISPLAY" corresponds to the time when the question is displayed
• MarkerName="SOC_ANSWER" corresponds to the time when the participant answer. The value of his perceived self-efficacy is given in the "MarkerDescription" column.

Each participant have subsequently done two sequence.

• Sequence 1 is composed of 6 assembly tasks :

  • P4-010, P4-011, P4-020, P4-030, P2-010, P2-020, P2-021, P2-022.

• Sequence 2 is composed of 9 assembly tasks :

  • P1-010, P1-020, P1-030, P3-010, P3-020, P3-030.

Situation in bold are designed as more complex than other as they synthetise assembly step. (example here)

Instruction sheet example can be found here.

File processing used is described in the paper and can be accessed here.

To cite this work, please use:

@INPROCEEDINGS{9757584,
  author={Bounhar, Thibaud and Yamak, Zaher and Havard, Vincent and Baudry, David},
  booktitle={2022 IEEE Conference on Virtual Reality and 3D User Interfaces Abstracts and Workshops (VRW)}, 
  title={A Dataset and Methodology for Self-Efficacy Feeling Prediction During Industry 4.0 VR Activity}, 
  year={2022},
  volume={},
  number={},
  pages={176-182},
  doi={10.1109/VRW55335.2022.00045}}

Publication available here https://ieeexplore.ieee.org/document/9757584