OpenPAV/OpenPAV-trajectory

---
pretty_name: OpenPAV-Trajectory
task_categories:
- tabular-regression
- time-series-forecasting
- other
tags:
- autonomous-driving
- transportation
- trajectory
- vehicle-dynamics
- csv
- tabular
- trajectory-modeling
- car-following-modeling
size_categories:
- 1M<n<10M
configs:
- config_name: Argoverse
data_files:
- split: train
path: data/Argoverse/*.csv
- config_name: CATS
data_files:
- split: train
path: data/CATS/*.csv
- config_name: MicroSimACC
data_files:
- split: train
path: data/MicroSimACC/*.csv
- config_name: Ohio
data_files:
- split: train
path: data/Ohio/*.csv
- config_name: OpenACC
data_files:
- split: train
path: data/OpenACC/*.csv
- config_name: Vanderbilt
data_files:
- split: train
path: data/Vanderbilt/*.csv
viewer: true
---

# OpenPAV-Trajectory

## Dataset Description

OpenPAV-Trajectory is a curated collection of longitudinal vehicle-following trajectories for production automated vehicles (PAVs). It is part of the OpenPAV platform, which supports data collection, behavior modeling, and performance evaluation for production automated driving systems.

This release standardizes public trajectory datasets into one common tabular schema centered on two vehicles: a lead vehicle (LV) and a following automated vehicle (FAV).

The dataset is intended for car-following analysis, trajectory modeling, calibration of behavioral models, benchmarking, and simulation-oriented automated driving research.

OpenPAV project page: <https://openpav.github.io/OpenPAV>

## Key Facts

- 12 CSV files from 6 data providers
- approximately 3,537,455 rows in total
- approximately 675 MB of raw CSV data
- unified schema across all files
- stored as provider-specific subsets for straightforward loading on the Hugging Face Hub

## Repository Structure

```text
OpenPAV-Trajectory/
├── README.md
├── Dataset.png
└── data/
├── Argoverse/
├── CATS/
├── MicroSimACC/
├── Ohio/
├── OpenACC/
└── Vanderbilt/
```

Each provider directory is exposed as a separate Hugging Face dataset configuration:

- `Argoverse`
- `CATS`
- `MicroSimACC`
- `Ohio`
- `OpenACC`
- `Vanderbilt`

## Load the Dataset

```python
from datasets import load_dataset

dataset = load_dataset("YOUR_USERNAME/OpenPAV-Trajectory", "OpenACC")
print(dataset["train"])
```

To load a specific CSV manually:

```python
from datasets import load_dataset

dataset = load_dataset(
"csv",
data_files="data/OpenACC/step3_ZalaZone.csv",
)
```

## Data Schema

All CSV files follow the same schema.

| Column | Description | Unit |
| --- | --- | --- |
| `Trajectory_ID` | Unique identifier of a longitudinal trajectory | N/A |
| `Time_Index` | Timestamp within a trajectory | s |
| `ID_LV` | Lead vehicle ID | N/A |
| `Type_LV` | Lead vehicle type: automated vehicle = 1, human-driven vehicle = 0 | N/A |
| `Pos_LV` | Lead vehicle position in Frenet coordinates | m |
| `Speed_LV` | Lead vehicle speed | m/s |
| `Acc_LV` | Lead vehicle acceleration | m/s^2 |
| `ID_FAV` | Following automated vehicle ID | N/A |
| `Pos_FAV` | Following automated vehicle position in Frenet coordinates | m |
| `Speed_FAV` | Following automated vehicle speed | m/s |
| `Acc_FAV` | Following automated vehicle acceleration | m/s^2 |
| `Spatial_Gap` | Bumper-to-bumper spacing between LV and FAV | m |
| `Spatial_Headway` | Center-to-center distance between LV and FAV | m |
| `Speed_Diff` | Relative speed defined as `Speed_LV - Speed_FAV` | m/s |

## Source Datasets

This integrated release currently standardizes public data from the following sources:

- Argoverse 2 Motion Forecasting Dataset
- CATS Open Datasets
- Central Ohio ACC Datasets
- MicroSimACC Dataset
- OpenACC Database
- Vanderbilt ACC Dataset

These sources cover multiple cities, road environments, and automated driving scenarios. The current repository contains transformed and harmonized trajectory tables derived from those public resources.

<img src="./dataset.jpg" alt="OpenPAV-Trajectory overview" width="700">

## Contributing Data

We welcome contributions of PAV trajectory datasets.

Please follow these steps:

1. Fork this dataset repository.
2. Upload your dataset following the structure described below.
3. Submit a Pull Request.
4. The maintainers will review and merge the dataset.

## Contributors

- [Hang Zhou](https://catslab.engr.wisc.edu/staff/zhou-hang/), Keke Long , Chengyuan Ma.