NeBula Multi-Robot Processor ROS2

Playback and analyze the NeBula multi-robot dataset with nebula_multirobot_processor.py in ROS2. Tested in ROS2 Humble.

Dataset structure and preparation

• Download the dataset and extract it.
• Convert robot.bag in the groundtruth folder and robot_odom.bag in the rosbag folder to ROS2 using the rosbags python package.

rosbags-convert husky4_odom.bag # for all bag files

You end up with the following folder structure, e.g. urban dataset. ROS2 folders and files are marked with a # ROS2 comment:

.
└── urban
    ├── g2o_pcd
    │   ├── pcd.zip
    │   └── result.g2o
    ├── ground_truth
    │   ├── husky1_odom             # ROS2
    │   │   ├── husky1_odom.db3     # ROS2
    │   │   ├── metadata.yaml       # ROS2
    │   ├── husky1_odom.bag
    │   ├── husky4_odom             # ROS2
    │   │   ├── husky4_odom.db3     # ROS2   
    │   │   ├── metadata.yaml       # ROS2
    │   ├── husky4_odom.bag
    │   ├── result.g2o
    │   ├── spot1_odom              # ROS2
    │   │   ├── metadata.yaml       # ROS2
    │   │   ├── spot1_odom.db3      # ROS2
    │   ├── spot1_odom.bag
    │   └── urban.pcd
    └── rosbag
        ├── husky1                  # ROS2
        │   ├── husky1.db3          # ROS2
        │   └── metadata.yaml       # ROS2
        ├── husky1.bag
        ├── husky4                  # ROS2
        │   ├── husky4.db3          # ROS2
        │   └── metadata.yaml       # ROS2
        ├── husky4.bag
        ├── spot1                   # ROS2
        │   ├── metadata.yaml       # ROS2
        │   └── spot1.db3           # ROS2
        └── spot1.bag

Usage

Prerequisites

ROS2 needs to be installed. I tested it with ROS2 Humble.

For playing back the data you need the pose_graph_msgs package from the lamp repository compiled and sourced. You can find a ROS2 version of the package in this repository. Take a look at this pose_graph_msgs README.md for more information.

Besides the ROS2 python packages you need the following python packages:

• rosbags for converting the rosbags to ROS2
• python-fire
• matplotlib
• numpy
• pyquaternion

pip3 install rosbags python-fire matplotlib numpy pyquaternion

Play back data of any number of robots

The play_rosbags function of the nebula_multirobot_processor.py script can be used to play back the data of multiple robots in order of their time stamps. It uses the python fire module to execute different functions from the command line.

The processor class is a ROS2 node which reads in the .db3 files of the robots and publishes the pointcloud and odometry data. The dataset_dir parameter needs to be set to the path of the dataset folder [urban, tunnel, prelim2, ku]. The rate parameter can be used to set the playback rate. The robot_names parameter is a list of the robot names which should be played back.

The point cloud of each keyframe will be published together with the closest odometry message. The topics can be set in the script.

python3 nebula_multirobot_processor.py play_rosbags --ros-args -p dataset_dir:=/path/to/data/dir/urban/ -p rate:=10.0 -p robot_names:="[husky1, husky4]"

Print dataset info

The print_info function of the nebula_multirobot_processor.py script can be used to print information about the dataset. The dataset_dir parameter needs to be set.

python3 nebula_multirobot_processor.py print_info --ros-args -p dataset_dir:=/path/to/data/dir/tunnel/ -p robot_names:="[husky3]"

Plot 3D trajectories

The plot_trajectories function of the nebula_multirobot_processor.py script can be used to plot the 3D trajectories of the robots, to get a general idea of the robots' trajectories.

python3 nebula_multirobot_processor.py plot_trajectories --ros-args -p dataset_dir:=/path/to/data/dir/tunnel/ -p robot_names:="[husky3, husky4]"

Ground truth pose file creation

The write_odom_groundtruth function function of the nebula_multirobot_processor.py script can be used to create ground truth pose files (timestamp x y z qx qy qz qw) for the robots. However for simplicity I added the pose files in the odom_groundtruth of this repo.