Note

This is a Hugging Face dataset. For large datasets, ensure huggingface_hub>=1.1.3 to avoid rate limits. Learn more in the Hugging Face integration docs.

Hugging Face

Dataset Card for ColoRadar Sample#

preview

This is a FiftyOne dataset with 7 samples. Each sample is one handheld sequence from ColoRadar, stored as a native multimodal MCAP episode.

The source is ColoRadar, the 3D millimetre-wave radar dataset from the Autonomous Robotics and Perception Group at the University of Colorado Boulder, converted from ROS 1 bags. A handheld rig carries two FMCW radars, a TI MMWCAS-RF-EVM cascaded imaging radar and a TI AWR1843BOOST single-chip radar, beside an Ouster OS1-64 LiDAR and a Lord Microstrain 3DM-GX5-25 inertial unit. The release holds 52 sequences recorded in seven places: hallways, a lab, a large motion-capture space, outdoor built environments, the narrow and the large passages of an underground mine, and a fast ride along paths and roads. This sample carries one sequence from each place, the shortest of its group.

Installation#

If you haven’t already, install FiftyOne:

pip install -U fiftyone

Usage#

import fiftyone as fo
import fiftyone.utils.huggingface as fouh

# Load the dataset
# Note: other available arguments include 'max_samples', etc
dataset = fouh.load_from_hub(
    "Voxel51/ColoRadar-Sample",
    name="ColoRadar-Sample",
    persistent=True,
)

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

Dataset Details#

Dataset Description#

7 sequences totalling 13m 46s of recording, 8,263 LiDAR scans holding 382.6 million points, 4,145 cascaded radar frames and 8,340 single-chip radar scans.

  • Curated by: Autonomous Robotics and Perception Group, University of Colorado Boulder (source release)

  • Funded by: [More Information Needed]

  • Shared by: Voxel51 (FiftyOne conversion)

  • Language(s): Not applicable (sensor data)

  • License: Apache 2.0

Dataset Sources#

Uses#

Direct Use#

Research on millimetre-wave radar, LiDAR and inertial sensing, including SLAM, in indoor, outdoor and subterranean environments, using the LiDAR-inertial ground-truth trajectory shipped with each sequence.

Out-of-Scope Use#

[More Information Needed]

Dataset Structure#

Topology

An ungrouped FiftyOne dataset with media_type="multimodal". One sample is one sequence, and its filepath is a .fo.mcap file. There are 7 samples, no sample tags, and an empty dataset.info. The dataset has no FiftyOne label fields; every signal lives inside the MCAP file as a channel and is shown by the App’s multimodal viewer.

Sample fields

Field

FiftyOne type

Description

id, filepath, tags, metadata, created_at, last_modified_at

built-in

Standard FiftyOne sample fields

sequence

StringField

Sequence name, e.g. arpg_lab_run2

environment

StringField

The release’s own description of the sequence’s group, e.g. Indoor built environment

recorded

StringField

Recording date

duration

FloatField

Sequence length in seconds

num_lidar_scans

IntField

LiDAR scans

num_lidar_points

IntField

LiDAR points across all scans

num_cascade_frames

IntField

Cascaded radar frames

num_cascade_points

IntField

Points derived from the cascaded radar heatmaps across all frames

num_single_chip_scans

IntField

Single-chip radar scans

num_single_chip_points

IntField

Single-chip radar points across all scans

num_imu_samples

IntField

Inertial samples

ground_truth_path_m

FloatField

Length of the LiDAR-inertial ground-truth trajectory in metres

has_motion_capture

BooleanField

Whether the sequence carries a motion-capture pose track

Episode contents

Each MCAP episode contains these channels:

Channel

Schema or content

/lidar-points

Ouster OS1-64 scans, foxglove.PointCloud with x, y, z, intensity, reflectivity and ring

/single-chip-radar-points

The AWR1843’s own point clouds, foxglove.PointCloud with x, y, z, intensity, range and doppler

/cascade-radar-points

Points derived from each cascaded radar heatmap, foxglove.PointCloud with x, y, z, intensity and doppler

/cascade-radar-bev

A top-down render of each cascaded radar heatmap, foxglove.CompressedVideo

/imu.plot

The Microstrain inertial unit at about 100 Hz

/ground-truth, /ground-truth.plot

The release’s LiDAR-inertial trajectory as foxglove.PoseInFrame, with its position as a plot

/motion-capture, /motion-capture.plot

The Vicon pose of the rig in the motion-capture space, with its position as a plot

/tf

Each sensor’s pose in the rig’s base frame, foxglove.FrameTransform

/sequence

Names the sequence and its environment

Sequences

Sequence

Environment

Recorded

Duration

LiDAR scans

Cascade frames

Single-chip scans

Path

arpg_lab_run2

Indoor built environment

2020-12-21

1m 41s

1,014

504

1,015

67 m

aspen_run9

Large motion capture space

2021-02-24

1m 16s

765

414

836

42 m

ec_hallways_run4

Indoor built environment

2020-12-21

1m 30s

899

447

902

103 m

edgar_army_run5

Large subterranean environment

2021-02-23

1m 54s

1,141

567

1,141

108 m

edgar_classroom_run4

Narrow subterranean environment

2021-02-23

2m 44s

1,639

816

1,639

184 m

longboard_run1

High-speed multi-use paths and roads

2021-02-22

2m 57s

1,767

882

1,770

766 m

outdoors_run8

Outdoor built environment

2021-03-02

1m 44s

1,038

515

1,037

121 m

Parsing decisions

  • Why MCAP: the dataset is one multimodal sample per sequence, so the radars, LiDAR, inertial unit, plots and poses play back on one shared clock.

  • ROS 1 bags: the bags are ROS 1 and were read without a ROS install.

  • LiDAR and single-chip radar: the LiDAR scans drop the returns the sensor marks empty, and the single-chip radar’s point clouds are carried as recorded.

  • Cascaded radar points: the cascaded radar publishes no point cloud, only a heatmap of 128 range by 128 azimuth by 32 elevation cells, each holding an intensity and a Doppler velocity. Its points are derived the way the release’s plotting tool derives them: every cell beyond the tenth range bin whose intensity exceeds a fifth of the way from the frame’s weakest cell to its strongest, placed at the cell’s centre. The release’s tool also thins them on a 0.3 m grid; these are not thinned.

  • Cascaded radar render: /cascade-radar-bev draws each heatmap from above, the strongest cell in each column of elevations on a scale of 30 dB below the frame’s peak, with the radar facing up.

  • Calibration: /tf is the release’s calibration, each sensor’s pose in the rig’s base frame, with the single-chip radar’s heatmap frame dca_link placed on its point-cloud frame as the bags place it.

  • Left out: the raw ADC samples of both radars and the single-chip radar’s heatmaps are not carried, nor are the bags’ own /tf stream and the LiDAR’s built-in inertial unit; all are in the release.

Dataset Creation#

Curation Rationale#

[More Information Needed]

Source Data#

Data Collection and Processing#

Recorded with a handheld rig carrying a TI MMWCAS-RF-EVM cascaded imaging radar, a TI AWR1843BOOST single-chip radar, an Ouster OS1-64 LiDAR and a Lord Microstrain 3DM-GX5-25 inertial unit. The FiftyOne conversion reads the ROS 1 bags without a ROS install and writes one MCAP episode per bag; see Parsing decisions above for the changes made.

Who are the source data producers?#

The Autonomous Robotics and Perception Group at the University of Colorado Boulder.

Annotations#

Annotation process#

The dataset has no human annotations. The ground-truth trajectory in each sequence is the release’s LiDAR-inertial trajectory.

Who are the annotators?#

Not applicable.

Personal and Sensitive Information#

[More Information Needed]

Citation#

Work using the dataset should cite:

BibTeX:

@article{kramer2022coloradar,
  title={ColoRadar: The direct 3D millimeter wave radar dataset},
  author={Kramer, Andrew and Harlow, Kyle and Williams, Christopher and Heckman, Christoffer},
  journal={The International Journal of Robotics Research},
  volume={41},
  number={4},
  pages={351--360},
  year={2022},
  doi={10.1177/02783649211068535}
}

APA:

Kramer, A., Harlow, K., Williams, C., & Heckman, C. (2022). ColoRadar: The direct 3D millimeter wave radar dataset. The International Journal of Robotics Research, 41(4), 351-360. https://doi.org/10.1177/02783649211068535

More Information#

The source release is distributed under the Apache License 2.0, and this conversion is distributed under the same license, whose text is in the LICENSE file beside this card.

Changes from the source: 7 of the release’s 52 sequences, converted from ROS 1 bags to the FiftyOne MCAP flavor, the LiDAR’s empty returns dropped, points derived from the cascaded radar’s heatmaps with a top-down render of each, the release’s calibration carried as transforms, and the raw ADC samples, the single-chip heatmaps and the bags’ own /tf left out.

Dataset Card Authors#

[More Information Needed]

Dataset Card Contact#

[More Information Needed]