Documentation Status

Cannot load information on name: tensor_field_nav, distro: electric, which means that it is not yet in our index. Please see this page for information on how to submit your repository to our index.
Cannot load information on name: tensor_field_nav, distro: fuerte, which means that it is not yet in our index. Please see this page for information on how to submit your repository to our index.
Cannot load information on name: tensor_field_nav, distro: groovy, which means that it is not yet in our index. Please see this page for information on how to submit your repository to our index.
Cannot load information on name: tensor_field_nav, distro: hydro, which means that it is not yet in our index. Please see this page for information on how to submit your repository to our index.
Cannot load information on name: tensor_field_nav, distro: indigo, which means that it is not yet in our index. Please see this page for information on how to submit your repository to our index.
Cannot load information on name: tensor_field_nav, distro: jade, which means that it is not yet in our index. Please see this page for information on how to submit your repository to our index.
Cannot load information on name: tensor_field_nav, distro: kinetic, which means that it is not yet in our index. Please see this page for information on how to submit your repository to our index.
Cannot load information on name: tensor_field_nav, distro: lunar, which means that it is not yet in our index. Please see this page for information on how to submit your repository to our index.

Introduction

Tensor_field_nav package is developed for autonomous mapping of unknown indoor scenes by a mobile robot holding an RGBD camera. The key idea is to utilize 2D directional fields to guide robot movement. We compute and update a geometry-aware tensor field constrained by the currently reconstructed scene. The 3D scene geometry (i.e., the known surfaces) is projected to the floor plane. A set of 2D tangential constraints along the projected boundaries is extracted and used to compute/update the tensor field. The robot path is formed by particle advection over the tensor field, which is inherently obstacle avoiding.

During online scanning, the tensor field is updated in real-time, conforming to the incrementally reconstructed scene. To ensure a smooth robot path when advecting over the time-varying field, we propose a space-time optimization of tensor fields via imposing both spatial smoothness and temporal coherence. There are several important advantages of tensor field guided navigation. First, tensor fields are orientation-free and thus contain much less singularities (degenerate points), as compared to vector fields which are predominantly used in the literature. Fewer singularities lead not only to smoother path advection, which is critical for quality reconstruction, but also to more efficient navigation due to less ambiguity. In addition, tensor fields are sink-free, avoiding the issue of local trapping. Most importantly, the topological skeleton of a tensor field, comprised of all degenerate points and the separatrices connecting them, can be viewed as a routing graph. With this global structure, one can achieve global path planning for efficient scene scanning.

Overview

Our tensor_field_nav package is composed of five sub-packages:

1) octomap_tensor_field for mapping based on Octomap

2) odom_tf_publish for monitoring robot position

3) pure_pursuit_controller for controlling robot movement

4) rviz_textured_quads for visualizing tensor field in rviz

5) tensor_field_nav_core is the core sub-package of our method used for computing and updating tensor field and achieving field-guided robot exploration and scanning.

Contributors and Publication

The contributors to this package are Lintao Zheng(lintaozheng1991@gmai.com) and Kai Xu(kevin.kai.xu@gmail.com). This method has been published at SIGGRAPH Asia 2017. Please refer to the project page. If you find the package useful, please consider citing the following paper:

@article {xu_siga16,
    title = {Autonomous Reconstruction of Unknown Indoor Scenes Guided by Time-varying Tensor Fields},
    author = {Kai Xu and Lintao Zheng and Zihao Yan and Guohang Yan and Eugene Zhang and Matthias Niessner and Oliver Deussen and Daniel Cohen-Or and Hui Huang},
    journal = {ACM Transactions on Graphics (Proc. of SIGGRAPH Asia 2017)},
    volume = {36},
    number = {6},
    pages = {Article No. ??},
    year = {2016}
}

Download and Install

Dependencies

This package depends on CUDA and OpenCV. Please install these two libraries first. CUDA version 7.5 and OpenCV 2.4.8 version were tested and thus recommended.

From source

Please download the source code from:

git clone https://github.com/zlt1991/tensor_field_nav.git

then

catkin_make

Run

$ roslaunch tensor_field_nav_core complete_run.launch
$ roslaunch octomap_tensor_field octomap_mapping.launch

Acknowledgements

The implementation of our tensor field navigation is based on the code of tensor field design provided by Eugene Zhang. If you use this code, please also cite the following paper:

@article{zhang2007interactive,
  title={Interactive tensor field design and visualization on surfaces},
  author={Zhang, Eugene and Hays, James and Turk, Greg},
  journal={IEEE Transactions on Visualization and Computer Graphics},
  volume={13},
  number={1},
  year={2007},
  publisher={IEEE}
}

Wiki: tensor_field_nav (last edited 2017-08-10 01:19:39 by lintao)