Add dataset card README

README.md

@@ -0,0 +1,283 @@
+---
+license: cc-by-nc-4.0
+task_categories:
+  - time-series-forecasting
+  - video-classification
+tags:
+  - autonomous-driving
+  - ADAS
+  - takeover
+  - driver-behavior
+  - openpilot
+  - time-series
+  - multimodal
+  - CAN-bus
+  - vehicle-dynamics
+size_categories:
+  - 10K<n<100K
+language:
+  - en
+pretty_name: "HADAS-TakeOver: Human-ADAS Driving Takeover Dataset"
+---
+
+# HADAS-TakeOver: Human-ADAS Driving Takeover Dataset
+
+## Dataset Summary
+
+**HADAS-TakeOver** is a large-scale, multimodal dataset of **15,705 real-world ADAS takeover events** captured from **327 drivers** across **163 vehicle models** from **23 manufacturers**. Each takeover event is a 20-second clip centered on the moment a driver disengages an Advanced Driver Assistance System (ADAS), providing synchronized video, vehicle dynamics, controller state, and sensor data.
+
+| Statistic | Value |
+|---|---|
+| Total takeover clips | 15,705 |
+| Unique drivers | 327 |
+| Unique driving routes | 2,312 |
+| Vehicle models | 163 |
+| Manufacturers | 23 |
+| Clip duration | 20 seconds (±10 s around takeover) |
+| Video resolution | Front-facing camera, 20 fps |
+| CAN/sensor signals | 10–100 Hz |
+| Total size | ~33 GB |
+
+## Motivation
+
+Understanding how and why drivers take over from automated driving systems is critical for improving ADAS safety, designing better human-machine interfaces, and developing predictive takeover models. Despite growing research interest, large-scale naturalistic takeover datasets remain scarce. HADAS-TakeOver addresses this gap by providing:
+
+- **Scale**: Over 15K events across hundreds of drivers and vehicles
+- **Diversity**: 163 vehicle models spanning sedans, SUVs, trucks, and EVs from 23 manufacturers
+- **Richness**: Synchronized video + 9 time-series signal files per event
+- **Naturalistic**: Real-world driving (not simulator), capturing genuine driver behavior
+
+### Use Cases
+
+- Takeover prediction and early warning systems
+- Driver behavior modeling during ADAS transitions
+- Analysis of ADAS disengagement patterns across vehicle types
+- Human factors research in automated driving
+- Multimodal time-series classification and forecasting
+
+## Dataset Structure
+
+```
+HADAS-TakeOver/
+  <CAR_MODEL>/                  # e.g., TOYOTA_PRIUS, TESLA_AP3_MODEL_3
+    <driver_XXX>/               # anonymized driver ID
+      <route_XXX>/              # anonymized route ID
+        <clip_id>/              # integer (0-indexed per route)
+          meta.json             # clip metadata
+          takeover.mp4          # 20-second front-camera video
+          carState.csv          # vehicle state signals
+          controlsState.csv     # ADAS controller state
+          carControl.csv        # control commands
+          carOutput.csv         # actuator outputs
+          drivingModelData.csv  # driving model predictions
+          radarState.csv        # radar / lead vehicle data
+          accelerometer.csv     # IMU accelerometer data
+          longitudinalPlan.csv  # longitudinal planner outputs
+```
+
+Each clip contains **10 files**: 1 video, 1 metadata JSON, and 8 CSV time-series files.
+
+## Takeover Event Definition
+
+A **takeover event** is defined as an ADAS ON → OFF transition where:
+
+- **ADAS engaged** = `controlsState.enabled` OR `carState.cruiseState.enabled`
+- **Minimum ON duration**: 2 seconds before disengagement
+- **Minimum OFF duration**: 2 seconds after disengagement
+- **Gap merging**: Transient gaps < 0.5 s are merged (to filter sensor noise)
+- **Clip window**: ±10 seconds centered on the ON→OFF transition (20 s total)
+
+The first ~10 seconds of each clip show ADAS-engaged driving; the remaining ~10 seconds show the driver resuming manual control.
+
+## Data Fields
+
+### meta.json
+
+| Field | Type | Description |
+|---|---|---|
+| `car_model` | string | Vehicle model identifier (e.g., `TOYOTA_PRIUS`) |
+| `dongle_id` | string | Anonymized driver ID (`driver_XXX`) |
+| `route_id` | string | Anonymized route ID (`route_XXX`) |
+| `log_kind` | string | Log source: `qlog` (10 Hz) or `rlog` (100 Hz) |
+| `log_hz` | int | Sampling rate of CAN signals (10 or 100) |
+| `vid_kind` | string | Video source: `qcamera` or `fcamera` |
+| `camera_fps` | int | Video frame rate (20 fps) |
+| `clip_id` | int | Clip index within the route (0-indexed) |
+| `event_mono` | int | Monotonic timestamp of the takeover event (nanoseconds) |
+| `video_time_s` | float | Time of takeover within the full route video (seconds) |
+| `clip_start_s` | float | Start time of the 20-second clip within the route (seconds) |
+| `clip_dur_s` | float | Clip duration (seconds, typically 20.0) |
+| `seg_nums_used` | list[int] | Openpilot segment numbers covering this route |
+
+### carState.csv — Vehicle State
+
+Driver inputs and ego vehicle dynamics.
+
+| Column | Description |
+|---|---|
+| `vEgo` | Ego vehicle speed (m/s) |
+| `aEgo` | Ego vehicle acceleration (m/s²) |
+| `steeringAngleDeg` | Steering wheel angle (degrees) |
+| `steeringTorque` | Steering torque applied by driver |
+| `steeringPressed` | Whether driver is actively steering (boolean) |
+| `gasPressed` | Whether gas pedal is pressed (boolean) |
+| `brakePressed` | Whether brake pedal is pressed (boolean) |
+| `cruiseState.enabled` | Whether cruise control / ADAS is engaged (boolean) |
+
+### controlsState.csv — ADAS Controller State
+
+| Column | Description |
+|---|---|
+| `enabled` | Whether openpilot ADAS is enabled (boolean) |
+| `active` | Whether ADAS is actively controlling the vehicle |
+| `curvature` | Current path curvature (1/m) |
+| `desiredCurvature` | Target path curvature from planner |
+| `vCruise` | Set cruise speed (m/s) |
+| `longControlState` | Longitudinal control state (enum) |
+| `alertText1` | Primary alert text displayed to driver |
+| `alertText2` | Secondary alert text |
+
+### carControl.csv — Control Commands
+
+| Column | Description |
+|---|---|
+| `latActive` | Whether lateral control is active |
+| `longActive` | Whether longitudinal control is active |
+| `actuators.accel` | Commanded acceleration (m/s²) |
+| `actuators.torque` | Commanded steering torque |
+| `actuators.curvature` | Commanded path curvature |
+
+### carOutput.csv — Actuator Outputs
+
+| Column | Description |
+|---|---|
+| `actuatorsOutput.accel` | Actual acceleration output |
+| `actuatorsOutput.brake` | Brake actuator output |
+| `actuatorsOutput.gas` | Gas actuator output |
+| `actuatorsOutput.steer` | Steering actuator output |
+| `actuatorsOutput.steerOutputCan` | Raw CAN steering output |
+| `actuatorsOutput.steeringAngleDeg` | Output steering angle (degrees) |
+
+### drivingModelData.csv — Driving Model Predictions
+
+| Column | Description |
+|---|---|
+| `action.desiredCurvature` | Model-predicted desired curvature |
+| `action.desiredAcceleration` | Model-predicted desired acceleration |
+| `laneLineMeta.leftProb` | Probability of left lane line detection |
+| `laneLineMeta.rightProb` | Probability of right lane line detection |
+
+### radarState.csv — Lead Vehicle Detection
+
+| Column | Description |
+|---|---|
+| `leadOne.dRel` | Distance to primary lead vehicle (m) |
+| `leadOne.vRel` | Relative velocity of lead vehicle (m/s) |
+| `leadOne.vLead` | Absolute velocity of lead vehicle (m/s) |
+| `leadOne.aLeadK` | Estimated acceleration of lead vehicle (m/s²) |
+| `leadTwo.*` | Same fields for secondary lead vehicle |
+
+### accelerometer.csv — IMU Data
+
+| Column | Description |
+|---|---|
+| `acceleration.v` | 3-axis acceleration vector (m/s²) |
+| `timestamp` | Sensor timestamp |
+
+### longitudinalPlan.csv — Planner Outputs
+
+| Column | Description |
+|---|---|
+| `aTarget` | Target acceleration from planner (m/s²) |
+| `hasLead` | Whether a lead vehicle is detected (boolean) |
+| `fcw` | Forward collision warning active (boolean) |
+| `speeds[]` | Planned speed profile |
+| `accels[]` | Planned acceleration profile |
+
+## Top Vehicle Models
+
+| Vehicle Model | Clips | | Vehicle Model | Clips |
+|---|---|---|---|---|
+| RIVIAN R1 GEN1 | 2,127 | | CHEVROLET BOLT EUV 2022 | 244 |
+| ACURA MDX 3G MMR | 1,863 | | TOYOTA RAV4 TSS2 2023 | 228 |
+| FORD F-150 MK14 | 1,226 | | RAM HD 5TH GEN | 221 |
+| CHEVROLET SILVERADO | 639 | | VOLKSWAGEN JETTA MK7 | 215 |
+| TOYOTA PRIUS | 482 | | KIA EV6 | 209 |
+| HONDA CIVIC | 470 | | VOLKSWAGEN GOLF MK7 | 192 |
+| TESLA AP3 MODEL 3 | 432 | | KIA NIRO EV | 185 |
+| FORD MAVERICK MK1 | 300 | | HYUNDAI IONIQ 6 | 177 |
+| HYUNDAI IONIQ 5 | 266 | | VOLKSWAGEN ATLAS MK1 | 153 |
+
+## Usage
+
+### Loading a Single Clip
+
+```python
+import json
+import pandas as pd
+from huggingface_hub import hf_hub_download
+
+repo_id = "HenryYHW/ADAS-TO"
+clip_path = "TOYOTA_PRIUS/driver_001/route_001/0"
+
+# Download metadata
+meta_path = hf_hub_download(repo_id, f"{clip_path}/meta.json", repo_type="dataset")
+with open(meta_path) as f:
+    meta = json.load(f)
+print(meta)
+
+# Load vehicle state signals
+car_state_path = hf_hub_download(repo_id, f"{clip_path}/carState.csv", repo_type="dataset")
+car_state = pd.read_csv(car_state_path)
+print(car_state[["vEgo", "aEgo", "steeringAngleDeg", "brakePressed"]].describe())
+
+# Load ADAS controller state
+controls_path = hf_hub_download(repo_id, f"{clip_path}/controlsState.csv", repo_type="dataset")
+controls = pd.read_csv(controls_path)
+```
+
+### Iterating Over All Clips
+
+```python
+from huggingface_hub import HfApi
+
+api = HfApi()
+files = api.list_repo_files("HenryYHW/ADAS-TO", repo_type="dataset")
+meta_files = [f for f in files if f.endswith("meta.json")]
+print(f"Total clips: {len(meta_files)}")
+```
+
+### Downloading the Full Dataset
+
+```bash
+# Using huggingface-cli
+huggingface-cli download HenryYHW/ADAS-TO --repo-type dataset --local-dir ./HADAS-TakeOver
+
+# Using git-lfs
+git lfs install
+git clone https://huggingface.co/datasets/HenryYHW/ADAS-TO
+```
+
+## Data Collection
+
+This dataset was built from driving logs collected by the [comma.ai](https://comma.ai/) community using [openpilot](https://github.com/commaai/openpilot), an open-source ADAS platform. Logs were processed to detect ADAS disengagement events, extract synchronized video and CAN-bus signals, and package them into standardized clips.
+
+**Privacy**: All driver and route identifiers have been anonymized. No personally identifiable information (PII) is included. Video data shows the forward road view only.
+
+## Citation
+
+If you use this dataset in your research, please cite:
+
+```bibtex
+@dataset{hadas_takeover_2025,
+  title={HADAS-TakeOver: A Large-Scale Naturalistic Dataset of Human-ADAS Driving Takeover Events},
+  author={Zhou, Haowei},
+  year={2025},
+  publisher={Hugging Face},
+  url={https://huggingface.co/datasets/HenryYHW/ADAS-TO}
+}
+```
+
+## License
+
+This dataset is released under [CC BY-NC 4.0](https://creativecommons.org/licenses/by-nc/4.0/). It is intended for academic and non-commercial research purposes.