GO-Mobility

AI at the service of public transport

Case studies and applications around the world

Artificial intelligence is already transforming public transport. According to the latest Knowledge Brief from UITP, AI applications in the sector are becoming increasingly concrete and measurable, thanks mainly to three key technologies: Large Language Models (LLM), video analytics, and predictive modeling. The applications are manifold: from increasingly popular chatbots for customer service to the development of digital avatars for sign language translation to driving behavior monitoring systems and much more. The implications of these tools have a significant impact on the accessibility, safety, and resilience of transport, but there are also critical issues and limitations. Let’s take a look at them together in this month’s article dedicated to concrete cases of AI application in public transport in Europe and around the world.

Inclusive and dynamic communication with LLMs

Thanks to LLMs, chatbots and virtual assistance systems have enhanced their ability to respond in an increasingly natural, accurate, and multilingual way, helping to break down linguistic and sensory barriers. Let’s look at some examples of application.

SBS Transit in Singapore has co-developed SiLViA, a digital sign language avatar based on advanced speech recognition algorithms that are used to instantly translate spoken or written words (e.g., announcements at stations or on board vehicles) into sign language. Similar trials are also underway at Belgrade Central Station and the Port Authority of New York and New Jersey.

PostBus in Switzerland provides its passengers with real-time audio announcements in multiple languages, including German, French, Italian, and English. These are automatic announcements relating to regular information and updates, as well as planned or unexpected events. To do this, it has implemented a framework consisting of a cloud backend and an on-board system in vehicles, using APIs that allow AI and machine learning capabilities to be used in a simple and scalable way. To ensure high-quality service, PostBus staff regularly audit samples of messages to assess their accuracy and clarity.

The Chicago Transit Authority (CTA) has launched a chatbot called “Chat with CTA” to report incidents in real time and collect urgent reports on issues such as maintenance, cleanliness, or disruptive passengers. The chatbot was developed using Google’s natural language processing (NLP) product and is currently able to handle 79% of customer requests. The planned update will integrate Google’s RAG (retrieval-augmented generation) technology to provide answers to frequently asked and simple questions about services, enabling the agency to offer useful answers in at least 95% of cases.

View this post on Instagram

A post shared by Chicago Transit Authority (@chicagocta)

UITP also mentions an Italian case: in January, Club Italia launched the open-source application Velvet, a chatbot capable of providing real-time updates, including service schedules and advice on the most suitable clothing based on the weather conditions forecast for the day of travel ☀️. The solution is currently being tested with several partner institutions, including IntercentER (Emilia-Romagna Regional Agency for Electronic Procurement), the online university UniMarconi, and the Veneto Region. The goal is to implement Velvet in central and southern Italy, with a particular focus on public transport operators in Sicily.

Finally, there is the interesting case of AC Transit in California, which has introduced a conversational chatbot to provide IT support to on-board staff. The company was struggling to provide IT support to drivers, whose shifts—especially night shifts—are often incompatible with those of IT support, creating difficulties and inefficiencies related to the need to resort to voicemail messages or forwarding requests. To address the problem, it introduced “IT Aimee,” a genAI-based chatbot that provides direct support and access to essential IT and organizational information. IT Aimee leverages resources such as articles from the Boss Desk knowledge base and troubleshooting documentation. Designed to provide immediate 24/7 assistance, IT Aimee has significantly improved resolution times and accessibility for operators.

Onboard safety and fare evasion prevention with video analytics

The use of AI in the field of video analytics is transforming cameras into intelligent tools capable of monitoring passenger flows, improving safety on board and near level crossings, detecting unauthorized access, and optimizing maintenance. The most recent experiments aim to identify abnormal behavior and dangerous abandoned objects in real time. Future prospects focus on local data processing (edge computing), integration with IoT sensors, and immediate incident reporting. It remains crucial to address issues related to image quality and the adaptability of AI models to constantly changing contexts. Let’s look at some cases.

The Public Transport Authority of Singapore (PTA) has addressed safety and efficiency issues by adopting advanced driver assistance systems, driver fatigue monitoring, blind spot detection, and a high-capacity video surveillance system developed by Streamax and installed on 5,500 buses. This upgrade has significantly improved safety, raising the quality of service and making the control of bus lanes more effective. Passengers also benefit, for example, from displays that show the availability of seats on the upper deck in real time.

The fight against fare evasion has also evolved with artificial intelligence. Since 2015, FGC Barcelona has adopted a system developed by AWAAIT, installing smart cameras in major urban stations. These detect passengers who pass through the gates without validating their tickets in real time and send notifications to an app used by inspectors for targeted checks. The cameras have proven to be an effective deterrent, contributing to a sharp reduction in fare evasion (-70%).

In New York, on the other hand, the Metropolitan Transportation Authority (MTA) is testing real-time detection systems for tunnel intrusions and fare evasion across the entire network. An AI system has been developed that can identify cases of illegal track crossing, report offenders, and generate video clips for in-depth analysis. This tool is based on object recognition and the collection of data on recurring behaviors for prevention purposes.

In Sofia, the Center for Urban Mobility uses real-time images from cameras already on board buses to classify vehicle occupancy into five levels. This system, developed by Theoremus, avoids the need to count passengers getting on and off, a method prone to cumulative errors and often requiring additional sensors. The solution has proven to be efficient: 50,000 images and a streamlined AI model were enough to obtain reliable results.

Once again, we mention AC Transit, which has developed a system to combat illegally parked vehicles at bus stops or in bus lanes, which are known to be an obstacle to the smooth running of the service and passenger safety. To do this, it implemented Hayden AI‘s computer vision technology on a large scale, consisting of front-mounted cameras on buses equipped with edge processing for automatic detection of violations and recording of short videos showing the license plate, time, and location of the vehicle. The system guarantees a much higher detection rate than manual checks, without invading privacy (it does not record the interior of the bus or use facial recognition).

Finally, in Boston, a mobile app was launched in 2024 to help blind people find their way to bus stops, using an autonomous AI-based system developed by the Schepens Institute, which uses the phone’s camera and an AI model trained on a set of images of bus stops to visually guide the user to the nearest stop. The result is a scalable, accessible, and autonomous solution that eliminates the need for external assistance and significantly improves the independence of users with visual impairments.

Predictive modeling to anticipate events and adapt data-driven services

Finally, predictive modeling, based on machine learning techniques, allows future events to be predicted or hidden patterns to be identified by analyzing large amounts of historical data. In the context of public transport, this technology is revolutionizing the planning of resources—staff, vehicles, services, and infrastructure—improving efficiency, service quality, and safety. Despite technological advances, challenges remain in terms of data availability and quality, model transparency, and user trust. Here too, we see the most telling examples.

Alsa Morocco has developed a machine learning algorithm designed to promote efficient and tailored driving styles. The algorithm, based on clustering, is capable of correlating vehicle telemetry data with external information such as traffic, passenger load, and weather conditions, taking into account differences between routes and different times of day, to identify driving behaviors appropriate to each context. To ensure impartial assessments and encourage improvement, it also applies gamification techniques with incentives and rewards, and organizes targeted training sessions based on recurring misconduct. The results? A reduction in fuel consumption of between 4% and 12% and a decrease of between 15% and 40% in vehicle misuse, with direct effects on safety, passenger comfort, and vehicle condition.

Arriva (Spain) has adopted the smart charging solution based on an AI model developed by Bia, trained on historical data from charger telemetry systems and fleet management. The platform dynamically optimizes charging cycles, scheduling them at times when rates are lowest and favoring slower charging speeds to extend battery life without compromising departure times. The results speak for themselves, with a 25% reduction in energy costs while fully complying with scheduled services.

In Hamburg, an artificial intelligence algorithm developed by PTV (a world leader in real-time traffic planning, simulation, and management software—widely used by GO-Mobility) generates short-term forecasts (5-30 minute horizon) on the impact of relevant events by combining a model based on individual mobility behaviors with real-time data. An accessible, user-centered map interface shows traffic, roadworks, accidents, and other relevant events in real time, providing operators and authorities with a shared, data-driven view to make more effective decisions in jointly resolving traffic issues.

The Irish National Transport Authority (NTA) has implemented a model for dynamic arrival forecasting, improving accuracy by 13%. Trapeze technology features a predictive engine based on machine learning that dynamically adapts the weight of short-term and historical data, while also integrating information on traffic, weather, and passenger load.

In Barcelona, predictive AI has been used to control ventilation in the metro. A predictive control system, developed by SENER, optimizes the ventilation strategy in real time, taking into account variables such as weather, indoor and outdoor air quality, energy consumption, fan performance, and energy costs. In addition to helping to contain the spread of COVID-19 during the pandemic, in 2022 it led to a +20.9% increase in fan efficiency, a 25.1% reduction in energy consumption, and a 10.7% increase in passenger satisfaction.

Lessons learned and next steps

In a rapidly evolving regulatory environment, it is important to proceed with caution when adopting artificial intelligence in public transport, with care, balance, and public oversight. The recent publication of the AI Safety Index Summer 2025, which assesses the safety practices of seven leading AI companies, has revealed alarming results regarding shortcomings in responsible AI development practices and the management of their systems’ safety.

In this sense, regulations such as the AI Act and the GDPR in Europe, or the Blueprint for an AI Bill of Rights in the United States, emphasize transparency, privacy protection, and ethical use of technologies. At the same time, the energy sustainability of AI models is becoming a key factor: hybrid solutions that combine large general models with leaner, more specific models allow for effective and less impactful implementation. Successful projects are based on strong collaboration between transport authorities and technology providers, dedicated internal teams, and attention to data quality. AI already offers concrete results in areas such as forecasting and operational management, but its true potential will be unleashed with a proactive, competent approach that is attentive to the regulatory and environmental context.