---
license: mit
dataset_info:
  features:
  - name: preview.png
    dtype: image
  - name: raw_log.bz2
    dtype: binary
  - name: video.hevc
    dtype: binary
  - name: processed_log.npz
    struct:
    - name: global_pose_frame_gps_times.npy
      list:
        list: float64
    - name: global_pose_frame_orientations.npy
      list:
        list: float64
    - name: global_pose_frame_positions.npy
      list:
        list: float64
    - name: global_pose_frame_times.npy
      list: float64
    - name: global_pose_frame_velocities.npy
      list:
        list: float64
    - name: processed_log_can_radar_t.npy
      list: float64
    - name: processed_log_can_radar_value.npy
      list:
        list: float64
    - name: processed_log_can_raw_can_address.npy
      list: int64
    - name: processed_log_can_raw_can_data.npy
      list: binary
    - name: processed_log_can_raw_can_src.npy
      list: int64
    - name: processed_log_can_raw_can_t.npy
      list: float64
    - name: processed_log_can_speed_t.npy
      list: float64
    - name: processed_log_can_speed_value.npy
      list:
        list: float64
    - name: processed_log_can_steering_angle_t.npy
      list: float64
    - name: processed_log_can_steering_angle_value.npy
      list: float64
    - name: processed_log_can_wheel_speed_t.npy
      list: float64
    - name: processed_log_can_wheel_speed_value.npy
      list:
        list: float64
    - name: processed_log_gnss_live_gnss_qcom_t.npy
      list: float64
    - name: processed_log_gnss_live_gnss_qcom_value.npy
      list:
        list: float64
    - name: processed_log_gnss_live_gnss_ublox_t.npy
      list: float64
    - name: processed_log_gnss_live_gnss_ublox_value.npy
      list:
        list: float64
    - name: processed_log_gnss_raw_gnss_qcom_t.npy
      list: float64
    - name: processed_log_gnss_raw_gnss_qcom_value.npy
      list:
        list: float64
    - name: processed_log_gnss_raw_gnss_ublox_t.npy
      list: float64
    - name: processed_log_gnss_raw_gnss_ublox_value.npy
      list:
        list: float64
    - name: processed_log_imu_accelerometer_t.npy
      list: float64
    - name: processed_log_imu_accelerometer_value.npy
      list:
        list: float64
    - name: processed_log_imu_gyro_bias_t.npy
      list: float64
    - name: processed_log_imu_gyro_bias_value.npy
      list:
        list: float64
    - name: processed_log_imu_gyro_t.npy
      list: float64
    - name: processed_log_imu_gyro_uncalibrated_t.npy
      list: float64
    - name: processed_log_imu_gyro_uncalibrated_value.npy
      list:
        list: float64
    - name: processed_log_imu_gyro_value.npy
      list:
        list: float64
  - name: __key__
    dtype: string
  - name: __url__
    dtype: string
  splits:
  - name: train
    num_bytes: 3023542189
    num_examples: 46
  download_size: 2305483724
  dataset_size: 3023542189
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
---

# comma2k19
[comma.ai](https://comma.ai) presents comma2k19, a dataset of over 33 hours of commute in California's 280 highway. This means 2019 segments, 1 minute long each, on a 20km section of highway driving between California's San Jose and San Francisco. comma2k19 is a fully reproducible and scalable dataset. The data was collected using comma [EONs](https://comma.ai/shop/products/eon-gold-dashcam-devkit/) that has sensors similar to those of any modern smartphone including a road-facing camera, phone GPS, thermometers and 9-axis IMU. Additionally, the EON captures raw GNSS measurements and all CAN data sent by the car with a comma [grey panda](https://comma.ai/shop/products/panda-obd-ii-dongle/). 

<img src="https://github.com/commaai/comma2k19/blob/master/assets/testmesh3d.png?raw=true"/>

Here we also introduced [Laika](https://github.com/commaai/laika), an open-source GNSS processing library. Laika produces 40% more accurate positions than the GNSS module used to collect the raw data. This dataset includes pose (position + orientation) estimates in a global reference frame of the recording camera. These poses were computed with a tightly coupled INS/GNSS/Vision optimizer that relies on data processed by Laika. comma2k19 is ideal for development and validation of tightly coupled GNSS algorithms and mapping algorithms that work with commodity sensors. 

<img src="https://github.com/commaai/comma2k19/blob/master/assets/merged.png?raw=true"/>

## Publication
For a detailed write-up about this dataset, please refer to our [paper](https://arxiv.org/abs/1812.05752v1). If you use comma2k19 or Laika in your research, please consider citing
```text
@misc{1812.05752,
Author = {Harald Schafer and Eder Santana and Andrew Haden and Riccardo Biasini},
Title = {A Commute in Data: The comma2k19 Dataset},
Year = {2018},
Eprint = {arXiv:1812.05752},
}
```