{"id":19858,"date":"2025-12-18T11:24:14","date_gmt":"2025-12-18T10:24:14","guid":{"rendered":"https:\/\/www.uni.lu\/snt-en\/?post_type=news&p=19858"},"modified":"2025-12-18T11:24:17","modified_gmt":"2025-12-18T10:24:17","slug":"collaborative-drone-swarms-for-airspace-protection","status":"publish","type":"news","link":"https:\/\/www.uni.lu\/snt-en\/news\/collaborative-drone-swarms-for-airspace-protection\/","title":{"rendered":"Collaborative drone swarms for airspace protection"},"content":{"rendered":"\n
\n

Rogue\u00a0drones\u00a0are bringing\u00a0chaos to airspace.\u00a0In a recent case in Belgium,\u00a0air traffic controllers scrambled to respond as\u00a0an\u00a0unauthorised\u00a0aircraft\u00a0threatened commercial flights and emergency services.<\/a> The incident\u00a0exposed\u00a0a serious problem: it only takes one drone to disrupt an entire system.<\/strong>\u00a0<\/p>\n\n\n\n

But what happens when dozens or even hundreds of drones work together?\u00a0<\/p>\n\n\n

\n \n\"a

\n PCOG drone swarm on the floor in their lab <\/p>\n <\/figure>\n\n\n

Researchers at the Parallel Computing and\u00a0Optimisation\u00a0Group (PCOG)<\/a> are tackling this exact challenge. As drone\u00a0intrusions\u00a0become more sophisticated, the team is developing artificial intelligence\u00a0methods\u00a0that\u00a0enable\u00a0coordinated\u00a0drone swarms to\u00a0respond\u00a0to complex\u00a0aerial threats.\u00a0They teach drones how to work as a team\u00a0through\u00a0advancing\u00a0multi-objective\u00a0reinforcement learning\u00a0(MORL)\u00a0by\u00a0drawing on concepts from\u00a0multi-objective\u00a0optimisation.\u00a0The goal is to\u00a0support\u00a0systems that can\u00a0assist\u00a0human operators in\u00a0intercepting,\u00a0containing, and escorting\u00a0hostile drones away from sensitive areas.\u00a0<\/p>\n\n\n\n

\nTraining drone swarms to act as a team <\/h2>\n\n\n\n

Current drone defense systems focus on stopping a single intruder. Once the drone appears, the system tracks it, identifies it, and neutralises it. But coordinated drone swarms are different. Multiple drones can split up, regroup, and change their strategy in real time. Stopping them requires defenders that can do the same. <\/p>\n\n\n\n

This is where reinforcement learning helps.\u00a0Reinforcement learning\u00a0trains\u00a0decision-making models\u00a0by letting them try actions\u00a0and\u00a0learn from the results in\u00a0a simulated environment.\u00a0\u00a0“By applying this\u00a0approach\u00a0to drone\u00a0swarms,\u00a0our\u00a0team\u00a0improves\u00a0swarm\u00a0coordination\u00a0by running\u00a0large numbers\u00a0of simulated missions,” said\u00a0Dr. Gr\u00e9goire\u00a0Danoy,<\/a> head of the PCOG research group.\u00a0\u00a0Every\u00a0simulation\u00a0provides\u00a0feedback on what worked\u00a0and\u00a0what\u00a0didn’t,\u00a0allowing\u00a0swarm\u00a0behaviours\u00a0to be refined\u00a0before they are used in real-world experiments.\u00a0<\/p>\n\n\n\n

The technology works like training a dog, but here it\u2019s applied to drones at massive scale and entirely in simulation. In early test runs, drones tend to cluster together and leave gaps in coverage. The AI gets a low performance score. As training continues, feedback helps drones move and work together more efficiently, which results in a better score for the AI. After millions of simulated practice runs, the swarm learns to maintain good coordination, adapt to failure, and respond to threats without the drones getting in each other’s way. <\/p>\n\n\n\n

The challenge grows when drones need to balance competing goals at once. A surveillance drone might need to\u00a0maximise\u00a0area\u00a0coverage\u00a0while conserving battery\u00a0power\u00a0and staying\u00a0in contact\u00a0with the swarm.\u00a0Traditional reinforcement learning often\u00a0turns all goals into one score, which makes it harder for the system to balance goals that conflict with each other.\u00a0 Within\u00a0the\u00a0PCOG research\u00a0team,\u00a0Dr. Florian\u00a0Felten\u00a0explored how combining\u00a0multi-objective\u00a0optimisation<\/a><\/strong>\u00a0with\u00a0reinforcement learning<\/strong>\u00a0<\/a>could help drone swarms handle the competing needs that arise during flight.\u00a0<\/p>\n\n\n\n

\nHow one PhD project redefined what drone swarms can learn <\/h2>\n\n\n\n

Felten carried out his doctoral research within the FNR-funded ADARS project (Automated Design of Autonomous Robot Swarms) led by Danoy, to explore how artificial intelligence can support the automated design and coordination of drone swarms. <\/p>\n\n\n\n

Felten focused on a central challenge: how could drone swarms learn to cooperate more effectively while balancing competing objectives?  <\/p>\n\n\n\n

His work improved multi-objective reinforcement learning by helping AI systems think more clearly about trade-offs like safety, energy use, and mission performance. Instead of condensing everything into a single score, Felten\u2019s approach helped learning systems understand how choices affect different goals. <\/p>\n\n\n\n

This work led to\u00a0several\u00a0high-impact scientific publications,\u00a0including\u00a0papers\u00a0in the Journal of Artificial Intelligence Research<\/a> and\u00a0NeurIPS<\/a>, and\u00a0to\u00a0the development of\u00a0an open-source simulation\u00a0environment for MORL\u00a0used worldwide.\u00a0<\/p>\n\n\n\n

In October 2025,\u00a0he\u00a0won\u00a0the Outstanding PhD Thesis Award<\/a>\u00a0from\u00a0the Luxembourg National Research Fund (FNR).<\/a>\u00a0\u00a0<\/p>\n\n\n\n

\n
\n
\n
\n \u201f<\/span>\n While large language models currently receive much public attention, reinforcement learning plays a crucial role when AI systems must\u00a0<\/em>control real robots and help automate complex\u00a0systems\u201d\n<\/blockquote>\n
\n \n\"\"\n <\/figure><\/div>\n\n
\n

Dr. Florian Felten<\/p>\n\n\n

Former SnT Doctoral Candidate, now at ETH Zurich<\/p>\n\n

\n \n \n View profile<\/span>\n <\/span>\n <\/use><\/svg> <\/a>\n<\/div>\n<\/div>\n <\/figure>\n <\/div>\n<\/section>\n\n
\n
\n
\n
\n