JUDOR:基于 CTDE 的多智能体强化学习框架,实现无人机运动与去中心化机会路由协同优化
JUDOR: CTDE-Based Multi-Agent Reinforcement Learning for the Co-Optimization of Joint UAV Motion and Decentralized Opportunistic Routing in Delay-Tolerant Network
AI 导读
arXiv 论文提出 JUDOR,一种基于 CTDE 的多智能体强化学习框架,联合优化无人机航向控制与 DTN 中的去中心化机会路由。该框架将问题建模为分解的部分可观测马尔可夫决策过程,采用去中心化执行加集中式 critic 训练,并支持多时域拥塞热点辅助监督。
正文 · 原文
Abstract:The growing deployment of delay-tolerant networks (DTNs) has made store-carry-forward (SCF) communication indispensable for operating under sparse intermittent connectivity. However, intermittent contact opportunities, finite node buffers, and limited message time-to-live (TTL) degrade message delivery performance and induce network congestion. Yet unmanned aerial vehicles (UAVs), with extended radio coverage and controllable trajectories, can bridge ground-node pairs that would otherwise suffer poor connectivity. Nevertheless, conventional routing schemes decouple UAV motion control from message replication, and consequently fail to leverage UAV mobility to proactively create future contact opportunities. This paper presents JUDOR (Joint-UAV-Decentralized-Opportunistic-Routing), a centralized training with decentralized execution (CTDE)-based multi-agent reinforcement learning framework optimized for joint discrete UAV heading control and decentralized opportunistic routing. Specifically, we model the tightly coupled UAV motion-routing co-optimization task as a factored partially observable Markov decision process with a shared team reward. JUDOR adopts decentralized actors for local-node execution and a centralized critic used exclusively in training, with optional multi-horizon congestion-hotspot auxiliary supervision. Through evaluation across four distinct traffic modes, JUDOR yields consistent performance gains over both classical and learning-based DTN baselines, including PRoPHET and MaxProp, whilst preserving contact-limited decentralized execution suitable for real-world DTN deployments.
| Subjects: | Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2608.04590 [cs.AI] |
| (or arXiv:2608.04590v2 [cs.AI] for this version) | |
| https://doi.org/10.48550/arXiv.2608.04590 arXiv-issued DOI via DataCite |
Submission history
From: Xiao Wang [view email]
[v1]
Wed, 5 Aug 2026 08:51:41 UTC (642 KB)
[v2]
Wed, 18 Nov 2026 09:05:12 UTC (627 KB)
来源:arXiv cs.AI · arxiv.org