README.md

---
license: cc-by-4.0
task_categories:
  - tabular-classification
  - tabular-regression
  - time-series-forecasting
language:
  - en
tags:
  - synthetic
  - autonomous-vehicles
  - adas
  - automotive
  - sensor-fusion
  - can-bus
  - lidar
  - radar
  - perception
  - safety-critical
  - anomaly-detection
  - edge-cases
  - robotics
pretty_name: Hyperion Drive AV Safety Intelligence Pack
size_categories:
  - 10K<n<100K
configs:
  - config_name: default
    data_files:
      - split: train
        path: hyperion_drive_sample.parquet
---

# Hyperion Drive AV Safety Intelligence Pack (Sample)

**A synthetic autonomous-vehicle telemetry dataset for ADAS validation, sensor-fusion model training, and edge-case safety research.** Each row is a complete driving event chain — from environmental trigger (black ice, glare, CAN bus glitch) through a cascading perception or control failure — with high-frequency CAN bus signals at each step, failure-class labels, causal-chain annotations, and a safety-critical outcome.

Built by [SolsticeAI](https://www.solsticestudio.ai/datasets) as a free sample of a larger commercial pack. 100% synthetic. No real vehicle logs, no real VINs (all VINs prefixed `SYN-VIN-` to disambiguate from real WMIs), no identifiable user data. Safe for model training, regulator evaluation, and public benchmark work.

## What is included

| File | Rows | Format | Purpose |
|---|---:|---|---|
| `hyperion_drive_sample.parquet` | 10,000 | Parquet | Columnar, typed, best for analytics |
| `hyperion_drive_sample.jsonl` | 10,000 | JSON Lines | Streaming / LLM training friendly |

**This sample:** 10,000 driving event chains, balanced across 4 safety-criticality tiers and 3 failure classes.  
**Safety-criticality tiers:** `low`, `medium`, `high`, `critical` (~2,500 each)  
**Failure classes:** `Sensor_Fusion_Conflict`, `Mechanical_Latency_Failure`, `Perception_Hallucination` (~3,300 each)  
**Vehicle models:** `EV_Sedan_Alpha`, `Autonomous_Shuttle_V2`, `Heavy_Duty_Truck`  
**Drive modes:** `manual_override`, `autopilot_assisted`, `autonomous`  
**Environmental chaos profiles:** `Black_Ice`, `Blinding_Glare`, `Intermittent_CAN_Bus_Error`, `Sensor_Ghosting`

## Record structure

Each record is one driving event chain with 7 top-level fields:

| Field | Type | Contents |
|---|---|---|
| `schema_version` | string | Pack schema version (`1.0.0-hyperion-drive-sample`) |
| `event` | struct | `id`, `trace_id`, `timestamp`, `safety_criticality`, `outcome`, `confidence` |
| `vehicle_context` | struct | `model`, `vin` (synthetic), `drive_mode`, `mileage_km` |
| `perception_logic` | struct | `failure_class`, `causal_chain[]`, `reaction_time_ms`, `tracked_signals[]` |
| `can_bus_telemetry` | list<struct> | Ordered CAN/Ethernet bus messages: `timestamp`, `vin`, `event_name`, `sensor_source`, `bus_id`, `signals` (wheel_speed_kph, brake_pedal_pos, steering_angle_deg, sensor_health_pct, camera_confidence_pct, lidar_noise_pct, radar_distance_m, brake_pressure_bar, traction_slip_pct) |
| `detection_logic` | struct | `signature`, `anomaly_score`, `baseline_deviation` |
| `simulation` | struct | `synthetic`, `engine`, `chaos_profile` |

See [SCHEMA.md](./SCHEMA.md) for the full nested field breakdown.

## Why this dataset is useful

Most public autonomous-driving datasets are either massive raw sensor blobs (nuScenes, Waymo Open, KITTI) that require hundreds of GB and deep infrastructure, or very narrow single-signal logs. This pack is shaped around a different need: decision-chain telemetry — the signal-level story of what happened, labeled with failure class and outcome — at a scale you can pull into a notebook in seconds.

- Complete event chains rather than isolated signal windows
- Balanced safety-criticality tiers across low → critical
- Causal-chain labels connecting trigger → cascade → outcome
- Per-step CAN / Ethernet signal snapshots at the moments the failure unfolds
- Cross-vehicle coverage (passenger EV, autonomous shuttle, heavy-duty truck)
- Scenario labels across ice, glare, bus glitch, sensor ghosting
- Compact (parquet + JSONL) so it fits inside CI and rapid prototyping flows

## Typical use cases

- ADAS and AV validation pipelines
- Sensor-fusion conflict detection
- Anomaly detection on CAN bus signals
- Failure-class classification models
- Reaction-time and control-delay modeling
- LLM fine-tuning on automotive incident narratives
- Scenario-mining prototypes for fleet debugging
- Dashboard / BI template development for AV observability

## Quick start

```python
import pandas as pd
import pyarrow.parquet as pq

df = pq.read_table("hyperion_drive_sample.parquet").to_pandas()

# Safety-criticality distribution (balanced)
print(df["event"].apply(lambda e: e["safety_criticality"]).value_counts())

# Reaction-time by failure class
df["failure"] = df["perception_logic"].apply(lambda p: p["failure_class"])
df["reaction_ms"] = df["perception_logic"].apply(lambda p: p["reaction_time_ms"])
print(df.groupby("failure")["reaction_ms"].mean().round(1))

# Drive-mode vs outcome cross-tab
df["drive_mode"] = df["vehicle_context"].apply(lambda v: v["drive_mode"])
df["outcome"] = df["event"].apply(lambda e: e["outcome"])
print(pd.crosstab(df["drive_mode"], df["outcome"]))
```

Streaming form:

```python
import json

with open("hyperion_drive_sample.jsonl") as f:
    for line in f:
        chain = json.loads(line)
        # one driving event chain per line
```

## Responsible use

This dataset is intended for **research, model training, and safety-evaluation** use cases around autonomous vehicles and ADAS systems. It contains synthesized CAN bus signals, synthetic VINs, and scenario labels — it does **not** contain real vehicle logs, real VINs, real user data, or identifiable information of any kind. Models trained on this data must be independently validated against real fleet telemetry under appropriate safety-assurance review before any deployment decisions.

## License

Released under **CC BY 4.0**. Use freely for research, ADAS prototyping, education, and commercial development with attribution.

## Get the full pack

This Hugging Face repo is a **10K-event-chain sample**. The production pack scales to 1M+ chains with finer-grained signal sampling (ms-level CAN traces), expanded scenario coverage (rain, snow, construction zones, V2X signal loss, cyberattacks on bus), additional vehicle classes (motorcycle, delivery robot, transit bus), per-sensor noise models, ROS-bag and CAN DBC-aligned delivery, and buyer-specific variants.

**Self-serve (Stripe checkout):**
- [**Sample Scale tier — $5,000**](https://buy.stripe.com/7sY5kD2j85QTfSb5lfeEo03) — ~25K records, one subject, 72-hour delivery.

**Full pack + enterprise scope:**
- [www.solsticestudio.ai/datasets](https://www.solsticestudio.ai/datasets) — per-SKU pricing across Starter / Professional / Enterprise tiers, plus commercial licensing, custom generation, and buyer-specific variants.

**Procurement catalog:**
- [SolsticeAI Data Storefront](https://solsticeai.mydatastorefront.com) — available via Datarade / Monda.

## Citation

```bibtex
@dataset{solstice_hyperion_drive_pack_2026,
  title        = {Hyperion Drive AV Safety Intelligence Pack (Sample)},
  author       = {SolsticeAI},
  year         = {2026},
  publisher    = {Hugging Face},
  url          = {https://huggingface.co/datasets/solsticestudioai/hyperion-drive-pack}
}
```