Accurate and scalable indoor pedestrian positioning remains challenging due to the absence of Global Navigation Satellite System (GNSS) signals and the complex spatial variability of indoor radio propagation. Wi-Fi fingerprinting is a widely adopted solution, estimating user’s position by matching received signal strength indicator (RSSI) measurements to a pre-recorded database of signal signatures. However, the construction of high-quality radio maps is limited by two fundamental factors: the lack of reliable ground-truth trajectories during data collection and the low temporal resolution of Wi-Fi scanning on consumer smartphones imposed by power management and operating system constraints.

Smartphone-based site surveys are therefore labor-intensive and result in sparse spatial sampling, with meter-scale spacing between consecutive fingerprints at typical walking speeds. This undersampling limits the ability to capture localized signal variations near doors, corridor transitions, and regions with strong RSSI gradients. Crowdsourced mapping approaches further suffer from inaccurate position labels derived from consumer-grade inertial sensors with rapidly accumulating drift, leading to degraded radio map quality.

This work presents a hybrid indoor mapping framework that decouples high-accuracy trajectory generation from scalable end-user localization. During an offline mapping phase, both pedestrian dead reckoning and Wi-Fi fingerprint collection are performed on a foot-mounted platform integrating an inertial measurement unit (IMU) and dual-band Wi-Fi hardware. The foot-mounted IMU exploits gait-induced zero-velocity updates (ZUPT) to generate drift-bounded pedestrian trajectories, while time-synchronized Wi-Fi measurements are collected along the same path. This synchronization enables automatic assignment of accurate position labels to each fingerprint without manual surveying. Continuous high-rate scanning yields spatially dense radio maps that better capture fine-grained signal variations compared to smartphone-based collection.

Dual-frequency Wi-Fi measurements in the 2.4 GHz and 5 GHz bands are modeled using Gaussian Process regression to account for spatial correlation and measurement uncertainty. In the online phase, arbitrary consumer smartphones perform localization using only ambient Wi-Fi observations and the pre-constructed radio map, without requiring specialized hardware or inertial sensing. Experiments conducted in a real-world office environment with multiple users and heterogeneous devices demonstrate meter-level positioning accuracy and consistent performance improvements over single-band Wi-Fi fingerprinting.

Most indoor position fixing methods rely on active radio frequency-emitting infrastructure such as WiFi access points, Bluetooth low energy beacons, or ultra-wideband anchors. In this paper, we present a novel and robust position fixing method which detects printed fiducial markers with known coordinates in 3D Light Detection and Ranging (LiDAR) point clouds. The method does not require active infrastructure, only an active sensor and power supply at the mobile robot. It is therefore well-suited for position fixing in challenging environments, as it supports navigation in the dark and in settings where powered infrastructure is unavailable.
Our method processes point clouds from a low-cost, non-repetitive 360° horizontal field of view LiDAR sensor, which are accumulated during short stationary phases of a mobile robot. The resulting point clouds are then converted to spherical images where the intensity of each individual LiDAR point is mapped as the gray value of the corresponding pixel. We apply image processing techniques and a custom kernel operation designed to fill the sparsely populated spherical image. Since the intensity values vary with the distance to the target and the angle of reflection of the emitted light pulse, we propose a robust processing chain to reliably detect the AprilTag markers in challenging conditions. The robot's position is then estimated using a weighted non-linear least-squares adjustment where the ranges to each detected marker corner represent the individual measurements.
We evaluate our approach in two steps using indoor test datasets with ground truth measurements from a camera tracking system. First, we determine the accuracy of individual position fixes with a varying number of tags in the robot's surroundings. In a second analysis, we show how the method can improve a LiDAR-based SLAM solution of the robot when it enters an environment with fiducial markers.

The deployment of Smart Road infrastructures relies on a dense network of RSUs (Roadside Units) to enable advanced ITS (Intelligent Transport Systems) services to enhance road safety by means communication-based approach for smart traffic management.
These RSUs are currently equipped with communication systems and perception sensors, as well as commercial GNSS receivers to support the timing requirements of the underlying telecommunications network. Due to the limited communication range of RSUs, their mutual spacing is generally small, resulting in an inherently dense and spatially distributed GNSS receiver network. This feature naturally enables the possibility of leveraging this dense network of stations to implement advanced GNSS augmentation services, ensuring high-accuracy and integrity localization for vehicles and road users. Furthermore, a potential evolution of this approach is the possibility of adopting multi–base station GNSS augmentation techniques, such as multibaseline PPP-RTK, which are specifically designed to leverage dense reference networks to enhance positioning performance.
While multibaseline PPP-RTK algorithms are well established in the literature, their applicability and benefits within Smart Road scenarios have not yet been thoroughly investigated.
The objective of this paper is to provide a comprehensive performance evaluation of a multibaseline PPP-RTK framework when the augmentation network is formed by GNSS receivers installed on Smart Road RSUs. In particular, the study focuses on the performance analysis of Smart Road architectures supporting multi-baseline PPP-RTK service, evaluating accuracy and integrity across different network topologies and varying inter-RSU distances. Furthermore, the impact of using a subset of RSUs for integrity monitoring of the reference network supporting PVT (position, velocity, and time) estimation is analyzed. From a system engineering perspective, the study also aims to investigate trade-offs between RSU density, performance improvements and costs associated with deploying and operating infrastructure-based augmentation services for smart road applications.
The analysis is conducted entirely in a simulated way through the VIRGILIO software, developed by Radiolabs which enables realistic GNSS PVT performance assessment in multi-sensor approach, and relevant GNSS data, acquired in real operational environments or synthetically generated to simulate complex GNSS scenarios. The results demonstrate that implementing a multibaseline PPP-RTK service over a Smart Road network significantly enhances both the accuracy and integrity of the PVT estimated by onboard systems, while providing quantitative insights to achieve an optimal trade-off between network density, system performance, and deployment and maintenance costs.

Reliable positioning in Global Navigation Satellite System (GNSS)-denied urban environments remains a critical capability gap for dismounted personnel and autonomous systems. However, most visual positioning systems require the query and database images to be taken from similar viewpoints. For military, emergency, and security applications, there is often a need to match forward-looking query images with a satellite-based database or a map. Using semantic feature descriptors, such as "building," "road," and "house," enables this cross-view image matching.

This paper proposes Angular Semantic Positioning, a method that extracts explicit angular bearings to static urban infrastructure from a single monocular image and compares them against lightweight 2D vector maps (e.g., OpenStreetMap). We employ a real-time semantic segmentation network to distinguish individual building instances from continuous urban blocks, enabling precise centroid-based bearing measurements. To handle the ambiguity of single-epoch matching, we enhance positioning accuracy by enforcing geometric depth-ordering constraints. For example, we use the physical prior that line-of-sight rays can traverse planar "road" surfaces but are obstructed by vertical "building" structures, allowing the algorithm to score position candidates based on free-space visibility.

This method offers two main advantages for real-time operation: (1) Robustness: By enforcing geometric consistency between the visual scene and the vector map, the system reduces the impact of transient segmentation errors; and (2) Efficiency: A forward-looking query image can be matched to a map in 2–6 ms on consumer hardware, minimizing Size, Weight, and Power (SWaP) requirements. This is significantly faster than emerging end-to-end machine learning approaches (e.g., Sarlin et al., 2023).

Experimental validation using query images from the Mapillary Vistas dataset (Neuhold et al., 2017) and map data from OpenStreetMap (Haklay & Weber, 2008) covering London and the surrounding area demonstrates that the algorithm successfully isolates the true position within the top 10% of candidates in 77% of trials. This effectively reduces the search space by an order of magnitude using only a single monocular frame, validating its suitability for resilient Positioning, Navigation, and Timing (PNT) in resource-constrained environments. By combining multiple image matches as the user moves, we would expect to obtain a unique position solution.