MAPLE-RF: Efficient Probabilistic RF Source Localization in Partially Explored Environments

Haozhe Lei; Sundeep Rangan

cs.ROSep 17, 2026

MAPLE-RF: Efficient Probabilistic RF Source Localization in Partially Explored Environments

Authors: Haozhe Lei, Sundeep Rangan

Organizations: NYU WIRELESS, Tandon School of Engineering, New York University, Brooklyn, NY 11201, USA

Abstract

Localizing a radio-frequency (RF) transmitter from received signals often requires a model of the environment to predict how obstacles block and reflect the signal. In many robotic applications, however, only a partial map is available, particularly when a robot localizes the source while exploring with simultaneous localization and mapping (SLAM). We study single-snapshot transmitter localization on such partially explored maps and compare two approaches that output a posterior over transmitter locations. The first extends a digital-twin method, which ray-traces every candidate location, to partial maps by treating unexplored space as free and training on mixed map coverage. The second, MAPLE-RF, encodes estimated path angles of arrival and signal-to-noise ratios as grid channels aligned with map knownness, occupancy, and line-of-sight visibility, and a U-Net scores all candidate positions in one pass without simulating propagation at inference. Ray-tracing simulations of indoor environments indicate that training on mixed map coverage is essential for both approaches. The digital-twin approach is more accurate on most single-snapshot metrics, while MAPLE-RF comes close at a query cost that does not depend on the propagation model and is more than two orders of magnitude below a fresh full-grid query with general-purpose ray tracing. Both outperform Gaussian and Gaussian-mixture baselines, and on exploration routes guided by its own estimates, fused MAPLE-RF posteriors place more probability near the source than the compared methods. Code and data will be released.

Figures & tables

Fig. 1: Illustration of MAPLE-RF for single-snapshot RF source localization with a partial map. One receiver (RX) provides its known pose and multipath angle-of-arrival (AoA) and signal-to-noise ratio (SNR) estimates. Map/visibility, geometry, and RF features are concatenated as $\mathbf{Z}$ . A residual U-Net scores all candidate transmitter (TX) positions in one pass; normalization yields the position posterior. The non-line-of-sight (NLOS) example uses Spatial supervision and Robust training. Gray marks unexplored space; green/orange indicate confirmed line-of-sight (LOS)/NLOS. Solid orange and dashed cyan contours outline observed and true obstacles. In-phase and quadrature (IQ) traces and feature icons are schematic. MAP denotes maximum a posteriori.

Fig. 2: Map and visibility information for a common RX position. (a) Three exploration routes in the same physical scene end at the RX; propagation paths are obtained by ray tracing (RT). Mi/Mo/Se denote Mild/Moderate/Severe route starts. (b) Knownness and occupancy grids. (c) Confirmed LOS/NLOS and unknown visibility. The coverage levels follow the definition in Section IV-B .

Fig. 3: Single-snapshot localization examples across map coverage. The first two columns show the available map and derived visibility; the remaining columns show estimated location probabilities on a common logarithmic color scale. Stars and crosses indicate the true TX and MAP estimates. Solid orange and dashed cyan contours outline observed and true obstacles. H/S denote Hard/Spatial supervision, and C/R denote Clean/Robust training.

Fig. 4: Single-snapshot performance across map coverage. Lines show means; shaded bands show 95% confidence intervals from 2,000 layout-family bootstrap resamples. Each layout–map unit averages 48 RF snapshots. Clean uses 240 units; each incomplete-map level uses its own 48-layout subset. H/S denote Hard/Spatial supervision; solid/dashed lines denote Robust/Clean training.

Quantity	Train	Val.	Test	Total
Layouts	1,920	240	240	2,400
Partial maps	1,272	96	144	1,512
RF–map examples	153,216	16,128	18,432	187,776

TABLE I: Dataset split and available RF–map examples

		Full distribution				MAP summaries		Cost
Model	Train	Mass NLL [nat] $\downarrow$	ES [m] $\downarrow$	Expected distance [m] $\downarrow$	1 m mass [%] $\uparrow$	Error [m] $\downarrow$	1 m recall [%] $\uparrow$	Query time [ms] $\downarrow$
LOCUS-DT	C	$5.26\pm 1.19$	$1.15\pm 0.52$	$2.21\pm 0.77$	$42.6\pm 15.9$	$1.75\pm 0.81$	$54.7\pm 15.6$	$3116.0\pm 403.5$
LOCUS-DT	R	$\mathbf{4.01\pm 0.88}$	$\mathbf{0.78\pm 0.35}$	$1.68\pm 0.63$	$\mathbf{57.0\pm 13.0}$	$1.32\pm 0.73$	$\mathbf{66.4\pm 14.7}$	$3116.0\pm 403.5$
LOCUS-DT $25^{2}$	R	$4.55\pm 0.75$	$0.85\pm 0.34$	$1.86\pm 0.61$	$52.2\pm 12.3$	$1.36\pm 0.69$	$65.4\pm 14.3$	$902.1\pm 111.7$
LOCUS-DT $13^{2}$	R	$5.45\pm 0.57$	$1.04\pm 0.32$	$2.26\pm 0.56$	$40.1\pm 9.9$	$1.60\pm 0.66$	$54.1\pm 13.6$	$353.9\pm 75.3$
SG	C	$7.52\pm 0.13$	$2.30\pm 0.18$	$4.59\pm 0.21$	$4.0\pm 0.5$	$3.86\pm 0.46$	$6.4\pm 3.5$	$11.6\pm 5.6$

TABLE II: Single-snapshot distribution quality, MAP localization summaries, and query cost on partially explored maps

Fig. 5: Visibility ablation of MAPLE-RF-S, using the same Mild example as Fig. 3 . Solid orange and dashed cyan contours outline observed and true obstacles. Clean mass uses all Clean test maps; other summaries use the partial-map units of Table II . Green marks MAPLE-RF-S (R).

Fig. 6: Current and accumulated MAPLE-RF-S (R) posteriors along one exploration trajectory. The current-snapshot MAP selects the route. Regional masses are compared at identical RX poses; the background transition marks the first true LOS observation. The 1 m disk is centered on the continuous TX location.

Fig. 7: Accumulated regional probability mass for five base models with Robust (R) training on 100 shared, initially NLOS sequences across 34 held-out test layouts with four obstacles each. Each sequence contains 16 observations. Lines show equally weighted sequence means; shading gives pointwise 95% confidence intervals from 2,000 shape-stratified layout-cluster bootstrap resamples. H/S denote Hard/Spatial supervision. The cone has a

30^{\circ}

full angle; its axis starts at 20%.

Explore similar work

RadioTrace: Transmitter-Aware Diffusion for Radio Map Estimation without Deployment-Time Fine-Tuning
Jul 23, 2026Liu Yang, Qiang Li, Zhuo Cao +3Wireless Communications Diffusion Posterior Sampling
School of Information and Communication Engineering, University of Electronic Science and Technology of China, Chengdu, China, 611731 · School of Information and Communication Engineering, University of Electronic Science and Technology of China, Chengdu, China, 611731, and also with Tianfu Jiangxi Laboratory, Chengdu, China, 641419
SymNet: A Multi-Task Network for Joint Radio Map Reconstruction and Transmitter Localization
Jul 30, 2026Lyuzhou Ye, Thanh Dat Le, Yan HuangMulti-Task Learning Wireless Communications
University of North Texas 1155 Union Circle Denton, Texas
PropSplat: Map-Free RF Field Reconstruction via 3D Gaussian Propagation Splatting
May 8, 2026William Bjorndahl, Maninder Pal Singh, Farhad Nouri +13D Gaussian Splatting Wireless Communications
Department of Electrical and Computer Engineering, Southern Methodist University, Dallas, TX, USA · Department of Engineering Technology, University of Houston, Houston, TX, USA

Jul 23, 2026J/K move · Enter open · S save

RadioTrace: Transmitter-Aware Diffusion for Radio Map Estimation without Deployment-Time Fine-Tuning

Liu Yang, Qiang Li, Zhuo Cao, Weijie Xiong+2

School of Information and Communication Engineering, University of Electronic Science and Technology of China, Chengdu, China, 611731 · School of Information and Communication Engineering, University of Electronic Science and Technology of China, Chengdu, China, 611731, and also with Tianfu Jiangxi Laboratory, Chengdu, China, 641419

Wireless Communications Diffusion Posterior Sampling

Open paper AskPDF arXiv

Radio map (RM) estimation aims to reconstruct the spatial distribution of wireless signal characteristics, such as received signal strength (RSS), from sparse measurements, a task that is critical for spectrum management, interference mitigation, and localization in modern wireless networks. Traditional approaches, including interpolation and deep learning, either struggle to capture complex propagation effects or require large-scale retraining for each new sampling pattern, which limits their generalization. More recently, prior-based methods have combined pre-trained generative models with measurements to reduce the need for deployment-time model fine-tuning, but they typically treat the prior as a simple regularizer and lack explicit transmitter-aware integration. In this paper, we propose RadioTrace, a novel RM estimation framework without deployment-time fine-tuning that tightly integrates sparse RSS measurements with a frozen pre-trained diffusion prior. RadioTrace incorporates transmitter (Tx) location estimation directly into the denoising loop, iteratively refining Tx coordinates based on reconstruction quality to guide the generative process. To further enhance robustness, we introduce a propagation-guided K-means initialization that mitigates poor local minima in the Tx update and provides a geometry-consistent starting point. Moreover, we provide a stochastic stability analysis for the Tx-coordinate refinement component, showing that the Tx update remains stable under perturbations induced by diffusion sampling and Tx-map relaxation. Extensive experiments demonstrate that RadioTrace achieves competitive performance with state-of-the-art learning-based methods under random sampling, and maintains strong reconstruction quality under restricted-area sampling, highlighting its adaptability, robustness, and practical relevance.

Abstract

Explore similar work

RadioTrace: Transmitter-Aware Diffusion for Radio Map Estimation without Deployment-Time Fine-Tuning

SymNet: A Multi-Task Network for Joint Radio Map Reconstruction and Transmitter Localization

PropSplat: Map-Free RF Field Reconstruction via 3D Gaussian Propagation Splatting