Fraunhofer Italia Behavior Tree Stack

Author: Michael Terzer - Fraunhofer Italia

This stack of ROS2 packages provides a framework to manage and execute behavior trees. It provides a plugin-based architecture to load custom behavior tree nodes, or so called skills. It is based on the BehaviorTree.CPP library and uses the interfaces of the BehaviorTree.ROS2 repository.

The stack provides following important features:

• Remote missions: send a tree’s full XML and its blackboard as an action goal. A Python client and CLI are included.
• Runtime control: run, pause, resume, stop, and step a tree one tick at a time for debugging.
• Automatic tree loading: trees are found recursively by name, and shared subtrees can be loaded from separate folders.
• Plugin skills: skill libraries are pluginlib plugins built on BehaviorTree.ROS2 wrappers, with a template package (bt_demo_skills) to start from.

The bt_executor package hosts several classes to execute a behavior tree and to manage the so called skill libraries, being behavior-tree-node-plugins that can be hosted in separate packages somewhere in your workspace. A demo of such skills can be found in bt_demo_skills.

The bt_action_server package implements a ROS2 action server that uses a bt_executor instance to run a tree. The tree can be loaded from a path or sent as a string, accompanied with a blackboard to the executor.

The bt_service_server package implements mutiple ROS2 service servers that wrap some base-class functions from the bt_executor package as ROS2 services.

The behavior_tree_msgs package collects message types that are used within this stack.

The bt_demo_skills package demonstrates the implementation of skills that are loaded by the bt_skill_manager from the bt_executor package.

For more detailed information, please see the specific package documentations.

Following graphics shows the class and package diagram.

Cloning

This repository comes with a vcs-tool setup. Import the needed repositories by:

vcs import < .repos

Docker

The stack comes with a docker-compose setup.

Build the docker

To build the docker, navigate to the docker directory and run:

cd docker
docker compose build

Run a simple demo in the docker container

Run the docker:

docker compose up 

Open a bash of the container:

docker exec -it bt_stack /bin/bash

Within the container, build the workspace:

colcon build

Run the demo application in the container:

source install/setup.bash && ros2 launch bt_demo_skills behavior_tree.launch.py

Run tests (inside container):

./src/run_tests.bash

Licence & Acknowledgement

The behavior_tree_stack is licensed under the terms of the Apache License 2.0. Parts of the code are inspired from the BehaviorTree.CPP and BehaviorTree.ROS2 libraries. BehaviorTree.ROS2 in turn was inspired from Nav2. For this reason, we keep the same license and copyright. For more information, consult the LICENCE file.

The project has recieved financial support by the Horizon 2020 EU Project CONCERT.

Citation

Please find parts of this work in the below article. Any citation would be very appreciated!

@article{Terzer2024,
  title = {Facilitated Construction Robot Programming based on Building Information Modelling},
  author = {Terzer, Michael and Flatscher, Tobit and Magri, Marco and Garbin, Simone and Emig, Julius and Giusti, Andrea},
  journal = {10th International Conference on Control, Decision and Information Technologies},
  year = {2024},
  publisher={IEEE}
}