JUDOR trains UAVs and routing together for delay-tolerant networks
A research team describes a multi-agent reinforcement learning framework that jointly controls UAV movement and decentralized message routing, reporting gains over established DTN baselines in simulation.
In plain words
Delay-tolerant networks move data through disconnected nodes that store messages and pass them along when contact happens. JUDOR is a training method in which drones learn both where to fly and how to forward messages, so their movement helps create future contact opportunities. The drones act independently at run time; the coordinating trainer is used only during learning.
What was announced
The authors frame the problem as a factored partially observable Markov decision process with a shared team reward, capturing how UAV trajectories and message replication decisions affect each other. During execution, each node runs a decentralized actor, while a centralized critic guides training only; the design also includes optional multi-horizon congestion-hotspot auxiliary supervision. According to the paper, this addresses a weakness of conventional schemes that decouple UAV motion control from message replication and so cannot use mobility to proactively create contact opportunities.
Evaluation across four distinct traffic modes reportedly shows consistent performance gains over classical and learning-based DTN baselines, including PRoPHET and MaxProp, while keeping contact-limited decentralized execution that the authors consider suitable for real-world deployments. The reported results come from the paper's own evaluation; the abstract does not state whether physical UAV trials were conducted, and the work is presented as a version-2 replacement on arXiv rather than a peer-reviewed publication.
What it does not show
- Evidence is limited to the paper's abstract, so the reported gains over PRoPHET, MaxProp and other baselines cannot be independently verified and no quantitative results are available.
- The abstract does not describe physical flight trials or real-world deployment results, only evaluation across four traffic modes.
For security and IT teams
- Scope
- The framework is designed to preserve contact-limited decentralized execution, meaning running nodes need no centralized coordinator, which the authors present as relevant to real-world DTN deployments.
Sources
Source-support review completed 20 Nov 2026. This story was drafted with AI assistance and checked against retained source passages. Automated checking is not independent verification.
Coverage: arXiv: abstract.
Updates and corrections
No corrections so far.
First published 20 Nov 2026
New to the subject? AI from the Ground Up starts from the beginning.