multi-robot-fabrics

Implementation of multi-robot-fabrics presented in our MRS 2023 paper "Multi-Robot Local Motion Planning Using Dynamic Optimization Fabrics"

In this paper, we address the problem of real-time motion planning for multiple robotic manipulators that operate in close proximity. We build upon the concept of dynamic fabrics and extend them to multi-robot systems, referred to as Multi-Robot Dynamic Fabrics (MRDF). This geometric method enables a very high planning frequency for high-dimensional systems at the expense of being reactive and prone to deadlocks. To detect and resolve deadlocks, we propose Rollout Fabrics where MRDF are forward simulated in a decentralized manner. We validate the methods in simulated close-proximity pick-and-place scenarios with multiple manipulators, showing high success rates and real-time performance.

A video showcasing the presented approach can be found here.

The current version of the paper can be cited using the following reference:

@inproceedings{bakker2023multi,
  title={Multi-Robot Local Motion Planning Using Dynamic Optimization Fabrics},
  author={Bakker, Saray and Knoedler, Luzia and Spahn, Max and B{\"o}hmer, Wendelin and Alonso-Mora, Javier},
  booktitle={2023 International Symposium on Multi-Robot and Multi-Agent Systems (MRS)},
  pages={149--155},
  year={2023},
  organization={IEEE}
}

This repository is meant to explore the use of fabrics for multiple mobile robots/robotic manipulators. Several flavors are explored:

Fabrics applied to multi-robot scenarios.
Rollout fabrics applied to multi-robot scenarios. (Forward predictions over a horizon to detect future deadlocks or other undesired states)

The 'examples' folder provides runnable examples of different scenarios.

Teaser

Options

This repository includes examples of the application of multi-robot fabrics to point robots and Panda robotic arms. The examples can be run 1) without rollouts (in the paper referred to as MRDF), 2) with rollout fabrics and deadlock resolution heuristic (in the paper referred to as RF), and 3) with rollout fabrics, constant velocity goal estimation and deadlock resolution (in the paper referred to as RF-CV)

While dynamic fabrics (Spahn2023) are applied in the paper, we also support static fabrics as introduced in Ratliff2020. The point-robot example was added for easy understanding and does currently not provide support for rollout fabrics and deadlock resolution.

Which configuration is used can be accessed in examples/configs. Here, also the number of robots and the number of collision spheres can be adapted. Further parameters can be adapted in 'parameters_manipulators.py'.

Point Robot	2 Panda Scenario	3 Panda Scenario

Installations

Clone this repository and go to its root:

git clone git@github.com:tud-amr/multi-robot-fabrics.git
cd multi-robot-fabrics

You can install the package using poetry. For more details on poetry see installation instructions.

poetry install

The requirements can be found in pyproject.toml.

Usage

Enter the virtual environment using:

poetry shell

In the folder multi_robot_fabrics run

python examples/<example-file-name>

E.g. to run the panda example python examples/example_pandas_Jointspace.py.

Troubleshooting

If you run into problems of any kind or have a question, do not hesitate to open an issue on this repository. Or have a look at the tips we summarized here.

tud-amr / multi-robot-fabrics

readme