Initiative details
Road safety begins with how drivers are trained and assessed, and that foundation is broken. In Spain, only 50–60% of candidates pass the practical driving test, and just 27% pass on the first attempt, according to DGT and sector data. The reason is structural: evaluation still relies on manual, subjective and non-reproducible judgement by individual examiners and instructors. This produces inconsistent results between comparable assessments, no traceability or auditability of decisions, and severe bottlenecks, with waiting lists reaching six months in major cities. Crucially, the data generated during training is never exploited to improve learning or prevent risk.
Beyond the test itself, fatigue and distraction remain leading causes of collisions, responsible for between 10% and 25% of road accidents according to the European Commission. Yet professional fleets and training centres lack objective, real-time tools to detect risky behaviour and correct it at source.
Mettis AI addresses these challenges with AI and computer vision that evaluate driver behaviour objectively in real time, removing subjectivity from assessment, standardising criteria, and producing better-prepared, safer drivers, improving road safety from its root: training.
Beyond the test itself, fatigue and distraction remain leading causes of collisions, responsible for between 10% and 25% of road accidents according to the European Commission. Yet professional fleets and training centres lack objective, real-time tools to detect risky behaviour and correct it at source.
Mettis AI addresses these challenges with AI and computer vision that evaluate driver behaviour objectively in real time, removing subjectivity from assessment, standardising criteria, and producing better-prepared, safer drivers, improving road safety from its root: training.
Initiative date
Who was/is your target audience?
Young adults 17-25
Adults
Parents
Company employees
Fleet operators
Car drivers
Car drivers – professional
Educational staff
Public transport
Van drivers
Lorry/truck drivers
Topic
Create awareness
Education in school or in community organizations
Improve vehicles and infrastructure
Knowledge building and sharing
Organisation details
Mettis AI
Enterprise
Spain
Pozuelo de Alarcón, Madrid
Contact name
Filippo Maria Brunelleschi
Telephone number
+34619716723
filippo@mettis.ai
Website link
Project activities
If you work together with external partners, list the most important partners and briefly describe their role.
1) Queclink Wireless Solutions: hardware supplier providing automotive-grade cameras and sensors with a modular, plug-and-play architecture.
2) NVIDIA Corporation: supplier of edge-computing technology enabling real-time, low-latency on-vehicle inference; we are also members of the NVIDIA Inception programme.
Amazon Web Services: cloud-infrastructure provider for large-scale data ingestion, storage and model training.
3) Science Park of the University of Alicante: academic/innovation collaborator on research and technology pilots.
4) EFA (European Driving Schools Association): sector association of which we are a partner, supporting the move toward objective driver evaluation across Europe; we also work with Spain's Confederación Nacional de Autoescuelas.
5) AAMVA (American Association of Motor Vehicle Administrators): North American association of motor-vehicle and driver-licensing authorities; partner supporting alignment toward objective, standardised driver evaluation.
2) NVIDIA Corporation: supplier of edge-computing technology enabling real-time, low-latency on-vehicle inference; we are also members of the NVIDIA Inception programme.
Amazon Web Services: cloud-infrastructure provider for large-scale data ingestion, storage and model training.
3) Science Park of the University of Alicante: academic/innovation collaborator on research and technology pilots.
4) EFA (European Driving Schools Association): sector association of which we are a partner, supporting the move toward objective driver evaluation across Europe; we also work with Spain's Confederación Nacional de Autoescuelas.
5) AAMVA (American Association of Motor Vehicle Administrators): North American association of motor-vehicle and driver-licensing authorities; partner supporting alignment toward objective, standardised driver evaluation.
Please describe the project activities you carried/are carrying out and the time period over which these were implemented.
Since founding in 2024, Mettis AI has built an AI and computer-vision system that evaluates driver behaviour objectively in real time, combining high-resolution cameras, motion/position sensors and OBD connectivity, installable in any vehicle without structural modification and processing data on-board without storing video, ensuring GDPR compliance.
Through 2024–2025 we developed and trained our perception models on proprietary multimodal datasets gathered in real driving conditions, and validated them in real-world environments. We started working with with leading Spanish driving-school groups (Gala, Torcal and Educatrafic). In the fleet vertical, we are working with Gesinflot providing a fleet monitoring system which detects aggressive or risky driving patterns and supports personalised training plans.
In October 2025 we closed an ~€800k pre-seed round (Easo Ventures, Dozen Investments, Cabify co-founders). We are now executing commercial roll-out, scaling toward driving-school cars and logistics vehicles through, and expanding across Europe (Italy, France).
Through 2024–2025 we developed and trained our perception models on proprietary multimodal datasets gathered in real driving conditions, and validated them in real-world environments. We started working with with leading Spanish driving-school groups (Gala, Torcal and Educatrafic). In the fleet vertical, we are working with Gesinflot providing a fleet monitoring system which detects aggressive or risky driving patterns and supports personalised training plans.
In October 2025 we closed an ~€800k pre-seed round (Easo Ventures, Dozen Investments, Cabify co-founders). We are now executing commercial roll-out, scaling toward driving-school cars and logistics vehicles through, and expanding across Europe (Italy, France).
In terms of implementation, what worked well and what challenges did you need to overcome?
A key success was the development our edge architecture. Running inference on-vehicle with NVIDIA edge hardware kept latency low enough for real-time feedback, while cloud infrastructure handles large-scale data ingestion. We are using CARLA simulator (Unreal Engine 5) and we are generating synthetic critical scenarios, testing and refining models against rare events before deploying them in real vehicles, complementing our real-world data. Operationally, holding cameras and capture units already stocked in our warehouses substantially de-risked rollout logistics and enabled fast installation.
The main technical challenge was coherently combining heterogeneous data (interior/exterior video, GPS, accelerometry and OBD telemetry) into a single, time-synchronised picture of driving; we addressed this with a dedicated on-vehicle capture-and-synchronisation layer. A second challenge has been making the device as plug-and-play as possible: our current system is fairly plug-and-play, but we are actively developing a more compact, even easier-to-install edge unit, specifically adapted to our functionalities to guarantee compatibility, simple installation and scalability.
The main technical challenge was coherently combining heterogeneous data (interior/exterior video, GPS, accelerometry and OBD telemetry) into a single, time-synchronised picture of driving; we addressed this with a dedicated on-vehicle capture-and-synchronisation layer. A second challenge has been making the device as plug-and-play as possible: our current system is fairly plug-and-play, but we are actively developing a more compact, even easier-to-install edge unit, specifically adapted to our functionalities to guarantee compatibility, simple installation and scalability.
Evaluation
Please summarise how you have evaluated the initiative’s impact (e.g. social media reach, survey, feedback forms, statistics).
We evaluate impact mainly through structured, continuous feedback collected directly from our user base (driving-school instructors, examiners, fleet managers and learner candidates). Rather than relying on one-off surveys, we have built feedback mechanisms into the platform interface and into the reports it generates, so we capture actionable insights in real time as the system is used.
A central source is the practice session itself. After each lesson, both the instructor and the student can rate the session and leave a comment. We analyse these scores and comments systematically to identify recurring issues, refine functionalities, improve the user experience and adapt our evaluation criteria to different operational and regulatory contexts.
We also track usage data through our instructor dashboard, which records each student's progress, accumulated practice hours, recurring errors and readiness for the exam, giving an objective, longitudinal view of how learners improve over time. Alongside this, we gather qualitative feedback from regular direct contact with our clients.
Together, these methods let us measure how the initiative is performing in real driving-school environments and feed every finding back into continuous improvement of the system.
A central source is the practice session itself. After each lesson, both the instructor and the student can rate the session and leave a comment. We analyse these scores and comments systematically to identify recurring issues, refine functionalities, improve the user experience and adapt our evaluation criteria to different operational and regulatory contexts.
We also track usage data through our instructor dashboard, which records each student's progress, accumulated practice hours, recurring errors and readiness for the exam, giving an objective, longitudinal view of how learners improve over time. Alongside this, we gather qualitative feedback from regular direct contact with our clients.
Together, these methods let us measure how the initiative is performing in real driving-school environments and feed every finding back into continuous improvement of the system.
What has been the effect of the activities?
The core effect is that driving assessment is now supported by objective and data-driven metrics. Instructors actively use our platform in real driving-school environments to evaluate students, generate reports and track readiness for the exam, making training more personalised and standardising evaluation criteria.
The impact is currently national, concentrated in Spain across driving-school groups
We are actively collecting data to estimate two direct road-safety effects: the improvement in pass rates for driving-school students, and the reduction in accident rates among professional drivers. Having begun commercial rollout only in 2026, we do not yet hold a large enough dataset for precise statistics, but these will follow over the course of this year.
The impact is currently national, concentrated in Spain across driving-school groups
We are actively collecting data to estimate two direct road-safety effects: the improvement in pass rates for driving-school students, and the reduction in accident rates among professional drivers. Having begun commercial rollout only in 2026, we do not yet hold a large enough dataset for precise statistics, but these will follow over the course of this year.
Please briefly explain why your initiative is a good example of improving road safety.
Our initiative is a strong example because it improves road safety at its source, driver competence and behaviour, rather than only reacting to incidents after they occur. By bringing AI and computer vision into the practical learning process, we detect driving errors in real time and give personalised corrections, helping produce not just candidates who pass the exam but better-prepared, safer drivers from day one.
It is preventive and covers the full driver lifecycle: the same technology trains new drivers in driving schools and monitors professional drivers in fleets, flagging fatigue, distraction and risky patterns, factors the European Commission links to between 10% and 25% of road crashes.
It is also scalable and aligned with where regulation is heading: our system installs in any vehicle without modification, processes data on-board without storing video (GDPR-compliant), and supports the EU direction toward in-cabin driver-monitoring systems in new vehicles.
Finally, by generating objective, standardised data on real driving behaviour, it builds a shared evidence base that benefits driving schools, fleets, insurers and, ultimately, safer roads across Europe.
It is preventive and covers the full driver lifecycle: the same technology trains new drivers in driving schools and monitors professional drivers in fleets, flagging fatigue, distraction and risky patterns, factors the European Commission links to between 10% and 25% of road crashes.
It is also scalable and aligned with where regulation is heading: our system installs in any vehicle without modification, processes data on-board without storing video (GDPR-compliant), and supports the EU direction toward in-cabin driver-monitoring systems in new vehicles.
Finally, by generating objective, standardised data on real driving behaviour, it builds a shared evidence base that benefits driving schools, fleets, insurers and, ultimately, safer roads across Europe.
How have you shared information about your project and its results?
We share our project actively within the road-safety and driver-training community. We regularly take part in conferences of European driving-school associations, most recently the Mobilians conference in Toulon in May, organised by EFA (European Driving Schools Association), where we present our approach to objective, AI-based driver evaluation. We also engage through start-up competitions, sectoral forums and trade shows, which let us reach driving schools, fleet operators and mobility stakeholders directly. Notably, we won the 2025 Grand Prix of the French Tech Train and took part in Tech Business PlaNET, an international technology-innovation showcase, further raising the project's visibility.
Supporting materials
