analiza aiqbench
The paper proposes CEDA, a multi-agent reinforcement learning method (a CTDE Double DQN with a QMIX-style mixing network) for routing drones that deliver medical supplies in disaster zones. Its results come from a simulated grid and PX4 software-in-the-loop tests, not real-world deployment, so the reported delivery rates and fairness scores need independent validation under physical hazards. For ML readers, the main contribution is a reward design that encodes triage priority and discourages starvation of low-priority patients.
A Cross-Layered Multi-Drone Coordination for Medical Supply Delivery during Disaster Response Management
arXiv:2605.09342v2 Announce Type: replace-cross Abstract: Autonomous drone fleets have potential in medical supply delivery during disaster incident response. However, coordinating multiple drones in such settings introduces compounding challenges: dynamic hazards such as wind, obstacles, and intermittent connectivity, constrained energy budgets, and the need to serve patients fairly under strict deadlines and triage-based priority. We present CEDA, a Centralized Training with Decentralized Execution (CTDE) Double Deep Q-Network algorithm augmented with a QMIX-style monotonic mixing network for cooperative multi-drone medical delivery, designed to jointly optimize triage-aware routing, multi-agent coordination, and energy-efficient navigation under dynamic uncertainty. CEDA introduces a Priority-Preserving Fair Scheduling strategy, in which a structured reward function encodes triage priority weights, outcome-level fairness and priority-ordering rewards, and complementary shaping mechanisms that discourage systematic starvation. We evaluate CEDA in a simulated grid environment with five drones serving up to fifty patients, featuring dynamic hazard zones, stochastic action failures, and dynamically spawning patients across three triage levels, and in PX4 software-in-the-loop (SITL) validation using five X500 quadrotors controlled via MAVSDK. Simulation results show that CEDA delivers over 95\% of spawned patients, reduces obstacle collisions by over 95\% across training, and achieves a weighted triage efficiency exceeding 0.90 alongside a Jain's Fairness Index of approximately 1.0 across triage classes. CEDA preserves clinical priority ordering, with Critical patients receiving the fastest response times, while achieving low and comparable mortality across all triage classes. PX4 SITL validation further shows that the learned policy retains its triage-aware behavior under continuous simulated flight and delayed-action conditions.
Abstract:Autonomous drone fleets have potential in medical supply delivery during disaster incident response. However, coordinating multiple drones in such settings introduces compounding challenges: dynamic hazards such as wind, obstacles, and intermittent connectivity, constrained energy budgets, and the need to serve patients fairly under strict deadlines and triage-based priority. We present CEDA, a Centralized Training with Decentralized Execution (CTDE) Double Deep Q-Network algorithm augmented with a QMIX-style monotonic mixing network for cooperative multi-drone medical delivery, designed to jointly optimize triage-aware routing, multi-agent coordination, and energy-efficient navigation under dynamic uncertainty. CEDA introduces a Priority-Preserving Fair Scheduling strategy, in which a structured reward function encodes triage priority weights, outcome-level fairness and priority-ordering rewards, and complementary shaping mechanisms that discourage systematic starvation. We evaluate CEDA in a simulated grid environment with five drones serving up to fifty patients, featuring dynamic hazard zones, stochastic action failures, and dynamically spawning patients across three triage levels, and in PX4 software-in-the-loop (SITL) validation using five X500 quadrotors controlled via MAVSDK. Simulation results show that CEDA delivers over 95\% of spawned patients, reduces obstacle collisions by over 95\% across training, and achieves a weighted triage efficiency exceeding 0.90 alongside a Jain's Fairness Index of approximately 1.0 across triage classes. CEDA preserves clinical priority ordering, with Critical patients receiving the fastest response times, while achieving low and comparable mortality across all triage classes. PX4 SITL validation further shows that the learned policy retains its triage-aware behavior under continuous simulated flight and delayed-action conditions.
Multiagent Systems (cs.MA); Machine Learning (cs.LG)
arXiv:2605.09342 [cs.MA]
(or arXiv:2605.09342v2 [cs.MA] for this version)
https://doi.org/10.48550/arXiv.2605.09342 Focus to learn more arXiv-issued DOI via DataCite
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Text integral sub licența deschisă a publicației · 338 cuvinte