In recent years, a various research projects and innovation activities such as [CLUG2], R2Dato] and [HELMET] have been undertaken with the aim of developing new concepts to enable global navigation satellite systems (GNSS) for high-accuracy and high-integrity railway operations. These concepts are considered to be mandatory for future railway signaling and control in Europe. All of the concepts employ GNSS in conjunction with various sets of other sensors, including inertial measurement units (IMU), odometers, radar, lidar, and digital map information. Additionally, augmentation services such as satellite-based augmentation systems (SBAS), local augmentation, and real-time kinematic (RTK) have been considered. Nevertheless, the selection of the optimal future solution is a complex undertaking. One challenge arises in the meaningful comparison of the various results. In [Grosch2025, RailgapD4.8], we presented a forward-backward smoothed RTK/INS solution with float ambiguities, which demonstrated the potential to achieve high-accuracy and high-integrity train trajectory determination capabilities enabling the required comparison. This solution utilizes GNSS, RTK and IMU only. The additional usage of the digital track map might increase the accuracy of the trajectory determination, however, ensuring the integrity of the final solution might be not possible. This is due to the fact that the question of how the quality and integrity of the map information will be ensured in the future system remains unanswered.

In this paper, we propose an enhancement to our methodology by incorporating ambiguity fixing techniques into the forward-backward smoothed RTK/INS approach. The implementation of appropriate ambiguity resolution has been demonstrated to enhance the accuracy of the solution, whilst concomitantly reducing the conservatism of confidence interval computation. It is evident that the determination of the corresponding confidence interval is contingent upon the consideration of the probability of correct ambiguity fixing [Garcia2024]. Consequently, the confidence interval is reduced while maintaining the integrity risk at the same level. The performance of the forward, backward and combined ambiguity-fixing strategies is compared. We discuss the potential accuracy improvement and confidence interval deduction by evaluating GNSS and IMU measurements for a large set of railway data collected within the RAILGAP project in Spain. The findings of this study provide a solid foundation for the establishment of a comprehensive and reliable railway ground truth. This is a prerequisite for the performance comparison of future railway systems. Additionally, this method can support the generation of a high-quality and high-integrity digital track map essential for the future railway system.

[CLUG2] – CLUG2 project website, https://www.clug2.eu/ ,[online], visited on 04.01.2026
[R2Dato] – R2Data project website, https://rail-research.europa.eu/rail-projects/fp2-r2dato/, ,[online], visited on 04.01.2026
[HELMET] - https://www.helmet-project.eu/, last visit 04.01.2026
[Grosch2025] A. Grosch and O. G. Crespillo, "Robust Railway Ground Truth and Confidence Bounds Estimation based on RTK/INS Forward-Backward Smoothing," 2025 IEEE/ION Position, Location and Navigation Symposium (PLANS), Salt Lake City, UT, USA, 2025, pp. 1336-1347, doi: 10.1109/PLANS61210.2025.11028193.
[RailGAPD4.8] - “RAILGAP report D4.8 - SENSOR FUSION ALGORITHMS DESIGN.” (2024), [online]. Available: https://railgap.eu/download.aspx?id=d7062d40-ae26-4fc3-9cf1-5ea656d3f244 (visited on 04/01/2026).
[Garcia2024] O. Garcia Crespillo, A. Brown, “Zero Baseline Evaluation of RTK Integrity Monitoring with Ambiguity Resolution”, Navitec 2024, 11-13 December 2024, ESA-ESTEC, NL

The European railway industry is actively developing an Advanced Safe Train Positioning (ASTP) capability for the future evolution of the European Rail Traffic Management System (ERTMS), providing safety-related train position, speed and acceleration to on-board functions including Automatic Train Protection (ATP) and Automatic Train Operation (ATO). The use of Global Navigation Satellite Systems (GNSS) in a multi-sensor architecture is considered one of the more promising approaches, where the European GNSS Navigation Overlay Service (EGNOS), the European Satellite-based Augmentation System (SBAS), plays a crucial role in enabling the ASTP to meet the stringent integrity requirements by protecting the user against satellite and ground segment fault conditions, and bounding residual orbit, clock and ionospheric pseudo-range errors.
Future railway Safety-of-Life (SoL) services are foreseen to provide commitments in the pseudo-range domain (excluding the impact of the local environment), quasi-independent of the user concept of operations, offering flexibility for the integration of GNSS into various solutions without a high level of prescription on the design. The ASTP is responsible for protecting users from local feared events, bounding residual errors due to the receiver and local environment, and computing parameters such as train position, velocity, acceleration, and their respective confidence intervals.
The use of EGNOS, however, imposes constraints on critical parameters of the user receiver configuration space to limit the complexity of the ground monitoring subsystem. These parameters include receiver pre-correlation filtering (bandwidth, roll-off and central frequency), differential group delay, and requirements on the tracking loop such as correlator spacing. They directly influence how errors in the corrected pseudo-range are translated from the threat space for fault modes such as signal distortions (Evil-Waveform). The design of GNSS augmentation ground monitors considers the full range of these critical parameters to ensure user protection against defined threat models.
Several options for safe-deviation from prescribed constraints have been investigated, such that the user does not rely on SBAS to protect against feared events linked with non-compliance of specific parameters and instead relies on additional barriers in the receiver to protect the user. This paper presents the options and trade-off analyses on possible railway GNSS receivers, described through examples of perimeters related to standardisation, bespoke development, railway supplier responsibility and potential usage of COTS (Commercial-Off-The-Shelf). Furthermore, a description of integrity algorithms is provided for railway GNSS receivers with safe-deviations supported by a justification of technical budgets (e.g., probability of misdetection and false alarm) for the additional barriers in order to fulfil integrity requirements. Finally, a preliminary quantitative assessment of the safe-deviations is presented, where the deviations and Evil-Waveform types are simulated in a dedicated GNSS-SDR receiver. The results highlight that the performance of the tested barriers is heavily impacted by local noise / multipath, limiting its efficacy in the railway environment. The paper concludes with a discussion on critical issues and further investigations on the candidate concepts.

Vehicle position and velocity information is vital for many digital railway use cases that determine the business of respective railway stakeholders. With the rise of affordable onboard GNSS (receivers and antennas) and IMU (inertial measurement units), more and more vehicle data become available in the rail domain. However, it is a complex task to convert said multi-sensor recordings with all their challenges and errors into position and velocity information of the required quality.

The paper shows how localization software for rail vehicle positioning developed by the DLR Institute of Transportation Systems can help solve this issue. A distinction between offline and online problems is highlighted. Modularity aspects with respect to different onboard sensor setups are discussed. For offline use cases a temporal segmentation into standstill and motion intervals is beneficial. Digital map data are crucial to exploit the constrained vehicle motion by estimating railway paths both online and offline. Path-constrained Kalman filter (KF) variants are especially useful to combine heterogeneous onboard data (different sources, rates, gaps, error characteristics) into track-selective position and velocity estimates with constant rates, interpretable estimation uncertainty, and robust performance. The above aspects advocate for a modular software architecture with separate problem domains and solutions with bounded context. On its basis code has been developed and prepared for the integration into real-world use cases derived from railway industry needs.

Advanced Safe Train Positioning (ASTP) is a key enabler to enhance the performance and cost-effectiveness of the European Rail Traffic Management System (ERTMS). By using new technologies such as global navigation satellite system (GNSS) based positioning, trains can estimate their position independently of infrastructure assets such as Eurobalises, which leads to higher capacities and reduced cost expenditure.
Any ASTP system will have to be designed to meet rigorous performance, safety, security, as well as availability and reliability targets. These requirements will have to be checked by a mix of theoretical and experimental means. To do this, a reference measurement is needed that is validated and at least an order of magnitude more precise than the targeted accuracy and confidence interval size.
Within the EGNSS MATE project, we have developed a Ground Truth algorithm based on the measurements of a ring-laser-gyro-based inertial navigation system aided by wheel odometry and real-time kinematic GNSS. In addition, a Eurobalise reader is installed on our test train that receives and timestamps the Eurobalise telegrams. Geocoordinates for each Eurobalise ID can be obtained from a digital map. Using map and route data from the test train, we then validate the INS positions in both longitudinal and lateral directions.
We first validate the route by checking that all Eurobalises are on the route. Where the route is validated, we project the INS position and check the cross-track error. In case the cross-track error is below a threshold, we check if when passing a Eurobalise the positions of the INS match the Eurobalise position. To also validate the along-track position between Eurobalises, we calculate the travelled distance between two Eurobalise readings from the INS and compare it to the Eurobalise distance on the route.
If any of these checks fail, the respective datapoints are excluded from the ground truth.

During the EGNSS MATE project, 400 train runs were processed with this ground truth algorithm, covering over 17,000 km. Our results show that the algorithm reliably detects errors and outliers in the sensor as well as in the map and route data.
Generally good ground truth coverage (over 70%) was obtained for most train runs. Challenging were the long Swiss rail tunnels like the Gotthard Base Tunnel (57km), where the INS drift cannot be compensated. Also, not up-to-date digital maps can lead to exclusions in areas with construction work. Missed Balise readings for Eurobalises on steel sleepers lead to poor ground truth availability in these scenarios.

The algorithm can also be used to assess the data quality of the digital map and is used now to correct erroneous position data of Eurobalises.

Continuous and reliable train localization is fundamental for modern railway train control systems for enable intelligent environment perception and train operation. Conventional train localization methods such as odometry, Balises, and Global Navigation Satellite Systems (GNSS) have inherent limitations in several environmental conditions. Exploring alternative or supplementary localization solutions using the environment features or information are of significant research and practical value. Vision-based perception has emerged as a promising solution for this challenge.
This paper proposed a localization framework using Stereo Visual Odometry (SVO) and trackside features. The framework aims to bound the SVO drift error by fusing opportunistic absolute updates from pre-surveyed trackside features, including Balises, and marked catenary poles. The framework using an enhanced one-stage object detection neural network adapted to railway scenes based on YOLO v8 to recognize opportunistic trackside features. The detector embeds lightweight priors including color-threshold masking and a position-threshold Return on Investment (ROI) gate, the constraints exploit scene regularities to filter implausible candidates, suppress false positives of the features. The features are used as opportunistic information to update the SVO drifts, the experiments under railway test track is studied, the results has shown that the vertical track error (VTE) can be reduced significantly compared with the SVO alone framework.