Data Usage Agreement (DUA) — Key Terms (public summary) Dataset: EmoRoad Multimodal Dataset (controlled access; includes identifiable or potentially re-identifiable human data such as facial videos and related behavioural/physiological signals). Data controller/provider: Prof. Stephen Jia Wang, The Hong Kong Polytechnic University (PolyU)DUA version/date: v1.2 (15/04/2026) By requesting access to, downloading, or using any part of this dataset, the Data User agrees to the following key terms: 1) Permitted use (internal research/analysis only)- Use is permitted for internal research and analysis (including methodological development/evaluation), subject to compliance with applicable laws and institutional policies.- The dataset itself must NOT be monetised, sold, licensed, or provided as a product/service/hosted dataset offering.- Legitimate internal R&D in academic or commercial settings is allowed provided there is no redistribution, no re-identification/contact, and no outputs that expose identifiable raw data. 2) No re-identification & no participant interaction- No attempts to identify or re-identify participants (including by linking with other datasets or using biometric identification such as face/voice recognition).- No contacting or attempting to contact participants.- If a participant’s identity is learned accidentally: cease analyses using identified information, do not record/disclose the identity, and notify the provider. 3) Data security & access control- Store and process data on secure, access-controlled systems (e.g., institutional servers or encrypted storage).- Limit access to authorised personnel working on the authorised project; ensure all such personnel are informed of and comply with these terms.- Do not store the dataset on public/insecure services or via publicly shared links. 4) No redistribution / onward sharing- No sharing/redistribution of the dataset (whole or part) to anyone not covered by an approved access agreement.- External collaborators must submit their own access request and agree to the DUA separately. 5) Outputs, publication, and citation- Only derived, non-identifiable outputs may be shared (e.g., aggregate statistics/plots/metrics/model outputs) and must not enable re-identification or disclose identifiable raw data (e.g., identifiable frames or raw signals that compromise privacy).- Publications must cite the dataset DOI shown on this record and any associated dataset paper/preprint where applicable.- Do not imply endorsement by the data providers or participants. 6) Incident reporting- Promptly notify the provider of any suspected/actual data breach, loss, unauthorised access, or accidental re-identification risk, and cooperate to mitigate risk. 7) Duration, termination, and deletion- Access may be revoked if terms are breached or if continued access poses unacceptable ethical/legal/security risks.- Upon request or termination, delete/destroy all copies (including backups) and confirm deletion if requested. Access procedure:- Use the “Request access” function on this Zenodo record. Provide name, affiliation, and a verifiable contact email/ORCID, and confirm agreement to the DUA terms above. Full agreement:- A complete DUA document is provided as `00_Data_Usage_Agreement_v2.pdf` (for reference).

Overview This dataset is a multimodal dataset collected under multiple driving tasks and conditions. It includes data from 50 participants and 8 driving tasks defined by a 2 × 2 × 2 factorial design: weather (2 levels), traffic volume (2 levels), and road scenario (2 levels). The recorded modalities include: facial video, in-car first-person screen recordings, car dynamics (e.g., speed, acceleration, etc.), EmoSense (steering-wheel touch sensor), EEG, eye tracking, and emotion annotations from both iMotions and participant self-reports. The dataset supports research on emotion and behavior modeling and analysis, and driving context's effects on emotion and behavior, with potential applications in driver state monitoring and adaptive vehicle system. Contents & file organization The upload contains two top-level folders: `Clip/` and `RawData/`. Clip/ `Clip/` provides task-aligned (clipped) data for all modalities. All signals/videos are clipped according to the valid driving-task start and end timestamps. The corresponding timestamps and helper code are provided in `ref_timestamp.py` in our GitHub repository (see “Repository URL” in this record). For EEG data in `Clip/`, we provide a preprocessed version with artifacts removed using ICA and amplitude-thresholding. The preprocessing scripts are also available in the GitHub repository. RawData/ `RawData/` provides representative raw recordings (including raw EEG), enabling users to apply their preferred preprocessing pipelines. Folder structure The overall folder structure is illustrated below. Data are organized by modality under `Clip/` or `RawData/`, using the prefix `Clip_` or `RawData_` followed by the modality name. Within each modality folder, data are organized as:- `P1/`–`P50/` (participants) - 8 driving tasks per participant, each containing the corresponding recordings for that task. Clip/├─ Clip_Car_Dynamics/├─ Clip_EEG_noArtifact/│ ├─ P1/│ │ ├─ 0925_p1_1.csv│ │ ├─ 0925_p1_2.csv│ │ ├─ ...│ │ └─ 0925_p1_8.csv│ ├─ P2/│ ├─ ...│ └─ P50/├─ Clip_EmoSense/├─ ...└─ Clip_ScreenRecord/ RawData/├─ RawData_Car_Dynamics/├─ RawData_EEG/├─ RawData_EmoSense/└─ RawData_iMotions/

Access notice: This dataset is available under restricted access. By requesting access, downloading, or using the data, you agree to comply with the Data Usage Agreement (DUA). Please review the “Data Usage Agreement (DUA) — Key Terms (public summary)” below. The full DUA text will be available to approved users, and in case of any discrepancy, the full DUA prevails.