VDB-EDT: An Efficient Euclidean Distance Transform Algorithm Based on VDB Data Structure

The video demonstration for ICRA-2021 paper can be found here https://youtu.be/Bojh6ylYUOo

VDB-EDT is an efficient and robust library for large-scale occupancy grid mapping and Euclidean distance transform. Compared with the state-of-the-art method, VDB-EDT can provide high-quality and more complete mapping results, while consuming less memory and processing time.

Installation

1. This package is developed based on ROS Kinetic and the Eigen3 library is also needed.

2. Dependencies for OpenVDB and the Catkin wrapper

sudo apt-get install libglfw3-dev libblosc-dev libopenexr-dev

sudo apt-get install liblog4cplus-dev

3. Download this package and go to the VDB-EDT folder

   cd src

   catkin_init_workspace

   cd ..

   catkin build (currently not support catkin_make command )

Download Dataset

1. Please click here to download the cow-and-lady dataset.

   

2. Please click here to download the large-scale SubT dataset (background grid is with a size of 50m*50m).

   

Run VDB-EDT

1. Make sure the current path is at VDB-EDT folder, i.e., in the catkin workspace

   soource devel/setup.bash

   roslaunch vdb_edt VDB-EDT.launch

2. Play the cow-and-lady dataset. There exist a severe synchronization problem in this dataset. To sufficiently process this dataset, a play rate of 0.1 is recommended

   rosbag play data.bag -r 0.1

3. After finishing testing cow-and-lady dataset, play the SubT dataset

   rosbag play SubT.bag

Results

Results on cow-and-lady dataset

Results on SubT dataset

Acknowledgement

The VDB Ros Wrapper inside this package is developed by Rohit Garg from Air Lab CMU based on the work of ETHZ ASL. We give our special acknowledgement to him.