The integration of the Advanced Safe Train Positioning (ASTP) based on GNSS and IMU into the ERTMS on board system requires robust, scalable, and certification-oriented testing methodologies. The challenge is to minimise the tests on field to reduce the validation costs by integrating with simulation methodologies. This hybrid approach has also the advantage to analyse the impacts caused by rare events which are difficult to evaluate with tests on field.
The VICE4RAIL (Hybrid Virtualized Testing for Certification of EGNSS in Railway Train Positioning) project, funded by the European Union Agency for the Space Programme (EUSPA), introduces a pioneering Hybrid Virtualized Certification Environment—HyVICE—designed to accelerate the certification and deployment of GNSS-based train positioning solutions. Central to this initiative is the On-Field/Mixed Reality Testing Platform, implemented at the Bologna San Donato Test Circuit, which serves as a real-world validation facility for innovative railway technologies. The Test Circuit, a 6 km electrified single-track loop managed by Rete Ferroviaria Italiana (RFI), provides a controlled yet operationally realistic setting for experimentation.
This paper focuses on the architecture and operational principles of the On-Field/Mixed Reality Testing Platform. Its architecture integrates a real-time Ground Truth (GT) acquisition using optoelectronic devices, a GNSS Radio Frequency (RF) signal generator and an Inertial Measurement Unit (IMU) sensor package. A Device Under Test (DUT) is installed on board of a test train operating on the San Donato Test Circuit. This configuration allows the DUT to experience authentic train dynamics while receiving synthetic GNSS signals, thereby ensuring consistency between real accelerations and computer-generated positioning data. These elements ensure precise temporal and kinematic alignment between physical motion and simulated satellite signals. The Testing Platform also includes the generation of Augmentation and Integrity Monitoring data as if they were provided by current and future SBASs as well as by local augmentation networks, providing DGNSS and RTK services. A key innovation of the platform is its ability to simulate global and local electromagnetic effects—including multipath, signal blockage, and interference—through advanced modelling and real-time fault injection. This hybrid approach enables performance evaluations from the antenna level to the complete system, supporting both nominal and degraded operational scenarios.
This architecture enables certification-oriented testing aligned with European interoperability standards, validating GNSS and IMU technologies for railway safety-critical applications. Through full-scale field trials, it lays the groundwork for certifying future GNSS-based train positioning in accordance with developing ASTP concepts.
By combining advanced simulation capabilities with on-field mixed reality experiments, the platform establishes itself as a vital contributor for the certification of the ASTP foreseen in the evolution of the ERTMS. Since cost is critical, it not only minimizes the need for extensive on-site testing but also delivers a robust, scalable, and cost-effective solution for manufacturers, system integrators, and certification bodies. This work represents a significant advancement toward harmonized certification methodologies and the widespread adoption of GNSS technologies in railway systems, paving the way for safer, more interoperable, and future-ready rail networks across Europe.

In the aviation domain, GNSS is already widely used for safety-critical applications. A similar adoption in the railway sector could reduce costs compared to current solutions while improving positioning system performance. However, this transition faces challenges due to the wide variety of environments traversed by trains (forests, open skies, urban areas, mountainous regions) and the local errors induced by these environments (multipath, interference, obstructions), which add to global errors (ionosphere, troposphere, clock biases, etc.). This combination makes GNSS performance difficult to predict and limits its integration into safety-critical applications.

Accurate characterization of these errors is therefore an essential step for operational deployment. However, real-world experiments present significant constraints in terms of cost, complexity, and reproducibility. Railway GNSS error characterization remains an issue that we need to solve to effectively demonstrate or assess GNSS-based solution performance. VICE4RAIL intends to address this gap by contributing to the development of a certification methodology including the GNSS-based solution in the ERTMS test-bed.

The HyVICE platform will be developed to cover both local and global GNSS errors. In this paper, we propose to assess the relevance of the use of 3D models and ray-tracing for local error representation, compare to with state-of-the-art studies relying on pure simulation or data-driven models. The approach will address the consistency between a GNSS analysis conducted in a simulated environment and real-world experiments. This study is based on a data acquisition campaign carried out on a railway line in Italy, along with a collaboration with Spirent Communications to generate corresponding GNSS signals in a 3D simulated environment. The objective is to determine the extent to which the errors observed in the field can be faithfully reproduced in simulation and formulate some recommendations for future use of such a technology in the HyVICE chain.

GNSS acquisitions are known to be sensitive to reflections or occlusions in dense urban areas. For train applications, bridges, train stations and high buildings are challenging structures, which create multipath and modifications of the pseudoranges. Navigation solutions must be robust to such events, and deviations need to be monitored during jointly with the estimation of the train state (position and velocity vectors).

R2DATO is an ongoing project funded by Europe’s Rail, focused on the development of digital automation technologies for train operations, ranging from automated to fully autonomous systems. The project aims to establish a new operational paradigm for railways by enhancing safety, flexibility, capacity, and overall performance, while reducing energy consumption and operational costs. Recent progressed have been made to increase the robustness of GNSS receivers: multi-sensor solutions, in the CLUG project, or Digital Maps, in the RAILGAP project.
Many GNSS solutions are map-agnostic. However, precise information about the rail tracks is now largely available (open-access data, Open Street Map, …). A simple solution is the projection of the position from a statistical estimator on the map to retrieve the most probable position of the train. However, the dynamic information is lost, and large errors projected on the map produce sudden jumps on the trajectory.

A more efficient procedure is to include the map information directly in the filtering process to produce coherent state vectors through time. A flexible framework for general filtering purposes is the Particle Filter, which manages complex transitions equations to produce multimodal non-gaussian posterior distributions of the state vector. This article introduces a probabilistic penalization of the particles of the filter with respect to the trajectories of the map. Robust measurements equations are coupled with the penalization inside the update step to modify the posterior distribution and increase the coherence of the estimations with the map, even in difficult situations.

This article presents a methodology for the design of the non-linear and non-Gaussian filtering process, with probabilistic fine-tuning of the parameters (acquisition noise variance and train dynamic). Results of the filters are tested on GNSS railway open access data acquired on the French trajectory and compared with classical algorithms (Weighted Least Square for instance), and naïve projections of the previous estimated states. Results highlight a reduction of number of outlier position estimates (in the case of multipath) on urban scenarios, and divergences of the filter when the map constraint is applied.

To provide economically viable railway transportation in sparsely populated areas, railway systems must evolve toward lower-cost architectures. A major lever for reducing infrastructure costs is the replacement of trackside Train Detection Devices (TDDs) with on-board GNSS-based localization. However, applying aerospace-derived GNSS safety standards directly to railway applications remains challenging.

In a classical GNSS receiver, signals are processed by extracting or estimating satellite clock errors, atmospheric corrections and ephemerides. In contrast, this paper introduces an alternative approach that treats the GNSS ground and satellite segments as non-cooperative systems by leveraging cooperation among receivers.

The proposed method relies on code-based double differencing to eliminate atmospheric effects and both satellite and receiver clock offsets. To recover the geometry matrix, we introduce a new technique: triple differencing between a rover and a base, between satellites, and across two epochs, which normally yields the rover’s displacement vector. This displacement can also be independently obtained using on-board tachometer measurements. The resulting over-constrained system enables estimation of the geometry matrix in the train-forward reference frame. This matrix can then be reused to compute a “probe” value via double differencing. The probe represents the forward component of the baseline between the rover and the base in the train’s reference frame and can be used as an input to a collision-avoidance system or a railway worker protection system. Furthermore, as a result, neither the space segment nor the ground segment of GNSS requires certification, shifting the certification burden entirely to railway-owned GNSS receivers.

In the paper, we will show an evaluation of the accuracy of the method using more than 205 hours of simulated data, as well as in-field railway measurements. Early results show a promising standard deviation of 22 meters.

Accurate track-selective positioning of railway vehicles using onboard sensor data is critical for many applications. The paper addresses the digital railway use case of track condition monitoring and noise mapping in European urban railway or tram networks with in-service vehicles that collect large amounts of microphone (MIC) and axle-box acceleration (ABA) data. These onboard sensors are complemented with affordable GNSS (Global Navigation Satellite Systems) equipment and inertial measurement units (IMU), enabling localization without any track-side positioning infrastructure. Furthermore, the use case allows for offline data processing which can exploit information of the complete vehicle trajectory
to reach a better localization accuracy than in the online case.

The paper presents algorithmic insights and experimental results from the recently concluded OnboardEU project with unified GNSS and IMU data from several vehicles in, for instance, Hannover (D), Graz (A) and Basel (CH). The employed modular processing pipeline is detailed, together with a description of the sensor and map data it utilizes and their roles in the pipeline. Novel practical challenges and implemented solutions regarding the differences between urban railway scenarios and previous work on shunting and mainline railways are explained. The feasibility for large-scale applications and adaptation for other sensor setups is highlighted.