dgural/PhysicalAI-Autonomous-Vehicles-Sample

Dataset Card for nvidia-physical-ai-sample

This dataset is a small curated sample (100 items) extracted from the full
NVIDIA PhysicalAI Autonomous Vehicles dataset.
It is intended for quick experimentation, tutorials, and FiftyOne integration demos without requiring the multi-terabyte original dataset.

---

Installation

If you haven't already, install FiftyOne:

pip install -U fiftyone

---

Usage

import fiftyone as fo
from fiftyone.utils.huggingface import load_from_hub

# Load the sample dataset
dataset = load_from_hub("dgural/PhysicalAI-Autonomous-Vehicles-Sample")

# Launch the App
session = fo.launch_app(dataset)

---

Dataset Details

Dataset Description

This dataset provides a representative slice of the NVIDIA PhysicalAI Autonomous Vehicles dataset, including:

• Camera
• A structure identical to the full dataset, suitable for:
  • Pipeline prototyping
  • Instructional demos
  • AV data exploration with FiftyOne
  • Quick testing of loaders/adapters/exporters

The full dataset is available at:
https://huggingface.co/datasets/nvidia/PhysicalAI-Autonomous-Vehicles

Curated by

Voxel51 (sample extraction), derived from NVIDIA’s original dataset.

Language(s)

• en (metadata)

License

Inherits licensing from the original NVIDIA dataset.
See the main dataset page for license details.

---

Dataset Sources

• Primary Dataset: NVIDIA PhysicalAI Autonomous Vehicles
• Sample Extraction: Voxel51 using FiftyOne + Physical AI Workbench pipelines
• Repository: https://huggingface.co/datasets/nvidia/PhysicalAI-Autonomous-Vehicles
• Demo Code: https://github.com/voxel51/fiftyone

---

Uses

Direct Use

Appropriate uses of this dataset include:

• Testing dataset import/export mechanisms
• Unit tests for dataset auditing logic
• Teaching users how to navigate AV sensor datasets
• Lightweight experimentation

Out-of-Scope Use

This sample is not suitable for:

• Training ML models
• Benchmarking performance
• Statistical analysis
• Scenario diversity evaluation
• Research intended to generalize across AV driving conditions

---

Dataset Structure

This sample preserves the same organizational layout as the full PhysicalAI dataset:

• Per-sample grouped data

Each sample corresponds to a discrete AV sensor datapoint.

---

Dataset Creation

Curation Rationale

The full PhysicalAI dataset is extremely large.
This sample provides a lightweight, highly portable subset that can be used for:

• Rapid experimentation
• Prototyping ingestion pipelines
• Teaching and demos
• Running on laptops or small instances

Source Data

The underlying data originates from NVIDIA’s PhysicalAI dataset.
The sample was created by subselecting a limited number of frames and repacking them while preserving field structure.

Source data produced by

NVIDIA Autonomous Vehicles & PhysicalAI teams.

---

Bias, Risks, and Limitations

Because this is a non-representative sample, it:

• Does not capture full scenario diversity
• Should not be used for model training
• Cannot support robust statistical evaluation
• May omit critical driving edge cases

It is designed solely for small-scale experimentation.

---

Citation

If you use this dataset or sample, cite the original:

NVIDIA PhysicalAI Autonomous Vehicles Dataset
https://huggingface.co/datasets/nvidia/PhysicalAI-Autonomous-Vehicles

---

Contact

For questions related to this sample or the Physical AI Workbench:
https://voxel51.com