The paper presents a practical and cost-efective approach to indoor navigation that reduces reliance on highprecision positioning systems. Developed within the LIFT project at the Warsaw University of Technology, the proposed system is tailored for metro environments and supports users, in navigating through stations with greater confidence and autonomy. By combining Bluetooth Low Energy ( BLE) -based positioning with landmark-based instructions, contextual images, and Augumented Reality (AR) elements, the system delivers intuitive guidance without requiring sub-meter accuracy. A three-layer spatial data model underpins the route generation and instruction framework, emphasizing user interpretability over constant real-time tracking. The mobile application, deployed and tested in the Warsaw Metro received positive feedback for its usability and accessibility. The findings demonstrate that efective indoor navigation can be achieved through intelligent instruction design and user-centric interface development, rather than solely through technical enhancements in positioning accuracy.
(M. Domagalski)
CEUR
ceur-ws.org
Faculties of Geodesy and Cartography, Electronics and Information Technology, Civil Engineering, and
Architecture. Beyond the development of technical solutions, the research has included study visits to
cities such as Barcelona and Los Angeles to examine efective indoor navigation practices. The project
also benefits from close collaboration with organizations representing people with disabilities, ensuring
that the solutions developed are grounded in real-world needs. Preliminary results were presented at
the CSUN Assistive Technology Conference [
This paper presents the conceptual foundation of a system designed to navigate users through metro stations eficiently, without requiring sub-meter positioning accuracy. The proposed solution is based on a customized spatial data model and an adaptive instruction generation algorithm.
The paper is structured as follows: Section 2 reviews the current landscape of indoor positioning systems and their limitations in enabling high-quality navigation. Section 3 introduces the developed system, and Section 4 evaluates its applicability in real-world environments.
Indoor navigation systems are widely used in transportation hubs such as airports, metro stations, and train terminals. These systems are implemented either via customized venue-specific apps or through general-purpose apps developed by third-party providers. Custom apps typically ofer deeper integration, incorporating maps, points of interest, and contextual features such as notifications. This makes navigation smoother and more intuitive. On the other hand, general apps provide only basic venue maps, requiring users to manually select their destination, often with limited context. The analysis of both types of apps shows that most existing solutions use traditional step-by-step navigation, which relies heavily on the precision of the underlying positioning system. When positioning fails or is not accurate enough, the instructions become unreliable.
Research on indoor navigation in airport environments shows a strong focus on improving positioning
systems, particularly in enhancing accuracy and reliability. For example, the solution presented in [
The metro stations and buildings are usually complex and architecturally distinct from other typical
indoor venues. They consist of numerous corridors, which are often narrow and prone to high pedestrian
density. Compared to airports, there are significantly fewer applications specifically designed for metro
navigation. In [
Extensive research has been conducted on how individuals navigate through indoor environments and
how architectural and signage design can facilitate intuitive wayfinding. In [
The previous chapter focused on reviewing literature relevant to the topic. While the solutions presented in these studies are not without limitations, they provide a solid foundation upon which the proposed system is built. The system developed in this work aims to address the issues identified above by designing an indoor navigation solution that does not rely on sub-meter positioning accuracy. Instead, it leverages the user’s ability to interpret contextual information by providing landmark-based instructions and images of the surroundings, enhanced with AR elements.
While the methodology for implementing the positioning module is not the primary focus of this paper, it is important to briefly describe the technology used and the achieved accuracy in order to provide context regarding the extent to which positioning inaccuracies can be mitigated by the applied methods.
The indoor positioning module is based on BLE beacons installed throughout the metro stations. The positioning technique used is fingerprinting [ 10][11], chosen primarily because the creation of the station’s signal map could be synchronized with the process of capturing images for the image-based instructions, thereby reducing the overall workload.
After collecting the signal data necessary for positioning, a machine learning model was trained. Three diferent algorithms were evaluated: • K-Nearest Neighbours (KNN) • Support Vector Machine (SVM) • Random Forest (RF) Among these, the RF algorithm yielded the highest classification accuracy of 77% resulting in the average distance between the user’s true position and the predicted node of the navigational graph of 6.15 meters This level of precision is suficient for the system’s needs, as the accuracy of user positioning is inherently constrained by the data model. Specifically, the user’s position is always snapped to the nearest node in the routing graph, where the average spacing between nodes is approximately 8.5 meters.
The system has been specifically designed for the Warsaw Metro, and the architectural analysis is therefore tailored to the characteristics of this transit network. Metro stations typically feature entrances with staircases that lead into narrow corridors, eventually connecting to the main station area with turnstiles [12]. The layout is generally linear, guiding passengers from the entrance to the platform and vice versa, which naturally reduces the need for high-precision indoor positioning.
Unlike ofice buildings or malls, metro stations lack complex branching paths or numerous individual rooms. In such buildings, users often need to be directed to specific rooms or areas, increasing the demand for precise localization. Metro stations, by contrast, are designed to accommodate large volumes of pedestrian trafic, resulting in wider corridors and open spaces. Additionally, the boundaries between functional areas in metro stations are less distinct—these are primarily transit spaces, and heavy spatial separation (e.g. via doors or partitions) would hinder passenger flow and reduce system eficiency.
Another important consideration in the system’s development was designing the user interface and application flow to function independently of the indoor positioning module [ 13]. By default, the user manually selects the starting point and destination of the journey, as they may not currently be located within the metro station. The application also provides an option to determine the user’s location via GPS, which is particularly useful when outdoors; in such cases, the nearest station entrance is automatically selected as the starting point.
While the indoor positioning module runs in the background, if it successfully detects the user’s location, a pop-up notification prompts the user to confirm whether they would like to see their location (Figure 2). Later on they can select it as a starting point of their journey. Importantly, the application does not continuously track the user’s movement during navigation. This decision is based on the rationale that displaying live location updates or highlighting real-time instructions may introduce confusion if the positioning data is inaccurate. Instead, the system relies on the user’s ability to interpret contextual cues and navigate based on provided instructions.
The generation of landmark-based instructions is closely linked to both the system’s data model and the algorithm used for generating the instructions. The overall process has been described in detail in a previous publication by the research team [14].
The data model consists of three distinct layers: • Transport Network • Room Topology • Topography
The Transport Network layer is primarily used for route calculation, as it contains the geometries of path segments. However, the layers most relevant to landmark-based instruction generation are the Room Topology and Topography layers.
The Room Topology layer identifies when a change in spatial context occurs—such as moving from one room to another, ascending a staircase, or similar transitions. At this stage, instructions may lack full contextual detail, but identifying such transitions is essential for meaningful navigation.
The Topography layer enriches instructions with contextual information. It includes details about interior features, named zones within stations, and descriptions of prominent landmarks. This layer plays a key role in generating intuitive and user-friendly guidance.
To ensure that users can navigate the metro system freely without relying on sub-meter positional accuracy, an additional solution has been implemented: the inclusion of visual cues. The data model has been extended to incorporate a link between the geometries of path segments and images representing the surrounding environment within the Transport Network layer. This allows the application to display relevant images to the user once a path has been determined, helping them visually confirm the correct route.
Additionally, the application utilizes AR to further assist passengers by overlaying navigational guidance directly onto the real-world environment (Figure 3).
Navigation instructions generated by the algorithm fall into five possible maneuver types: • right turn • mild right turn • left turn • mild left turn • straight
The type of maneuver selected depends on the distance over which the user must make the turn. If the turn is executed over a longer distance, a mild turn (implying a smaller angle) is selected. For shorter distances, a sharper turn is expected, so a standard left or right turn is chosen.
Usability tests were conducted with over 30 participants representing diverse backgrounds, age groups, and familiarity levels with the metro system, with additional feedback collected from real-life users via the publicly available application on the Apple App Store and Google Play Store. To ensure objectivity, none of the core test participants had prior experience with the mobile application. Each participant completed a designated nine-stop metro journey, including one line transfer, both with and without the app. The route, shown in Figure 4, was intentionally designed to involve elevator navigation and a transfer to reflect realistic, moderately complex travel scenarios. After a brief introduction, participants used the in-app map to locate the nearest elevator and initiate the “to platform” navigation, which provided on-screen guidance from the surface to the metro platform. Following a 15-minute journey and a line change, they arrived at the destination station, where they used the “to surface” function to set route parameters and receive step-by-step instructions to the target tram stop. Upon completing the journey, participants filled out a short questionnaire, which asked whether using the mobile app afected their comfort with the metro system, whether a more precise indoor positioning module would improve or worsen the experience, and whether they relied on text-based landmark instructions, image-based instructions, or both while navigating.
An exact explanation of the process of analysis of the results has been omitted due to the page limitation of the paper. Collected feedback allowed us to draw the following conclusions: • Overall, user feedback was positive. Most users reported that the application increased their comfort and confidence when using the metro. None of the participants indicated issues related to their positioning within metro stations or expressed a need for a highly precise indoor positioning system. This confirms the main thesis of this paper: sub-meter accuracy is not required for efective metro navigation. Among the implemented features, landmark-based instructions received the most praise. Users appreciated being informed about their surroundings and what to expect along their journey, which helped them avoid major navigational errors. • Positive feedback was also received from users with visual impairments. Although they could not benefit from image-based instructions, the landmark-based guidance significantly enhanced their ability to visualize the route and increased their travel confidence. Unlike standard indoor navigation apps that rely on rigid, step-by-step instructions, this system ofers a more intuitive and lfexible approach. Users are not forced to follow a predefined path precisely, reducing dependency on the system and decreasing the risk of accidents caused by incorrect positioning—particularly important for visually impaired users who rely more heavily on the navigation system. Another benefit for this group is the ability to pre-plan routes at home without needing an active indoor positioning module, allowing them to mentally prepare by visualizing the environment beforehand.
Despite the generally positive feedback, several limitations of the application were identified. One notable issue was a decrease in user attention to their physical surroundings during navigation. The use of the mobile device led some users to focus more on the screen than on their environment, increasing the likelihood of minor mistakes that may not have occurred otherwise. Additionally, in certain instances, users made significant navigational errors without being immediately aware of them. When this happened, the visual or textual instructions provided by the application no longer matched the user’s surroundings, resulting in confusion and a loss of time as users attempted to determine whether the instructions were incorrect. In some cases, this led to users arriving at unintended locations.
To address these limitations, future development could incorporate a detour detection mechanism. This feature would monitor a user’s position relative to the planned route and detect significant deviations. Upon detecting a detour, the system could issue timely alerts to guide the user back to the correct path or, at the very least, inform them that they are of-course. Such a capability would reduce confusion and minimize time lost in returning to the intended route. Another improvement involves mitigating ambiguity in complex environments, such as areas with similar-looking doors or adjacent elevators. By incorporating a spatial analysis algorithm capable of detecting such situations, the system could either alert users to potential confusion or ofer more detailed instructions, thereby enhancing clarity and user confidence during navigation.
The author(s) have not employed any Generative AI tools. [9] I. Fellner, H. Huang, G. Gartner, “turn left after the wc, and use the lift to go to the 2nd floor”—generation of landmark-based route instructions for indoor navigation, ISPRS International Journal of GeoInformation 6 (2017). URL: https://www.mdpi.com/2220-9964/6/6/183. doi:10.3390/ijgi6060183. [10] P. I. Philippopoulos, K. N. Koutrakis, E. D. Tsafaras, E. G. Papadopoulou, D. Sigalas, N. D. Tselikas, S. Ougiaroglou, C. Vassilakis, Cost-eficient rssi-based indoor proximity positioning, for large/complex museum exhibition spaces, Sensors 25 (2025). URL: https://www.mdpi.com/1424-8220/ 25/9/2713. doi:10.3390/s25092713. [11] E. Skýpalová, M. Boroš, T. Loveček, A. Veľas, Innovative indoor positioning: Ble beacons for healthcare tracking, Electronics 14 (2025). URL: https://www.mdpi.com/2079-9292/14/10/2018. doi:10.3390/electronics14102018. [12] A. Tofiluk, M. Domagalski, B. Wiktorzak, Analiza wybranych aspektów przestrzeni warszawskiego metra w kontekście dostosowania do potrzeb osób z niepełnosprawnością ruchu, Builder 27 (2023) 20–26. [13] G. Szewczuk, R. Olszewski, Development of methodology for designing indoor cartographic visualizations for use in navigation for persons with special needs, Polish Cartographical Review 56 (2025) 115–133. [14] J. B. Marciniak, B. Wiktorzak, Automatic generation of guidance for indoor navigation at metro stations, Applied Sciences 14 (2024). URL: https://www.mdpi.com/2076-3417/14/22/10252. doi:10. 3390/app142210252.