# OctoSense: Synchronized Multi-Sensor Driving Dataset with Event Cameras

## Facts

- Provider: Anthony Bisulco et al. (University of Pennsylvania)
- License: MIT (https://huggingface.co/datasets/anthonytec2/OctoSense)
- Size: 3,058 files
- Format: PNG, MP4, HDF5 (in sampled files)
- Stored at: Hugging Face, https://huggingface.co/datasets/anthonytec2/OctoSense
- Access: public, readable without an account
- Category: Robotics
- Source: https://arxiv.org/abs/2606.27317
- Page: https://scigantic.com/archive/57f45ad6-eaef-5961-91c2-ac998f9dbc68
- Keywords: Robotics, image segmentation, LiDAR, PNG, MP4, HDF5, Hugging Face

## About

OctoSense is a time-synchronised, calibrated multi-sensor dataset in which a car, a boat and a Unitree Go2-W quadruped share the same sensor rig (arXiv:2606.27317). Most of it is driving: 371 sequences, 59 hours and 2,474 km of urban, suburban and rural driving on Long Island and in Philadelphia at sunrise, day, sunset and night. Every sequence carries stereo RGB, stereo event cameras, infrared, Ouster LiDAR and IMU; driving sequences add GPS, CAN, ground truth for depth, ego-motion optical flow and semantic segmentation, and a fused GPS plus LiDAR-inertial odometry trajectory. Nine boat sequences and two quadruped sequences (with per-joint robot state) have no perception ground truth. Sequences are keyed by recording datetime under car/, boat/ and unitree/, with per-sequence data in HDF5 (data.h5), calibrations/, and metadata.parquet listing platform and attributes. Measured 2026-11-14 at about 8.5 TB. Read with h5py and pandas; the authors provide a Colab getting-started notebook. Useful for event-based vision, depth and flow estimation, SLAM and night driving. Released under the MIT licence; cite the OctoSense paper.

## Load the data

```python
from huggingface_hub import hf_hub_download
path = hf_hub_download(repo_id="anthonytec2/OctoSense", filename="boat/sess1/rosbag2_2025_11_20-08_48_14/data.h5", repo_type="dataset")
print(path)
```

## Data structure

File types in sampled files: PNG, MP4, HDF5

Example file: `boat/sess1/rosbag2_2025_11_20-08_48_14/data.h5`

## Open it

- In a hosted notebook with the data mounted: https://scigantic.com/archive/57f45ad6-eaef-5961-91c2-ac998f9dbc68
- Full record as JSON (schema card included): https://api.scigantic.com/api/archives/57f45ad6-eaef-5961-91c2-ac998f9dbc68
- Over MCP: `get_schema_card` with id `57f45ad6-eaef-5961-91c2-ac998f9dbc68` at https://api.scigantic.com/mcp

## Related topics

- [Image segmentation datasets](https://scigantic.com/topics/image-segmentation)
- [PNG datasets](https://scigantic.com/topics/png)
- [HDF5 datasets](https://scigantic.com/topics/hdf5)
- [LiDAR datasets](https://scigantic.com/topics/lidar)
- [Robotics datasets](https://scigantic.com/topics/robotics)
- [MP4 datasets](https://scigantic.com/topics/mp4)
