Organizations: The authors are with the Key Laboratory of Intelligent Perception and Human-Machine Collaboration - ShanghaiTech University, Ministry of Education, China.
WiFi localization offers a lightweight global prior for robots in GPS-denied environments, but fingerprinting requires dense surveys and generalizes poorly beyond surveyed areas. We model received signal strength indicator (RSSI) measurements using OpenStreetMap Area Graph (osmAG) structural priors. Offline, known anchor access points (APs) support sequential calibration of line-of-sight (LOS) propagation and wall attenuation; unknown AP locations are then estimated by alternating ray-wall intersection checks, signal compensation, and least-squares fitting. The resulting AP map supports online robot position estimation using the same structure-aware model. In experiments in an 11,025m2 multi-floor environment, our method reduces the mean AP localization error from 5.86m to 3.79m. Mean robot localization errors are 3.12m in fingerprinted areas and 3.83m in unfingerprinted areas, where four baselines yield 11.00--20.21m. The WiFi-related map representation achieves approximately 251:1 compression relative to the raw fingerprint database. These results support structural modeling for compact global localization beyond surveyed regions. Code and data: https://github.com/XuMa369/osmag-wifi-localization
Figures & tables
Fig. 1: Beyond surveyed areas. Top: fingerprinting depends on historical scans and can fail outside their coverage. Bottom: osmAG wall intersections (green nodes) determine attenuation along signal rays, enabling structure-aware localization.
Fig. 2: Structure-aware WiFi localization. Offline calibration learns line-of-sight (LOS) and wall-loss parameters from anchor APs, then maps unknown APs. Online localization combines averaged RSSI with the augmented map. Both solvers alternate ray casting, virtual-LOS compensation, and least-squares fitting; Algorithm 1 and Section III-C specify the updates.
Fig. 3: Augmented osmAG map: grey structures represent walls and rooms; orange nodes represent estimated APs. Semantic tags associate each AP’s coordinates with its BSSID and floor level.
Fig. 4: Single-AP refinement. (a) LOS-only initialization. (b) Estimate after ten wall-aware iterations. Blue points are robot positions; the green star is AP ground truth. The red box marks the final-estimate zoom.
Fig. 5: Experimental osmAG layouts across Buildings A and B: (a) Floor 1; (b) Floor 2. Red RP markers indicate robot positions, and blue AP markers indicate surveyed AP locations.
Method
Fingerprinted Areas (In-Distribution)
Unfingerprinted Areas (Out-of-Distribution, N=7)
Mean
Std.
RMSE
P95
Mean
Std.
RMSE
P95
(m)
(m)
(m)
(m)
(m)
(m)
(m)
(m)
KNN Fingerprinting
3.42
2.50
4.24
7.85
20.21
9.48
22.32
36.18
GP-PL [ 4 ]
3.18
2.81
4.24
6.50
13.70
11.50
17.89
22.90
GPoE [ 5 ]
3.12
1.77
3.59
6.63
14.71
2.67
14.95
18.72
IRPLS-ABCL [ 3 ]
3.05
1.88
3.59
6.63
11.00
2.41
11.26
13.65
TABLE I: Quantitative Comparison of Robot Localization Performance Across Different Scenarios.
Fig. 6: Per-AP errors for 49 non-anchor APs: geometry-unaware trilateration versus structure-aware refinement.
Test set
Mean (ms)
P95 (ms)
Max. (ms)
Fingerprinted Areas
276.9
341.7
394.6
Unfingerprinted Areas
289.9
370.6
381.5
All positions
278.2
343.9
394.6
TABLE II: Per-query solver computation time over five repeated passes.
Representation
Points
Size (KB)
Per-Unit
Fingerprint DB
1,491 scans
11,068
7.42 KB/scan
osmAG + AP (Ours)
49 APs
44
< 1 KB/AP
TABLE III: Storage Efficiency Comparison of WiFi Map Representations.
Fig. 7: Robot test positions across two floors: 66 fingerprinted positions (red) and 7 unfingerprinted positions (blue), including cross-building corridors and atriums.
Indoor navigation to task-relevant targets remains difficult when the target is small, occluded, or visually indistinct. Meanwhile, smart appliances, mobile devices, and IoT tags emit Wi-Fi signals during everyday communication, providing a pervasive but underused spatial cue. We study visual-Wi-Fi navigation, where a robot reaches a signal-emitting device in an unmapped environment using onboard visual observations and Wi-Fi packets overheard from the device, without connecting to the device or modifying infrastructure. Realizing this requires pose-dependent Wi-Fi observations during closed-loop policy learning and reliable spatial cues from bearing measurements corrupted by indoor multipath propagation on a moving robot. We present WiFiSpaces, which extends the Habitat embodied simulator with online ray-traced Wi-Fi sensing, rendering Wi-Fi observations for a robot-mounted antenna array at continuous poses during policy rollouts. On top of it, we build ViFiNav, a localization-aware navigation framework: an online source-localization algorithm exploits robot motion to disambiguate and robustly aggregate bearings from multiple viewpoints into a confidence-aware source estimate, and a learned policy fuses this evolving estimate with depth observations and learns to act on it as it is refined. Experiments show that exploiting robot motion is key to accurate source estimates, and that fine-tuning on ray-traced Wi-Fi observations in WiFiSpaces improves the success rate of ViFiNav. The policy transfers zero-shot to unseen scene datasets, and we deploy it on a real robot with a Wi-Fi sensor built from a commodity router, without any real-world training.
WiFi fingerprint-based indoor localization has been widely studied, but most existing approaches focus on absolute positioning and rely on dense coordinate annotations, which are costly to obtain at scale. In this paper, we study a fundamentally different problem: relative localization, where the goal is to directly estimate the displacement between two WiFi fingerprint traces without predicting their absolute positions. To reduce annotation overhead, we adopt weak supervision in the form of stepwise motion vectors obtained from inertial sensing. We propose Intersection Pathway (IP), a cross-modal learning framework that aligns fingerprint traces (f-traces) and displacement traces (d-traces) in a shared latent space. The key idea is to enforce an additive structure in the latent space, such that latent addition and subtraction correspond to physical motion composition, enabling direct relative-displacement inference. Experiments on a synthesized dataset derived from real measurements demonstrate that the proposed method learns displacement-aware WiFi representations and achieves accurate relative localization across varying displacement ranges. Furthermore, the learned model can be extended to few-shot absolute localization with sparse anchors.
Tzu-Ti Wei, Po-Cheng Chen, Yu-Chee Tseng +1
College of AI, National Yang Ming Chiao Tung University, Taiwan
Deep learning models can achieve high accuracy for indoor localization, but their black-box nature limits interpretability and the reuse of learned information. We propose a hierarchical deep learning framework for WiFi fingerprint-based indoor localization that jointly predicts user location and learns an effective geometry of the surrounding access points (APs). Physics-informed decoders infer this geometry directly from RSSI measurements and labelled user positions, without requiring the true AP coordinates during training. The learned geometry is then used to rank and prune APs. On the UJIIndoorLoc dataset, the proposed chained model achieves a mean 3D localization error of 7.07 m, reducing error by 26% to 36% compared with baseline models. Previously published methods evaluated on the same official split report errors 10.6% to 31.0% higher. Pruning 35% or 50% of the APs causes only a small loss in localization accuracy. The inferred geometry also enables Fisher-information-based AP ranking even when fingerprint databases do not contain surveyed AP coordinates. Experiments on the Tampere/TUT and UTSIndoorLoc datasets show that geometry-guided AP selection performs comparably to selectors built directly from labelled data. These results show that physics-informed interpretability can improve indoor localization while also supporting effective feature selection.
Arshia Eftekhari zadeh, Rezvan Nasiri, Hadi Moradi
School of Electrical and Computer Engineering University of Tehran Tehran, Iran · Advanced Robotics and Intelligent Systems Laboratory School of Electrical and Computer Engineering University of Tehran Tehran, Iran