Language-Guided Pattern Formation for Swarm Robotics with Multi-Agent Reinforcement Learning
IROS2024
TL;DR This paper explores how to leverage the vast knowledge encoded in large language models to tackle pattern formation challenges for swarm robotics systems.
Overview
This paper explores leveraging the vast knowledge encoded in Large Language Models (LLMs) to tackle pattern formation challenges for swarm robotics systems. A new framework, named LGPF (Language-Guided Pattern Formation), is proposed to address these challenges. The framework breaks down the pattern formation into two key components: pattern synthesis and swarm robotics control. For the former, this study utilizes the exceptional few-shot generalizability of LLMs to translate high-level natural language descriptions into the desired spatial pattern coordinates. This approach allows for overcoming previous limitations in representing and designing complex patterns. The framework further employs a centralized training with decentralized execution (CTDE) based multi-agent reinforcement learning (MARL) approach to control the swarm robots in forming the specified pattern while avoiding collisions. The decentralized policies learned with the CTDE-based MARL algorithm consider coordination between robots without direct communication under a partially observable setup. To validate the effectiveness of our framework, we perform extensive experiments in both simulation and real-world environments. These experiments validate LGPF's effectiveness in accurately and safely forming diverse user-specified patterns.
Video
An LLM as Swarm Pattern Generator
The following image shows an overview of our LGPF. Given a desired pattern in language (e.g., circle, tree, house), we ask an LLM to generate coordinates of swarm robots. Then, the robots are ordered to move to assigned coordinates. The robots are trained by a multi-agent RL algorithm so that they move smoothly to target positions while avoiding collisions and solving partial observability.
Example Patterns Generated by an LLM
We tested the ability of an LLM to generate patterns for swarm robots. As the following image shows, an LLM is able to generate diverse patterns, ranging from alphabets, geometric shapes, and even complex objects such as houses. This is surprising since, in contrast to line drawing, an LLM needs to take into account the number of robots so that dots representing an object indeed look like a specified object.
Swarm Robot Patterns in a Simulator
Generating coodinates alone is far less sufficient for actually letting swarm robots to form a pattern. We trained the swarm robots to move to target positions while avoiding collisions with other robots and solving partial observability due to local observation of each robot. The following image shows actuall patterns formed by the swarm robots in a simulated environment.
Swarm Robot Patterns in a Real Environment
Finally, we transferred policies of swarm robots to a real environment. The following image shows how our robots react to changing orders and how they move to form desired shapes. For details of the robots we used, please refer to the "maru" paper cited at the bottom of this project page.
Acknowledgement
The authors gratefully acknowledge the support from the National Science and Technology Council (NSTC) in Taiwan under grant numbers MOST 111-2223-E-002-011-MY3, NSTC 113-2221-E-002-212-MY3, and NSTC 113-2640-E-002-003. The authors would also like to express their appreciation for the donation of the GPUs from NVIDIA Corporation and NVIDIA AI Technology Center (NVAITC) used in this work. Furthermore, the authors extend their gratitude to the National Center for High-Performance Computing (NCHC) for providing the necessary computational and storage resources.
OMRON SINIC X is Hiring!
OMRON SINIC X continuously hires internship students. We are a completely research-oriented company aiming at publishing papers in top conferences and journals such as IROS. We encourage internship students to present their results by themselves and provide travel support. As a matter of course, we provide competitive salary, accomodation, and visa support. If you live in Japan, remote work is also possible.
For details, please refer to call for interns.
Contact
Citation
@inproceedings{liu2024language,
title={Language-Guided Pattern Formation for Swarm Robotics with Multi-Agent Reinforcement Learning},
author={Liu, Hsu-Shen and Kuroki, So and Kozuno, Tadashi and Sun, Wei-Fang and Lee, Chun-Yi},
booktitle={2024 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS)},
year={2024}
}
Relevant Projects
maru: a miniature-sized wheeled robot for swarm robotics research
"maru" (= miniature assemblage adaptive robot unit) is a custom-made, miniature-sized, two-wheeled robot designed specifically for tabletop swarm robotics research.
Multi-Agent Behavior Retrieval: Retrieval-Augmented Policy Training for Cooperative Push Manipulation by Mobile Robots
The multi-agent coordination skill database allows multiple mobile robots to efficiently use past memories to adapt to new tasks.