This is a ROS node (targeting ROS noetic) for playing the Lidar, Camera, and GPS/INS modalities of the Oxford Radar RobotCar Dataset in real-time. It supports playing multiple Robotcars simultaneously within their own ROS namespaces, e.g. to test collaborative SLAM systems.
The player was created as a testbed for CURB-OSG, a collaborative open-vocabulary scene graph mapping system focussed on urban driving scenes.
Note: This code is adapted from radar-robotcar-dataset-ros by Rongxi Zhang.
Changes:
-
No saving to rosbag
-
No radar support
-
Can play multiple sequences in parallel to simulate a multi-agent setup
-
Sensor streams play synchronously by coupling playback to real-time
-
Static TF publishing via URDF
-
Delayed playback and 'clipping' feature to start in the middle of a sequence (see parameters in launch file)
-
Some refactoring and renaming
-
Uses ROS logging interface
Obtain the data and other resources from here (requires permission):
- Oxford Radar RobotCar Dataset: Download for any dataset you want: files for mono cameras, stereo camera, lidar left and right (processed point clouds in .bin format), and GPS/INS. Extract to the same folder, then set the path in launch file.
- Base Robotcar SDK: Important for getting camera calibrations, no code is used. Clone and set path in launch file.
- Radar Robotcar SDK Extension: Not required, but includes a downloader.
Image credit: the authors of Radar RobotCar
Only the following sensors are played (no 2D lidars, no radar):
-
Cameras: 1 x Point Grey Bumblebee XB3 3 x Point Grey Grasshopper2
-
GPS/INS: 1 x NovAtel SPAN-CPT ALIGN inertial and GPS navigation system
-
3D LiDAR: 2 x Velodyne HDL-32E - Mounted to the left and right of the Navtech CTS350-X radar.
To play more than one dataset at once (e.g. to simulate a multi-agent setup), very fast disk read speed is required.
- Download the dataset and SDK
- Place this ROS package in the
catkin_ws/srcdirectory of your catkin workspace. - Install dependencies manually (See CMakeLists.txt and package.xml) or via rosdep:
rosdep install --from-paths -i -y catkin_ws/src - Build the workspace:
catkin build -DCMAKE_BUILD_TYPE=Releaseorcatkin build -DCMAKE_BUILD_TYPE=Debug - Make sure to source your
catkin_ws/devel/setup.bashagain. - Adjust the settings in
launch/radar_robotcar_player.launchand run:roslaunch radar_robotcar_player radar_robotcar_player.launch
The included Rviz config displays all six cameras and the Lidar:
rviz -d ./launch/rviz/radar_robotcar_player.rviz
Frames:
The frames are configured in src/radar_robotcar_player.cpp and launch/urdf/robotcar.xml:
const std::string mono_left_frame = "mono_left",
mono_right_frame = "mono_right",
mono_rear_frame = "mono_rear",
stereo_left_frame = "stereo_left",
stereo_centre_frame = "stereo_centre",
stereo_right_frame = "stereo_right",
lidar_left_frame = "velodyne_left",
lidar_right_frame = "velodyne_right",
gps_frame = "gps_ins",
initial_frame = "initial";They will be prefixed by the agent_name, e.g. robotcar_0/mono_left. The launch file will start an URDF publisher for each robotcar instance. A base_link frame is placed approximately at the GPS frame, and the transform between world and base_link will be read from the GPS/INS solution and published. The initial GPS/INS pose will be repeatedly published via TF.
Topics:
# GPS/INS
gps/gps
ins/gps
ins/imu
ins/pose
# Lidar
lidar/left
lidar/right
# Mono info
mono/info/left
mono/info/rear
mono/info/right
# Mono images
mono/left
mono/rear
mono/right
# Stereo info
stereo/info/narrow/left
stereo/info/narrow/right
stereo/info/wide/left
stereo/info/wide/right
# Stereo images
stereo/centre
stereo/left
stereo/right