LocDreamer: World Model-Based Learning for Joint Indoor Tracking and Anchor Scheduling

TL;DR

LocDreamer addresses indoor localization and tracking under constrained anchor usage by learning a world-model-based latent dynamics representation and an RL-based anchor scheduler. It combines a Deep State Space Model with imagination-driven training to generate synthetic measurements for unseen anchor configurations, optimizing the marginal likelihood $\log p_{\theta}(\mathbf{d}_{1:T} | \boldsymbol{\alpha}_{1:T})$ under the constraint $\sum_k \alpha_t^k = K_t$. The approach yields a two-stage learning process: pre-train the DSSM on real data, then train with imagined measurements to jointly optimize tracking and scheduling, with an actor-critic scheduler guided by reconstruction-based rewards. On a real indoor UWB dataset, LocDreamer outperforms a model-based tracker with random scheduling by 37% in tracking accuracy and reaches 86% of the accuracy of the same model trained directly on real data, demonstrating strong data efficiency and cross-environment generalization.

Abstract

Accurate, resource-efficient localization and tracking enables numerous location-aware services in next-generation wireless networks. However, existing machine learning-based methods often require large labeled datasets while overlooking spectrum and energy efficiencies. To fill this gap, we propose LocDreamer, a world model (WM)-based framework for joint target tracking and scheduling of localization anchors. LocDreamer learns a WM that captures the latent representation of the target motion and localization environment, thereby generating synthetic measurements to imagine arbitrary anchor deployments. These measurements enable imagination-driven training of both the tracking model and the reinforcement learning (RL)-based anchor scheduler that activates only the most informative anchors, which significantly reduce energy and signaling costs while preserving high tracking accuracy. Experiments on a real-world indoor dataset demonstrate that LocDreamer substantially improves data efficiency and generalization, outperforming conventional Bayesian filter with random scheduling by 37% in tracking accuracy, and achieving 86% of the accuracy of same model trained directly on real data.

Figures (5)

• Figure 1: Illustration of the proposed LocDreamer.
• Figure 2: Graphical model of LocDreamer. Diamond and circle represent deterministic and stochastic variables. Arrows indicate conditional dependencies. Solid and dashed lines represent the DSSM modeling the tracking system and the scheduling decision making. Green, red and blue highlight DSSM, observation and anchor scheduler components, respectively.
• Figure 3: Estimated trajectories with different methods.
• Figure 4: An anchor scheduling heatmap over spatial locations where different color represents different selected anchor sets $\mathcal{K}_t$.
• Figure 5: Validation loss and test MAE of LocDreamer - scheduling versus imagination epochs $E_\text{imagine}$.