Beyond the vehicle: autonomous driving as the new architecture of mobility

A conversation with Sergio Matteo Savaresi, Director of the Department of Electronics, Information and Bioengineering of the Politecnico di Milano and co-founder of Niulinx.

Interview by Paolo Barbato with Sergio Matteo Savaresi.

The number of highly automated vehicles in circulation worldwide was projected at around 34,000 units by the end of 2025 and around 43,000 by the end of 2026, concentrated almost exclusively between the United States and China. Service volumes, however, tell a story of growth that is already real: robotaxi rides on paid commercial services have multiplied rapidly. Waymo alone went from around 4 million rides in 2024 to over 14 million in 2025; at the start of 2026, between Waymo (around 26 million rides a year at the current pace) and Baidu Apollo Go (3.2 million in the first quarter of 2026 alone), the global volume comfortably exceeds 100 million rides a year. On a weekly basis, the services already active across the United States and China together move more than 800,000 rides a week at the start of 2026, driven by Waymo (around 500,000 paid weekly rides in the United States) and Baidu Apollo Go (peaks above 350,000 a week in China in March 2026), not counting the other Chinese operators. In Europe, the first commercial launches of urban robotaxis are expected between 2030 and 2035: a delay of four to six years.

Italy starts from a mobility model that makes this gap particularly costly. There are 701 passenger cars for every 1,000 inhabitants, the highest rate in the European Union against an average of 578 (ISTAT, Vehicle Fleet Indicators, 2024); the fleet exceeds 41.3 million cars, with an average age of 13 years, and those cars travel almost empty, with an average occupancy in cities of 1.3 people per vehicle. We own more cars than anyone in Europe, we keep them longer and we barely fill them. It is a model that is hard to dismantle, because the alternative, outside the large cities equipped with a metro network, often does not exist or is not competitive.

It is within this picture that the Italian bet on autonomous driving sits. In April 2026 Niulinx, a spin-off of the Politecnico di Milano born from the AIDA (Artificial Intelligence Driving Autonomous) research group, closed a 38 million euro round led by A2A and CDP Venture Capital, with the participation of Ferrovie dello Stato Italiane, Pirelli, MOST and other industrial players. It is the largest Italian investment ever made in the sector. The model that Niulinx pursues, called “robosharing”, does not aim to replace public transport but to fill the gaps that today make it a non-competitive alternative for many daily journeys.

Sergio Matteo Savaresi coordinates the AIDA research group and is co-founder of Niulinx. In a recent keynote in Milan we discussed the long-term vision, the numbers of urban scalability, the regulatory knot and the economic model. What follows is the thread of the conversation.

The fork between function and emotion

Paolo Barbato: Professor, let’s start with the long-term vision. I often say that the automobile, taken in isolation, is an engineering masterpiece: a century of refinement makes it an extraordinary object. Placed within the mobility system, however, it becomes an engineering disaster: we keep it parked 95% of the time and when it moves it travels with 1.3 people on board. When you talk about autonomous driving you are not talking about an incremental improvement of the vehicle, but about a transformation of the mobility model itself. What exactly do you mean?

Sergio Matteo Savaresi: The private car today contains within it two dimensions that coexist: a functional one (I have to go from A to B) and an emotional one (driving as pleasure, the car as an identity object). Over the next 20-30 years these two dimensions will fork. The functional part will be satisfied by robotaxis and similar services, where emotion does not exist: it is pure transfer. The emotional part will gradually slide toward the tracks, where driving will remain a sport or pastime. The analogy is that of the horse: from a functional mobility tool to an emotional object confined to racecourses.

There is a generational data point that strikes me every time I tell this vision. People over 50, myself included, tend to shudder at the idea of losing their own car. The new generations have the opposite attitude: in my research group there are young researchers who do not have a driving licence and will not get one. For them the private car is not an untouchable icon, it is one service among others.

Cannibalization of public transport: the misunderstanding to clear up

Paolo Barbato: On our blog we often talk about multimodality. The typical concern of those who plan mobility and approach these technologies is simple: a comfortable, fast and affordable robotaxi ends up emptying the metro, trains and buses. Is it a real risk?

Sergio Matteo Savaresi: The answer is no, and let me argue it. Trains and metros are extremely efficient: they handle enormous demand peaks on fixed axes, at very low unit costs. They are untouchable. The problem is the bus in medium-small cities: it typically travels almost empty, has high costs and a very low capacity for calibration.

The model we imagine does not replace public transport, it extends it downward. At the top remain trains and metros on the main routes. Below, a continuum of on-demand road transport: from smaller buses (10-20 seats instead of today’s 50) down to the two-seat electric microcar, which is in effect an on-demand module for almost single use.

The data point I love to cite: the average occupancy of a car in cities is 1.3 people per vehicle. A two-seat microcar is already above the real occupancy average. Our ambition is to bring individual motorized mobility inside multimodality, not to push it out. I call it “multimodality by design”: each leg has its optimal mode, and today the private car sits outside this reasoning because it is a fixed asset, not a variable service.

Scalability in mid-sized Italian cities

Paolo Barbato: One of the recurring objections is that these technologies only work in large metropolises. An important share of users is in mid-sized cities or in peripheral industrial hubs, where demand density is much lower. Do you have data on scalability in these contexts?

Sergio Matteo Savaresi: We have done a lot of analysis on insurance telematics box data. Let’s set aside Milan, which is a special case. The example I always bring is Brescia: 200,000 inhabitants, a single metro line, relatively low public transport coverage. Just 40-50 cars are enough to guarantee a waiting time of 3-4 minutes maximum. A small number, a very high level of service.

The other case we studied is Darfo Boario Terme: a scattered, mountainous context, where the first mile toward the railway station is today an obstacle that pushes people toward the private car. In that context a few units are enough to make multimodality truly viable.

The message for those who plan mobility is twofold. The barrier to entry is low: you do not need thousands of vehicles to start. At the same time, below a certain fleet threshold the service does not work, because the waiting time becomes prohibitive. It is a delicate point for any planning on a territorial scale.

The regulatory knot and European type-approval

Paolo Barbato: The public narrative on autonomous driving in Europe emphasizes the regulatory block. As an insider, is it really the main obstacle?

Sergio Matteo Savaresi: Regulation is a bit of a myth. The truth is that in Europe it is already technically feasible to go to type-approval. Difficult, yes. Impossible, no. France, Croatia and Austria have already carried out the review of their national regulations.

Niulinx does not have an Italian ambition, it has a European ambition. We will soon open offices in some European states and we will aim for type-approval in several countries simultaneously. The goal is to remove the safety driver and make the service truly available to citizens by the end of 2028 or the start of 2029. Ambitious, but feasible.

On the industrial side it is worth clarifying one point: 38 million is a lot, but to go all the way you need much more. The order of magnitude of these technologies is billions, not hundreds of millions. We are already planning a significantly larger second round, a third round with one more zero, and so on. It is a capital journey measured over ten years, not three.

Critical infrastructure and technological sovereignty

Paolo Barbato: You have defined autonomous driving as critical infrastructure. What does this mean in terms of governance of the service? Will it be managed by private operators, by municipal companies, or by a combination of the two?

Sergio Matteo Savaresi: It is a political choice, and I am not talking about right or left: I am talking about concrete political choices. Today, at the highest institutional level, the idea is taking hold that this type of technology is a critical infrastructure, over which governments must have technological sovereignty and control capability.

You cannot leave a fleet of hundreds of thousands of autonomous vehicles operating in cities in the free hands of the market. They can be municipal companies, they can be private operators with strong public control over the technology, they can be hybrid models. What is certain is that some form of public control will have to exist, because matters of national security, of algorithmic ethics, of access to mobility data are at stake.

The economics of the service: the break-even point

Paolo Barbato: A question that recurs among those who plan mobility and in company management is the final price for the citizen. If the service costs as much as a taxi it does not shift behaviour. What are the real numbers?

Sergio Matteo Savaresi: The taxi today is a premium service that costs 4-5 euro per kilometre. It is another world. The break-even point to shift behaviour is the per-kilometre cost of the underused private car.

And here there is a data point that few people know. A family with a second car, a small new compact used 5,000-6,000 kilometres a year, has a per-kilometre cost above 1 euro. Almost no one knows this, because people think only about the cost of petrol or charging and forget the depreciation of the asset, the insurance, the maintenance. To buy the car you make an outlay of 15,000-20,000 euro that then has to be spread over the actual kilometres.

Our analyses indicate that robosharing can comfortably drop below one euro per kilometre: we are talking about 0.7-0.8, ideally 0.5. With an enormous advantage over the private car: there is no entry point. You do not have to face an initial expense of 20,000 euro that you then have to depreciate. You pay only for the kilometres you travel, like a public transport ticket. This is extremely inclusive, especially for those who today are excluded from individual mobility: the elderly, people with disabilities, low-income groups.

What it means for those planning mobility today

Paolo Barbato: Let’s conclude. Those who plan the mobility of an industrial hub or a medium-small metropolitan area look at these technologies with curiosity, but also with the feeling that their horizon is too far off for today’s decisions. What makes sense to do now?

Sergio Matteo Savaresi: The first point is about timing. The end of the decade is not far off: those who today build plans on a 2025-2028 horizon are already reasoning about the period in which the first commercial services, at least in some European countries, will be operational. Taking it into account as a scenario, if only to avoid being caught unprepared, already makes sense now.

There is then a question of priorities. The piece that robosharing will fill first is the first and last mile toward railway stations. For many areas outside urban centres it is exactly the constraint that today pushes people to use the car for the entire journey, even when part of it would be better covered by rail. Mapping that segment of the network precisely today means being ready to integrate it when the alternatives arrive.

And finally there is inclusivity, which is not a rhetorical abstraction. The Italian population is ageing: I am a baby boomer, and within 10-15 years my generation will have difficulty with individual mobility. Those who today look at mobility only as an environmental theme will soon have to look at it also as a theme of accessibility and welfare. Autonomous driving is one of the technologies that will make that transition manageable, and it is worth starting to build the framework while there is still time to do so.

Sergio Matteo Savaresi

Sergio Matteo Savaresi is Director of the Department of Electronics, Information and Bioengineering of the Politecnico di Milano and coordinator of the AIDA - Artificial Intelligence Driving Autonomous research group. He is co-founder of Niulinx, a spin-off of the Politecnico di Milano dedicated to autonomous driving, which in April 2026 closed the largest Italian round in the sector with A2A and CDP Venture Capital as lead investors.