Skip to content

Implementation of the SADG RHC feedback control scheme to reduce route completion times of delayed agents following MAPF plans.

License

Notifications You must be signed in to change notification settings

alexberndt/sadg-controller

Repository files navigation

SADG Controller

sadg-controller is a Python-library containing an implementation of the Switchable Action Dependency Graph (SADG) receding horizon controller (RHC) feedback scheme presented in:

A. Berndt, N. van Duijkeren, L. Palmieri, A. Kleiner, T. Keviczky, "Receding Horizon Re-ordering of Multi-Agent Execution Schedules," to appear in Transactions on Robotics.

SADG Receding Horizon Feedback Control Scheme Typical MAPF Execution Schemes

Our approach significantly reduces the cumulative route completion of agents subjected to large delays by optimizing the ordering of agents based on their progress in a receding horizon fashion, while maintaining collision- and deadlock-free plan execution guarantees.

Switchable Action Dependency Graph Roadmap

Purpose of the Project

This software is a research prototype, solely developed for and published as part of the publication cited above. It is not foreseen to be either actively maintained or monitored in any way.

List of Functionalities

  1. Python-based interface for interacting with MAPF planners from libMultiRobotPlanning
  2. SADG Receding Horizon feedback control scheme implementation

Install with Docker

The most straightforward way to get started is using Docker. This way, you can use the software in Ubuntu, Windows and Mac. To this end first install Docker.

Then, build the image:

docker build -t sadg-controller .

Subsequently, you can run the image:

xhost +  # only for system running an X server 
docker run --rm -it -e DISPLAY=$DISPLAY -v /tmp/.X11-unix:/tmp/.X11-unix:ro sadg-controller bash;
xhost -  # only for system running an X server

From inside the image you can follow the instructions in the Examples section.

Install without Docker

Before proceeding, consider using the Docker-based approach. We present here the typical ROS2-based installation approach that will also prepare the dependency libMultiRobotPlanning automatically.

Install ROS2

We currently support the following version of ROS2:

Make sure to install colcon and rosdep:

sudo apt install python3-rosdep python3-colcon-common-extensions
sudo rosdep init
rosdep update

Prepare workspace

Create a workspace (e.g., in your home folder) and clone sadg-controller.

mkdir -p ~/sadg_ws/src
cd ~/sadg_ws/src
git clone --recurse-submodules https://github.com/alexberndt/sadg-controller.git
cd ~/sadg_ws

Install dependencies

Then, from ~/sadg_ws, execute:

rosdep install --from-paths src --ignore-src -r -y

Install python dependencies:

/usr/bin/python3 -m pip install -r src/sadg-controller/requirements.txt

Build workspace

From ~/sadg_ws, execute:

colcon build --symlink-install  # --symlink-install is optional

Examples

To start a simulation, run the following

source install/setup.bash
ros2 launch sadg_controller scenario.launch.xml agent_count:=8 roadmap:=test visualize_sadg:=True

More generally, one can start scenarios by

source install/setup.bash
ros2 launch sadg_controller scenario.launch.xml agent_count:=<number of agents> roadmap:={test, cover, full_maze, half_maze, islands, warehouse} ecbs_sub_factor:=<ecbs suboptimality factor> visualize_sadg:={True, False}

It is also possible to start the agent nodes, simulator and controller separately using the launch files:

  • ros2 sadg_controller agents.launch.py agent_count:=...
  • ros2 sadg_controller simulation.launch.py agent_count:=... roadmap:=...
  • ros2 sadg_controller controller.launch.py agent_count:=... roadmap:=... ecbs_sub_factor:=... visualize_sadg:=...

The controller generates random start and goal points, which may cause issues in certain configurations. Some configurations that we have managed to run relatively reliably can be found in the table below, note that the limitation here is in fact the randomness (which may be fixed).

roadmap agent_count
test 2, 4, 8, 10
cover 8
full_maze 60
half_maze 40
islands 40
warehouse 40, 60

Example Results

Full Maze Half Maze Warehouse Islands
Full Maze Half Maze Half Maze Half Maze

About

Maintainers

3rd Party Licenses

Details about the 3rd party licenses can be found in 3rd-party-licenses.txt. In short, this repository contains a submodule for the library libMultiRobotPlanning, which comes under the MIT license.

License

sadg-controller comes under the GNU Affero General Public License, Version 3, see LICENSE.

About

Implementation of the SADG RHC feedback control scheme to reduce route completion times of delayed agents following MAPF plans.

Topics

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages