After receiving $150,000 from the Defence Innovation Partnership’s (DIP) Collaborative Research Fund (CRF) under round 7 in January 2024, collaborators DSTG, the University of Adelaide, the University of South Australia and DEWC Services (DEWC) are getting started on exciting new research examining autonomous command and control (C2) of drone swarms.

This project is among six that secured nearly $900k in funding earlier this year through round 7 of DIPs CRF.

Take a read below as David Kilmartin from DSTG explores more about how the project is shaping up.

Novel hybrid machine-learning approach

According to Dr Tim McKay, Director S&T at DEWC, the project will use an unclassified tactical scenario of fire-fighting to explore how innovative artificial intelligence (AI) and machine learning techniques can guide a fleet of self-guided drones to detect and fight bushfires, efficiently and collaboratively. The plan is for DEWC to then take the output of the project and translate it into Defence capability.

Hybrid approach uses the best of both worlds

University of Adelaide Associate Professor Claudia Szabo (Lead Complex Systems Research Group, School of Computer and Mathematical Sciences) has a long-term research interest in drone swarm control for years. She will be providing academic expertise into the project and explains that a novel hybrid approach to multi-agent reinforcement learning will be trialled.

“The use of multi-agent reinforcement learning is appealing in complex, dynamic environments because it adapts and learns very quickly with very little data,’ says Dr Szabo.

“But the challenge is that when you apply it into a real system, it takes a long time to train: there might not be enough data, or the training events might be very sparse. So we’re developing a hybrid approach by adding heuristics.”

Heuristics refers to rules of thumb and expert knowledge about certain situations (e.g. how to fight a fire). Many heuristics exist out there for specific situations, but Dr Szabo says the problem is that they are static; useful once and perhaps not applicable the next time.

“So in this project, we’re basically merging the two techniques and we’re improving the way the reinforcement learning algorithm decides what to do by adding expert system rules and optimisation heuristics.”

The novelty of this hybrid approach, is that it is more accurate and adapts better to changes in the environment; that’s perfect for the highly complex and contested environments that Defence commonly operates in.

Defence scientist Dr Anton Uzunov explains that the main problems that his team tries to solve for its ADF clients require multiple software agents to coordinate activities towards a shared (non-kinetic) outcome, using a variety of autonomous decision-making approaches.

“What we are looking to acquire from this DIP project, specifically, are valuable insights and algorithms that can inform our R&D efforts towards using teams of hybrid learning agents to complement our existing suite of solutions,” says Dr Uzunov.

“DSTG’s involvement in this project will be the contribution of subject-matter expertise and will also shape the research through a knowledge of the relevant Defence problems.”

Fellow Defence scientist Dr Eranda Galhenage says he is seeking to introduce advanced decision-making technologies to ADF clients.

“These technologies are often backed by machine intelligence and AI to gain a competitive advantage for Defence operators confronted by existing and potential problems, providing an upper-hand when it comes to making the best decisions. With the DIP-funded project we intend to find some answers that can extend and enhance our internal R&D efforts and at the same time complement our existing family of algorithms.”

Firefighting scenario to showcase hybrid benefits

Dr McKay says DEWC will transform Claudia’s team’s novel ideas and technology into a prototype that will be tested and demonstrated in a simulated firefighting scenario using AI-enabled drones.

“We are looking at how these novel AI and machine learning approaches could enable a swarm of drones to work as a team to identify the source of the fire and fight the fire by delivering the exact the amount of water, at the right spot, and do it in the quickest possible time,” he says.

“Working closely with the researchers, DEWC Services will take these novel ideas and approaches and, through coding and prototyping, demonstrate their potential in a scenario. This project will serve as a surrogate for Defence projects. The technology has the potential to greatly enhance the resilience of Defence’s C2 and autonomous operations, particularly in denied or degraded environments.”