Learning Ad Hoc Network Dynamics via Graph-Structured World Models

2026-04-16

AI Summary

This work addresses the challenge of modeling highly coupled node mobility, energy depletion, and topological dynamics in self-organizing wireless networks—a problem exacerbated by the neglect of structural information in existing approaches. To overcome this, the authors propose G-RSSM, a Graph-structured Recurrent State Space Model that jointly learns network dynamics from offline trajectories using node-level latent states and cross-node multi-head attention. Within a structure-aware world model, an imagination-based rollout mechanism trains a cluster-head selection policy. Notably, this is the first application of a multiphysics graph world model to scale-invariant, node-level combinatorial decision-making. Evaluated across 27 diverse scenarios—including MANETs, VANETs, FANETs, WSNs, and tactical networks with 30–1000 nodes—the policy trained on only 50 nodes consistently maintains high network connectivity.

Technology Category

• Planning, Routing, and Scheduling: Model-Based Reasoning
• Machine Learning: Graph-based Machine Learning
• Intelligent Robots: State Estimation

Application Category

• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications
• Economics, Online Markets and Human Computation: Incentives in network design for Web infrastructures and ecosystems

Abstract

Ad hoc wireless networks exhibit complex, innate and coupled dynamics: node mobility, energy depletion and topology change that are difficult to model analytically. Model-free deep reinforcement learning requires sustained online interaction whereas existing model based approaches use flat state representations that lose per node structure. Therefore we propose G-RSSM, a graph structured recurrent state space model that maintains per node latent states with cross node multi head attention to learn the dynamics jointly from offline trajectories. We apply the proposed method to the downstream task clustering where a cluster head selection policy trains entirely through imagined rollouts in the learned world model. Across 27 evaluation scenarios spanning MANET, VANET, FANET, WSN and tactical networks with N=30 to 1000 nodes, the learned policy maintains high connectivity with only trained for N=50. Herein, we propose the first multi physics graph structured world model applied to combinatorial per node decision making in size agnostic wireless ad hoc networks.

Problem

• ad hoc networks
• network dynamics
• graph-structured models
• combinatorial decision making
• size-agnostic

Innovation

• graph-structured world model
• multi-head attention
• offline reinforcement learning
• size-agnostic networks
• ad hoc network dynamics