CLI Tools¶

Usage

The CLI tools use tyro:

Positional arguments ("required") are passed as positional command line arguments
Named arguments are passed as flagged command line arguments

`roverd blobify`¶

Convert a trace to use blob channels.

uv run roverd blobify /path/to/src /path/to/dst

Warning

This script requires ffmpeg to be installed, which we use as a fast way to split mjpeg video files into individual frames.

Danger

Blob channels store each sample as a separate file, which can lead to a very large number of files and high file system overhead. This should only done on a proper blob storage backend, e.g. AWS S3 or Azure Blob Storage! Local file systems or HPC storage systems should stick to the default channel types.

Parameters:

Name	Type	Description	Default
`src`	`str`	path to the source trace.	required
`dst`	`str`	path to the output trace.	required
`workers`	`int`	number of worker threads for writing blobs.	`64`

Source code in format/src/roverd/_cli/blobify.py

def cli_blobify(src: str, dst: str, /, workers: int = 64) -> None:
    """Convert a trace to use blob channels.

    ```sh
    uv run roverd blobify /path/to/src /path/to/dst
    ```

    !!! warning

        This script requires `ffmpeg` to be installed, which we use as a fast
        way to split `mjpeg` video files into individual frames.

    !!! danger

        Blob channels store each sample as a separate file, which can lead to
        a very large number of files and high file system overhead. This should
        only done on a proper blob storage backend, e.g. AWS S3 or Azure Blob
        Storage! Local file systems or HPC storage systems should stick to
        the default channel types.

    Args:
        src: path to the source trace.
        dst: path to the output trace.
        workers: number of worker threads for writing blobs.
    """
    def _copy(file: str) -> None:
        shutil.copy(os.path.join(src, file), os.path.join(dst, file))

    trace = Trace.from_config(src, sensors={
        "_camera": None, "radar": None, "lidar": None,
        "camera": None, "imu": None,
    })

    os.makedirs(dst, exist_ok=True)
    for name, sensor in trace.sensors.items():
        assert isinstance(sensor, DynamicSensor)
        s_copy = DynamicSensor(
            os.path.join(dst, name), create=True, exist_ok=True)

        for ch_name, channel in sensor.channels.items():
            if ch_name in {"ts", "valid", "rot", "acc", "avel"}:
                ch_copy = s_copy.create(ch_name, sensor.config[ch_name])
                ch_copy.write(channel.read(start=0, samples=-1))
            else:
                cfg = {
                    "type": sensor.config[ch_name]['type'],
                    "shape": channel.shape
                }
                if ch_name == 'video.avi':
                    cfg["format"] = "jpg"
                    ch_copy = s_copy.create(ch_name, cfg)
                    os.makedirs(ch_copy.path, exist_ok=True)
                    subprocess.call(
                        f"ffmpeg -i {channel.path} -c:v copy -f image2 "
                        f"{ch_copy.path}/%06d.jpg", shell=True)
                elif ch_name == 'iq':
                    cfg["format"] = "npz"
                    ch_copy = s_copy.create(ch_name, cfg, args={
                        "compress": False, "workers": workers})
                    ch_copy.write(channel.read())
                else:
                    cfg["format"] = "npz"
                    ch_copy = s_copy.create(ch_name, cfg, args={
                        "compress": True, "workers": workers})
                    ch_copy.write(channel.read())

    _copy("lidar/lidar.json")
    _copy("radar/radar.json")
    os.makedirs(os.path.join(dst, "_radar"), exist_ok=True)
    _copy("_radar/pose.npz")
    _copy("config.yaml")

`roverd extract`¶

Extract a subset of a trace.

$ uv run roverd extract /data/grt/bike/point.out extracted-trace \
    --start 0.5 --length 0.001 --relative
$ uv run roverd info extracted-trace
start    1727901429.808
length   1.775
...

Info

Two of start, end, and length must be specified. If relative, these values are specified as a proportion of the trace duration.

Parameters:

Name	Type	Description	Default
`src`	`str`	path to the trace directory.	required
`dst`	`str`	output trace directory.	required
`start`	`float \| None`	start time offset relative to the trace start.	`None`
`end`	`float \| None`	end time offset relative to the trace start (if positive) or trace end (if negative).	`None`
`length`	`float \| None`	length of the extracted trace in seconds.	`None`
`relative`	`bool`	whether the start/end/length values are relative to the trace duration, in seconds.	`False`

Source code in format/src/roverd/_cli/extract.py

def cli_extract(
    src: str, dst: str, /, start: float | None = None, end: float | None = None,
    length: float | None = None, relative: bool = False
) -> None:
    r"""Extract a subset of a trace.

    ```sh
    $ uv run roverd extract /data/grt/bike/point.out extracted-trace \
        --start 0.5 --length 0.001 --relative
    $ uv run roverd info extracted-trace
    start    1727901429.808
    length   1.775
    ...
    ```

    !!! info

        Two of `start`, `end`, and `length` must be specified. If `relative`,
        these values are specified as a proportion of the trace duration.

    Args:
        src: path to the trace directory.
        dst: output trace directory.
        start: start time offset relative to the trace start.
        end: end time offset relative to the trace start (if positive) or
            trace end (if negative).
        length: length of the extracted trace in seconds.
        relative: whether the start/end/length values are relative to the trace
            duration, in seconds.
    """
    trace = Trace.from_config(src)

    trace_start = max(v.metadata.timestamps[0] for v in trace.sensors.values())
    trace_end = min(v.metadata.timestamps[-1] for v in trace.sensors.values())
    duration = trace_end - trace_start

    if sum(x is not None for x in (start, end, length)) < 2:
        raise ValueError(
            "Two of `start`, `end`, and `length` must be specified.")

    if end is None and start is not None and length is not None:
        end = start + length
    if start is None and end is not None and length is not None:
        start = end - length

    if relative:
        start = trace_start + (start * duration if start is not None else 0)
        end = trace_start + (end * duration if end is not None else 0)

    if os.path.exists(dst):
        raise FileExistsError(f"Output directory {dst} already exists.")

    os.makedirs(dst)
    for s_name, sensor in trace.sensors.items():
        assert isinstance(sensor, DynamicSensor)
        s_copy = DynamicSensor(os.path.join(dst, s_name), create=True)

        i_start, i_end = np.searchsorted(
            sensor.metadata.timestamps, np.array([start, end]))

        for ch_name, channel in sensor.channels.items():
            ch_copy = s_copy.create(ch_name, sensor.config[ch_name])
            ch_copy.write(channel.read(i_start, samples=i_end - i_start))

`roverd info`¶

Print trace metadata.

Only metadata for non-virtual (originally collected) sensors is shown.

uv run roverd info /data/grt/bike/point.out

Sample output

$ roverd info /data/grt/bike/point.out
start    1727900496.810
length   1865.947

total    61.2 GB (32.8 MB/s)
radar    29.3 GB (15.7 MB/s, n=37320, t=1865.9s)
    ts        299 KB (rate= 160  B/s)
    iq       29.3 GB (rate=15.7 MB/s)
    valid    37.3 KB (rate=  20  B/s)
camera   21.1 GB (11.3 MB/s, n=56053, t=1868.3s)
    ts        448 KB (rate= 240  B/s)
    video    21.1 GB (raw= 349 GB, ratio=16.52, rate=11.3 MB/s)
lidar    10.8 GB (5.78 MB/s, n=18570, t=1867.1s)
    ts        149 KB (rate=79.6  B/s)
    rfl      2.03 GB (raw=4.87 GB, ratio= 2.39, rate=1.09 MB/s)
    nir      5.27 GB (raw=9.74 GB, ratio= 1.85, rate=2.82 MB/s)
    rng      3.48 GB (raw=9.74 GB, ratio= 2.80, rate=1.86 MB/s)
imu      8.22 MB ( 4.4 KB/s, n=186847, t=1868.3s)
    ts       1.49 MB (rate= 800  B/s)
    rot      2.24 MB (rate= 1.2 KB/s)
    acc      2.24 MB (rate= 1.2 KB/s)
    avel     2.24 MB (rate= 1.2 KB/s)

Parameters:

Name	Type	Description	Default
`path`	`str`	path to the trace directory.	required

Source code in format/src/roverd/_cli/info.py

def cli_info(path: str, /,) -> None:
    """Print trace metadata.

    Only metadata for non-virtual (originally collected) sensors is shown.

    ```sh
    uv run roverd info /data/grt/bike/point.out
    ```

    ??? quote "Sample output"

        ```
        $ roverd info /data/grt/bike/point.out
        start    1727900496.810
        length   1865.947

        total    61.2 GB (32.8 MB/s)
        radar    29.3 GB (15.7 MB/s, n=37320, t=1865.9s)
            ts        299 KB (rate= 160  B/s)
            iq       29.3 GB (rate=15.7 MB/s)
            valid    37.3 KB (rate=  20  B/s)
        camera   21.1 GB (11.3 MB/s, n=56053, t=1868.3s)
            ts        448 KB (rate= 240  B/s)
            video    21.1 GB (raw= 349 GB, ratio=16.52, rate=11.3 MB/s)
        lidar    10.8 GB (5.78 MB/s, n=18570, t=1867.1s)
            ts        149 KB (rate=79.6  B/s)
            rfl      2.03 GB (raw=4.87 GB, ratio= 2.39, rate=1.09 MB/s)
            nir      5.27 GB (raw=9.74 GB, ratio= 1.85, rate=2.82 MB/s)
            rng      3.48 GB (raw=9.74 GB, ratio= 2.80, rate=1.86 MB/s)
        imu      8.22 MB ( 4.4 KB/s, n=186847, t=1868.3s)
            ts       1.49 MB (rate= 800  B/s)
            rot      2.24 MB (rate= 1.2 KB/s)
            acc      2.24 MB (rate= 1.2 KB/s)
            avel     2.24 MB (rate= 1.2 KB/s)
        ```

    Args:
        path: path to the trace directory.
    """
    ds = Trace.from_config(path)

    start = max(v.metadata.timestamps[0] for v in ds.sensors.values())
    end = min(v.metadata.timestamps[-1] for v in ds.sensors.values())
    print(f"start    {start:.3f}")
    print(f"length   {end - start:.3f}")
    print("")

    print("total    {} ({})".format(
        _size(ds.filesize),
        _size(ds.datarate, suffix='/s')))
    for sname, sensor in ds.sensors.items():
        sensor = cast(sensors.DynamicSensor, sensor)
        print("{:8} {} ({}, n={}, t={:.1f}s)".format(
            sname,
            _size(sensor.filesize),
            _size(sensor.datarate, suffix='/s'),
            len(sensor),
            sensor.duration))
        for cname, channel in sensor.channels.items():
            if sensor.config[cname]['format'] != 'raw':
                raw = channel.size * len(sensor)
                print("    {:8} {} (raw={}, ratio={:5.2f}, rate={})".format(
                    cname.split('.')[0],
                    _size(channel.filesize),
                    _size(raw),
                    raw / channel.filesize,
                    _size(channel.filesize / sensor.duration, suffix='/s')))
            else:
                print("    {:8} {} (rate={})".format(
                    cname.split('.')[0],
                    _size(channel.filesize),
                    _size(channel.filesize / sensor.duration, suffix='/s')))

`roverd list`¶

List traces (recursively) in a directory by looking for config.yaml.

$ uv run roverd list /path/to/datasets
example_a/trace1
example_a/trace2
example_b/trace1
...

Tip

This CLI is intended to be piped to other commands (and roverd CLI tools), e.g.:

# Count traces
uv run roverd list /path/to/traces | wc -l
# Loop over traces
for trace in `uv run roverd list /path/to/traces`; do echo $trace; done

Parameters:

Name	Type	Description	Default
`path`	`str`	directory to search inside.	required
`follow_symlinks`	`bool`	whether to follow symlinks when searching.	`False`

Source code in format/src/roverd/_cli/list.py

def cli_list(path: str, /, follow_symlinks: bool = False) -> None:
    """List traces (recursively) in a directory by looking for `config.yaml`.

    ```sh
    $ uv run roverd list /path/to/datasets
    example_a/trace1
    example_a/trace2
    example_b/trace1
    ...
    ```

    !!! tip

        This CLI is intended to be piped to other commands (and `roverd` CLI
        tools), e.g.:
        ```sh
        # Count traces
        uv run roverd list /path/to/traces | wc -l
        # Loop over traces
        for trace in `uv run roverd list /path/to/traces`; do echo $trace; done
        ```

    Args:
        path: directory to search inside.
        follow_symlinks: whether to follow symlinks when searching.
    """
    traces = Dataset.find_traces(path, follow_symlinks=follow_symlinks)
    relative_traces = [os.path.relpath(trace, path) for trace in traces]
    print('\n'.join(relative_traces))

`roverd validate`¶

Validate dataset files.

$ uv run roverd validate /data/grt
Validate: 166 traces with 0 containing errors.
$ echo $?
0  # would be 1 if there were any errors

Usage with custom schema

grt.yaml

camera: ["ts", "video.avi"]
_camera: ["ts", "segment"]
lidar: ["ts", "rng"]
radar: ["ts", "iq"]
imu: ["ts", "rot", "acc", "avel"]
_:
- _camera/pose.npz
- _fusion/indices.npz
- _lidar/pose.npz
- _radar/pose.npz
- _slam/trajectory.csv

/bin/sh

$ roverd validate /data/grt --schema grt.yaml
Validate: 166 traces with 0 containing errors.

Parameters:

Name	Type	Description	Default
`path`	`Sequence[str]`	Target path or list of paths to validate.	required
`schema`	`str \| None`	Dataset file schema (yaml) to check. If not specified, uses a default schema which corresponds to raw files which are expected to be collected by the `red-rover` rig.	`None`
`fix_errors`	`bool`	If `True`, fix consistency errors (data not present, but metadata is present).	`False`

Source code in format/src/roverd/_cli/validate.py

def cli_validate(
    path: Sequence[str], /, schema: str | None = None,
    fix_errors: bool = False
) -> None:
    """Validate dataset files.

    ```sh
    $ uv run roverd validate /data/grt
    Validate: 166 traces with 0 containing errors.
    $ echo $?
    0  # would be 1 if there were any errors
    ```

    ??? info "Usage with custom schema"

        ```yaml title="grt.yaml"
        camera: ["ts", "video.avi"]
        _camera: ["ts", "segment"]
        lidar: ["ts", "rng"]
        radar: ["ts", "iq"]
        imu: ["ts", "rot", "acc", "avel"]
        _:
        - _camera/pose.npz
        - _fusion/indices.npz
        - _lidar/pose.npz
        - _radar/pose.npz
        - _slam/trajectory.csv
        ```

        ```sh title="/bin/sh"
        $ roverd validate /data/grt --schema grt.yaml
        Validate: 166 traces with 0 containing errors.
        ```

    Args:
        path: Target path or list of paths to validate.
        schema: Dataset file schema (yaml) to check. If not specified, uses a
            default schema which corresponds to raw files which are expected
            to be collected by the `red-rover` rig.
        fix_errors: If `True`, fix consistency errors (data not present, but
            metadata is present).
    """
    # We'll get a lot of warings for missing timestamps. Don't warn, since
    # the schema will explicitly catch them if the user cares.
    warnings.filterwarnings(
        "ignore", message="Sensor metadata does not contain 'ts' channel")

    if schema is None:
        _schema = {
            "lidar": ["ts", "rfl", "nir", "rng"],
            "radar": ["ts", "iq", "valid"],
            "camera": ["ts", "video.avi"],
            "imu": ["ts", "rot", "acc", "avel"]
        }
    else:
        with open(schema) as f:
            _schema = yaml.load(f, Loader=yaml.SafeLoader)

    datasets = Dataset.find_traces(*path)
    n_errors = 0
    for path in datasets:
        errors = _validate_schema(path, _schema)
        errors += _validate_consistency(path, fix_errors=fix_errors)
        if errors:
            n_errors += 1
            print(path)
            print('\n'.join("    " + x for x in errors))

    if n_errors > 0:
        print("")
    print(
        f"Validate: {len(datasets)} traces with {n_errors} containing "
        "errors.")
    if n_errors > 0:
        exit(1)

`roverd rosbag`¶

Write lidar and IMU data to a ROS 1 bag.

uv run --extra ros roverd rosbag data/wiselab --min_range 0.5

Warning

This CLI command requires the roverd[ros] extra to be installed.

Danger

Rosbags are incredibly inefficient; expect for the output bag file to be ~10x larger than the input depth maps!

Parameters:

Name	Type	Description	Default
`path`	`str`	data path.	required
`out`	`str \| None`	output rosbag file path; if `None`, uses `_scratch/lidar.bag` in the dataset directory.	`None`
`min_range`	`float \| None`	minimum range (in meters) for lidar points.	`None`

Source code in format/src/roverd/_cli/rosbag.py

def cli_rosbag(
    path: str, /,
    out: str | None = None, min_range: float | None = None
) -> None:
    """Write lidar and IMU data to a ROS 1 bag.

    ```sh
    uv run --extra ros roverd rosbag data/wiselab --min_range 0.5
    ```

    !!! warning

        This CLI command requires the `roverd[ros]` extra to be installed.

    !!! danger

        Rosbags are incredibly inefficient; expect for the output bag file to
        be ~10x larger than the input depth maps!

    Args:
        path: data path.
        out: output rosbag file path; if `None`, uses `_scratch/lidar.bag` in
            the dataset directory.
        min_range: minimum range (in meters) for lidar points.
    """
    try:
        from roverd.transforms.ros import rover_to_rosbag
    except ImportError as e:
        raise ImportError(
            f"Could not import `rover_to_ros` ({e}). Make sure the `ros` "
            f"extra is installed (i.e., `pip install roverd[ros]`).")

    if out is None:
        out = os.path.join(path, "_scratch", "lidar.bag")

    trace = Trace.from_config(
        path, sensors={
            "lidar": partial(sensors.OSLidarDepth, correction="auto"),
            "imu": partial(sensors.IMU, correction="auto")
        })
    lidar = cast(sensors.OSLidarDepth, trace["lidar"])
    imu = cast(sensors.IMU, trace["imu"])

    rover_to_rosbag(out=out, lidar=lidar, imu=imu, min_range=min_range)

CLI Tools¶

roverd blobify¶

roverd extract¶

roverd info¶

roverd list¶

roverd validate¶

roverd rosbag¶

`roverd blobify`¶

`roverd extract`¶

`roverd info`¶

`roverd list`¶

`roverd validate`¶

`roverd rosbag`¶