Mobile Robot Navigation in 3D Meshes

Tutorials • Documentation • Videos • Move Base Flex • Mesh Tools

The Mesh Navigation bundle provides software for efficient robot navigation on 2D manifolds, which are represented in 3D as triangle meshes. It enables safe navigation in various complex outdoor environments by using a modularly extensible layered mesh map. Layers can be loaded as plugins representing specific geometric or semantic metrics of the terrain. This allows the incorporation of obstacles in these complex outdoor environments into path and motion motion planning. The layered Mesh Map is integrated with Move Base Flex (MBF), which provides a universal ROS action interface for path planning, motion control, and for recovery behaviors. We also provide additional planner and controller plugins that run on the layered mesh map.

Publications
Installation
Usage Examples and Demos
Software Stack
Mesh Map
Planners
Controllers
Simulation
Maintain and Contribute
Build Status

Publications

Please reference the following papers when using the navigation stack in your scientific work.

Continuous Shortest Path Vector Field Navigation on 3D Triangular Meshes for Mobile Robots

@inproceedings{puetz21cvp,
    author = {Pütz, Sebastian and Wiemann, Thomas and Kleine Piening, Malte and Hertzberg, Joachim},
    title = {Continuous Shortest Path Vector Field Navigation on 3D Triangular Meshes for Mobile Robots},
    booktitle = {2021 IEEE International Conference on Robotics and Automation (ICRA)},
    year = 2021,
    url = {https://github.com/uos/mesh_navigation},
    note = {Software available at \url{https://github.com/uos/mesh_navigation}}
}

Move Base Flex: A Highly Flexible Navigation Framework for Mobile Robots

@inproceedings{puetz18mbf,
    author = {Sebastian Pütz and Jorge Santos Simón and Joachim Hertzberg},
    title = {{Move Base Flex}: A Highly Flexible Navigation Framework for Mobile Robots},
    booktitle = {2018 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS)},
    year = 2018,
    month = {October},
    url = {https://github.com/magazino/move_base_flex},
    note = {Software available at \url{https://github.com/magazino/move_base_flex}}
}

Installation

ROS Version

This is the active ROS2 branch of this repository, which targets humble. If your are looking for the old ROS1 version, checkout the noetic branch.

Installation from source

Prerequisite: A working ROS 2 installation
Go into a ROS 2 workspace's source directory cd $YOUR_ROS_WS/src.
Clone the repo git clone git@github.com:naturerobots/mesh_navigation.git
Get the tutorial's ROS 2 dependencies
- Clone source dependencies: Run vcs import --input mesh_navigation/source_dependencies.yaml in your ROS 2 workspace source directory.
- Get packaged dependencies: Run rosdep install --from-paths . --ignore-src -r -y from within your ROS 2 workspace source directory.
Build: Go to workspace root cd $YOUR_ROS_WS and run colcon build --packages-up-to mesh_navigation.

Use the pluto_robot package for example HDF5 map datasets, Gazebo simulations, and example configurations.

Usage Examples and Demos

Recommended entrypoint for new users: Check out the mesh_navigation_tutorials for a ready-to-use mesh navigation stack. Complete with simulated environment, rviz config, mesh nav config, etc.

See the pluto_robot bundle for example configurations of the mesh navigatoin stack and usage (ROS 1).

Demos

In the following demo videos we used the developed VFP, i.e., the wavefront_propagatn_planner. It will be renamed soon to vector_field_planner.

Dataset and Description	Demo Video
Botanical Garden of Osnabrück University
Stone Quarry in the Forest Brockum

Stone Quarry in the Forest in Brockum

Colored Point Cloud	Height Diff Layer	RGB Vertex Colors

Software Stack

This mesh_navigation stack provides a navigation server for Move Base Flex (MBF). It provides a couple of configuration files and launch files to start the navigation server with the configured layer plugins for the layered mesh map, and the configured planners and controller to perform path planning and motion control in 3D (or more specifically on 2D-manifold).

The package structure is as follows:

mesh_navigation The corresponding ROS meta package.
mbf_mesh_core contains the plugin interfaces derived from the abstract MBF plugin interfaces to initialize planner and controller plugins with one mesh_map instance. It provides the following three interfaces:
- MeshPlanner - mbf_mesh_core/mesh_planner.h
- MeshController - mbf_mesh_core/mesh_controller.h
- MeshRecovery - mbf_mesh_core/mesh_recovery.h
mbf_mesh_nav contains the mesh navigation server which is built on top of the abstract MBF navigation server. It uses the plugin interfaces in mbf_mesh_core to load and initialize plugins of the types described above.
mesh_map contains an implementation of a mesh map representation building on top of the mesh data structures in lvr2. This package provides a layered mesh map implementation. Layers can be loaded as plugins to allow a highly configurable 3D navigation stack for robots traversing on the ground in outdoor and rough terrain.
mesh_layers The package provides a couple of mesh layers to compute the trafficability of the terrain. Furthermore, these plugins have access to the HDF5 map file and can load and store layer information. The mesh layers can be configured for the robots abilities and needs. Currently we provide the following layer plugins:
- HeightDiffLayer - mesh_layers/HeightDiffLayer
- RoughnessLayer - mesh_layers/RoughnessLayer
- SteepnessLayer - mesh_layers/SteepnessLayer
- RidgeLayer - mesh_layer/RidgeLayer
- InflationLayer - mesh_layers/InflationLayer
dijkstra_mesh_planner contains a mesh planner plugin providing a path planning method based on Dijkstra's algorithm. It plans by using the edges of the mesh map. The propagation start a the goal pose, thus a path from every accessed vertex to the goal pose can be computed. This leads to a sub-optimal potential field, which highly depends on the mesh structure.
cvp_mesh_planner contains a Fast Marching Method (FMM) wave front path planner to take the 2D-manifold into account. This planner is able to plan over the surface, due to that it results in shorter paths than the dijkstra_mesh_planner, since it is not restricted to the edges or topology of the mesh. A comparison is shown below. Please refer to the paper Continuous Shortest Path Vector Field Navigation on 3D Triangular Meshes for Mobile Robots which is stated above.
mesh_client Is an experimental package to additionally load navigation meshes from a server.

Mesh Map

Mesh Layers

The following table gives an overview of all currently implemented layer plugins available in the stack and the corresponding types to specify for usage in the mesh map configuration. An example mesh map configuration is shown below.

Overview of all layers

Layer	Plugin Type Specifier	Description of Cost Computation
HeightDiffLayer	`mesh_layers/HeightDiffLayer`	local radius based height differences
RoughnessLayer	`mesh_layers/RoughnessLayer`	local radius based normal fluctuation
SteepnessLayer	`mesh_layers/SteepnessLayer`	arccos of the normal's z coordinate
RidgeLayer	`mesh_layer/RidgeLayer`	local radius based distance along normal
InflationLayer	`mesh_layers/InflationLayer`	by distance to a lethal vertex

Planners

Currently the following planners are available:

Dijkstra Mesh Planner

  - name: 'dijkstra_mesh_planner'
    type: 'dijkstra_mesh_planner/DijkstraMeshPlanner'

Continuous Vector Field Planner

  - name: 'cvp_mesh_planner'
    type: 'cvp_mesh_planner/CVPMeshPlanner'

MMP Planner

  - name: 'mmp_planner'
    type: 'mmp_planner/MMPPlanner'

The planners are compared to each other.

Vector Field Planner	Dijkstra Mesh Planner	ROS Global Planner on 2.5D DEM

Controllers

Simulation

If you want to test the mesh navigation stack with Pluto please use the simulation setup and the corresponding launch files below for the respective outdoor or rough terrain environment. The mesh tools have to be installed. We developed the Mesh Tools as a package consisting of message definitions, RViz plugins and tools, as well as a persistence layer to store such maps. These tools make the benefits of annotated triangle maps available in ROS and allow to publish, edit and inspect such maps within the existing ROS software stack.

Mesh Localization

For the necessary localization of the robot relative to the mesh, we recommend using RMCL: https://github.com/uos/rmcl. We presented the combination of both software packages at ROSCon 2023: