A2D2 to ROS

Utilities for converting A2D2 data sets to ROS bags.

There is an executuable for each sensor modality: camera, lidar, and bus. Bag files are generated for these modalities independently.

Step-by-step example

For a step-by-step example of how to use the converters, see docs/WALKTHROUGH.md.

Requirements

This package has the following dependencies in addition to standard ROS dependencies:

• RapidJSON: used to load, parse, and validate the JSON data files
• ROS CNPY: used to load .npz files for lidar data

The ROS CNPY package can be downloaded at the above link, and RapidJSON can be installed with:

$ rosdep install a2d2_to_ros --ignore-src -r -y

Recommendations

• Use a dedicated download utility, such as wget, for downloading the data set files. Using a browser can be unreliable, likely due in part to the large size of the files.
• Use the --start-time, --min-time-offset, and --duration converter options to split the converted bag files into smaller timespans (see scripts/convert.sh for an example of doing this). The rosbag utility does a poor job handling extremely large log files, so splitting them up can make them easier to use.

FAQ

docs/FAQ.md contains common questions about the A2D2 data set.

Converters

• Sensor Fusion > Camera
• Sensor Fusion > Lidar
• Sensor Fusion > Bus Signal

Note: The Bus Signal converter also creates a bag file that publishes the TF tree for the vehicle.

Batch conversion

For convenience, a shell script is provided in scripts/convert.sh that can batich convert an entire sensor fusion data set for a single city. Before running the script, be sure to open it and set the configuration options appropriately.

#
# START: CONFIGURATION OPTIONS
#

# The below two paths should point to the data set root and ROS package
package_source=~/catkin_ws/src/a2d2_to_ros
data_root=~/data/a2d2

# Duration (in integer seconds) to record to a single bag file before splitting off a new one
split_duration=7

# Relative location of the sensor fusion data set being converted
# Set 'sensor_data' to the desired location
munich_data= #TODO
gaimersheim_data= #TODO
ingolstadt_data=camera_lidar/20190401_145936
sensor_data=$ingolstadt_data

# Earliest time in the data set (in microseconds) for which all sensor modalities provide data
# Set 'record_start_time' to the appropriate time for the dataset being converted
munich_start_time= #TODO
gaimersheim_start_time= #TODO
ingolstadt_start_time=1554121593489120
record_start_time=$ingolstadt_start_time

# Approximate duration (in integer seconds) of the data set
# Set 'data_set_duration' to the appropriate duration
munich_duration= #TODO
gaimersheim_duration= #TODO
ingolstadt_duration=746
data_set_duration=$ingolstadt_duration

#
# END: CONFIGURATION OPTIONS
#

To run the script:

$ rosrun a2d2_to_ros convert.sh

Note: Be aware that the batch converter skips a bit of time at the beginning of certain sensor logs in order to enusre that split bag files are better time aligned; not all logs begin at the same time. See docs/FAQ.md.

Note: Do not run this script on the preview data set. The preview data set uses a slightly different file layout, so it will not work.

Visualization

An example RViz config is included along a convenience launch file:

$ roslaunch a2d2_to_ros visualize.launch

This launches RViz pre-configured to visualize the TF tree, and the front-facing camera and lidar data.

Compatibility

This code is built and tested under:

• ROS Melodic with Ubuntu 18.04.4
• Clang 6.0.0 with -std=c++14

There is nothing very platform specific, so other reasonably similar system configurations should work.

TODO

A list of open issues for the converter can be found here.

Contact

For any questions or comments about the converter, please reach out to [email protected].