Using AI to automate real-time decision making for swarm robots

Researchers earn IEEE MRS Best Paper Award for work with broad applications in disaster response, environment monitoring, and military operations

Students work in a lab.

Researchers earned a Best Paper Award from IEEE for their work on the use of AI to automate swarm robotics, a combination of small ground robots and drones, for a wide range of real-world applications. Photo by Douglas Levere.

by Nicole Capozziello

Published January 7, 2022

A team of University at Buffalo researchers received the Best Paper award at the 2021 Institute for Electrical and Electronics Engineers (IEEE) International Symposium on Multi-Robot and Multi-Agent Systems (MRS). 

Print
Souma Chowdhury.
“Our research outcomes could potentially enhance the confidence of the scientific community in translating swarm robot technology to the real-world, as well as bring to light the promise of AI to play a pivotal role in making this happen.”
Souma Chowdhury, associate professor
Department of Mechanical and Aerospace Engineering

The prestigious, single-track conference in the field of robotics takes place every two years and has a paper acceptance rate of around 30%. The Best Paper Award is given based on both the live presentation made at the conference and the assessment of the paper.

The paper, “Learning Robot Swarm Tactics over Complex Adversarial Environments,” lies in the area of swarm robotics, a critical sub-field of robotics. It will be included in the conference proceedings, which are forthcoming.

Swarm robotics is the study of using large scalable teams of simple robots – generally small ground robots and drones – in a wide range of applications, such as disaster response, environment monitoring, military operations and space exploration. 

“It’s really exciting to be recognized for our work in this field,” says Souma Chowdhury, an associate professor of mechanical and aerospace engineering. “Our research is among the first to develop AI models to perform automated tactical level decision-making for ongoing swarm operations, including deciding how to divide the swarm into squads, what tasks to allocate to each squad and the degree of risks each squad is allowed to take subject to environmental adversities.” 

The research was accomplished through the Defense Advance Research Project Agency’s (DARPA) OFFensive Swarm-Enabled Tactics (OFFSET) program. Teams in the program had the opportunity to test their work during experiments performed by industry-led integrator teams, who used around 60 ground and aerial vehicles at the Leschi Town Combined Arms Collective Training Facility, located near Tacoma, Washington, in the fall of 2020.

“I am very proud of this work because we not only developed novel end-learning techniques for robot swarms but also demonstrated them in DARPA’s simulation and multi-robot testbed environment. These demonstrations required immense collaboration between many master's and PhD students across the two departments, and required a massive software engineering effort,” says Karthik Dantu, an associate professor of computer science and engineering and co-author of the paper.

Members of the team will continue to build on this work in 2022 through a newer Defense University Research Instrumentation Program (DURIP) grant. 

Chowdhury says, “Our research outcomes could potentially enhance the confidence of the scientific community in translating swarm robot technology to the real-world, as well as bring to light the promise of AI to play a pivotal role in making this happen.”

The paper was the work of faculty, students and recent graduates from the Departments of Mechanical and Aerospace Engineering (MAE) and Computer Science and Engineering (CSE) in the School of Engineering and Applied Sciences.

In addition to Chowdhury, MAE authors include Ehsan Esfahani, associate professor; Amir Behjat, Hemanth Manjunatha and Payam Ghassemi, all recent PhD graduates; Prajit Krisshna Kumar and Joseph Distefano, both current PhD students; and Apurv Jani and Leighton Collins, both recent master’s graduates. CSE authors include Dantu and David Doermann, SUNY Empire Innovation Professor and director of the Institute of Artificial Intelligence and Data Science.

“The significant contributions made by a team of exceptionally persistent and enterprising graduate students demonstrates that this work is a springboard for future engineering and research leaders,” says Chowdhury. “Their expertise will be especially needed in the intersection of robotics and AI, an area in which the United States is keen on retaining and expanding its global leadership.”

The IEEE MRS Conference was held virtually and in person at the University of Cambridge, UK, from November 4-5, 2021. The team attended virtually.