When we take a look at the vehicles available in the various catalogues, we are no longer surprised to read among their features systems such as “automatic emergency braking”, “adaptive cruise control” or even “automatic lane change”. These technologies, unthinkable just a few decades ago, are a good example of the innovation that is driving the automotive industry. In this way, they increase both safety and the enjoyment of travelling in a personal vehicle, increasingly freeing the driver from the inherent actions of driving and, by extension, from the possible associated mistakes that can be made.
The current trend is therefore to create vehicles that are increasingly intelligent and autonomous, reducing the number and severity of road accidents that kill thousands of people around the world every year. But this is not a simple path. The development of autonomous driving faces enormous challenges, both engineering and legal, before it can set foot on and conquer the roads. For this reason, a growing number of projects are trying to find the keys that will shape the mobility of the future. Many of them in our country.
INSIDE. Artificial intelligence leads to the new mobility
Almost 140 years have passed since Carl Benz patented his “petrol-powered motor vehicle” on 29 January 1886, an invention that brought about a paradigm shift in the way people move around. Automobiles transformed the urban environment and brought cities closer together, expanding the boundaries of people's lives. They also emptied them of draught animals, improving their levels of sanitation. As the chronicles of the time indicate, by the end of the 19th century the big cities were in the midst of a major health crisis brought about by the bestiary. In London, for example, the 50,000 draught animals produced around half a million kilos of manure every day, which had to be removed from the streets. As a result, many families decided to raise their houses to prevent the manure from reaching the doors. In New York, early urban planning meetings focused on finding solutions to these excrements, and in Memphis, outbreaks of cholera and typhoid fever were increasingly common. Thus, the advent of the automobile, whose only excreta are volatile compounds, was a revolution.[1].
Since its adoption in the early 20th century, the automotive industry has continued to innovate to provide an increasingly attractive and safe driver experience. Today's cars feature insulated and comfortable interiors, with features such as climate control and in-vehicle entertainment systems that bear little resemblance to what Benz devised. Seat belts, multiple airbags, and crash-absorbing body designs have increased safety. These systems have been supplemented by driving aids in which the user has to perform fewer and fewer actions.
This quest for comfort and safety is being translated into the development of autonomous vehicles, capable of transforming mobility once again. With the intention of enabling people to enjoy the journey without the need to drive, the industry is facing the challenge of creating cars that are capable of operating completely independently, opening the door to a new era of transport.
Vehicles can be classified according to their automation in 6 levels. The first, L0, encompasses all fully manual vehicles, while L5 would include those that are capable of navigating any road and situation without any human intervention. This type of vehicle is what the automotive industry aspires to, one that has no steering wheel or pedals and where the human only indicates their destination and arrives efficiently without having to take any action, just enjoying the journey. However, this technology presents both technological and logistical challenges including, among others, the presence of other vehicles, pedestrians and road situations that depend on a quick response.
In order to carry out these actions, autonomous vehicles are equipped with pieces of technology that far exceed human capabilities. In humans, it is the senses (mainly vision and hearing) that do the work of gathering information about what is happening on the road and transmitting it to the brain. There, it is processed by the neural networks and a response is issued that modifies the traffic flow, whether it is braking for a red light or a quick lane change to avoid an accident or a breakdown on the road. The response will depend on many factors, such as the driver's experience, state of mind, fatigue or possible distractions. However, in an autonomous vehicle, both data collection and processing are radically different.
To detect obstacles or events that occur while driving, autonomous vehicles use systems that far exceed human capabilities. Typically, vehicles use cameras, radar, LiDAR or any combination of these to gather information about their surroundings. Each of these technologies has its advantages and disadvantages, and their combination usually serves as a redundant mechanism to clarify a situation that may occur while driving.
In L0 vehicles these systems do not exist, as their operation is entirely dependent on humans, and L1 and most L2 vehicles do not generally rely entirely on these systems, with the driver still doing most of the work. But looking to the future[2], With L3 vehicles onwards, these devices and the applications associated with them will be increasingly present in our driving stock.[3]. Figure 1 gives an accurate picture of the functioning mechanisms of this complex ecosystem.
Cameras capture light from the sun (or any other source) that bounces off the surface of an object and enters the detector. In most cameras, this light is in the visible light spectrum, which is the same light that our eyes can see. In this way, like a human, artificial devices can see roadside obstacles or, using multiple cameras in different locations, triangulate the position in three dimensions of both static objects and those approaching the vehicle on potential collision paths. Cameras in autonomous vehicles are similar to those available in any current phone, although they tend to specialise in maintaining a delicate balance between resolution, power consumption and robustness, as they will face harsh climates, vibrations and potential bumps from insects and small stones while driving.
One of the biggest advantages of the machine vision technology that autonomous vehicles use is that they are not subject to the vicissitudes of biology, and to what evolution has selected to be most advantageous in nature. Any machine can be custom designed and built to fulfil a mission in the most efficient way possible. That is why, by applying ingenuity, the automotive industry has adapted technologies that previously had other uses to autonomous vehicles. By adding another type of sensor, capable of seeing light that is invisible to humans, vehicles can have an “enhanced eye” with abilities that far surpass any biological vision organ.
A clear example is infrared cameras, which detect the radiation that escapes from any object because it has a certain temperature. As the temperature rises, the heat exchange with the medium increases, resulting in a colour gradient, usually from black (cold) to white (hot). In this way, the vehicle can “see” in complete darkness at night, or behind a dense smoke screen, because each object, depending on whether it is exposed to the sun or covered with snow, has a certain temperature. This technology is especially useful for detecting humans, who have a temperature of around 35°C on their exposed skin, or the exhaust pipes of the vehicles around them, which are heated by the exhaust gases coming out of the engine.[4].
Both normal and infrared cameras are passive systems and therefore, as mentioned, detect the light or radiation emitted or reflected by objects. Therefore, the former depend on the presence of light in order to function, and the latter on the presence of temperature differences in their environment. If these conditions are not met, the cameras are blinded. Autonomous vehicles therefore use other systems that move the light source to the vehicle itself. A system that demonstrates how human ingenuity can find solutions to any obstacle it faces.
To understand how these systems work, let's do a little imagination exercise. Let's place ourselves in a valley formed by two mountain ranges. One of the walls is almost vertical, and the other offers a slope and an accessible path. We don't know how we got there, but of course we want to know where we are. So we lace up our shoes and start to climb the accessible slope to try to get to the top and see some trace of civilisation on the horizon. We start to climb, and climb, and climb, and climb. And, when we make a rest stop, we would like to know how far we are from the wall. We don't have any device other than a stopwatch and a pocket calculator, so we'll have to be creative to find out.
If we turn and look at the distant mountain wall, we can estimate how far away it is by emitting a scream. The sound travels through the air at approximately 343 metres per second, or in other words, every second it travels 343 metres. Therefore, by analysing the echo, or in other words, how long it takes for the sound to bounce off the wall and reach our ears again, we can get an idea of how far away the wall is. If it takes two seconds, the wall is exactly 343 metres away (since it takes one second to reach the wall and one second to return). If it takes more, or less, we can make a simple rule of three to find out the approximate distance.
This is, in a very simplified form, how active sonar works, a mechanism that has been widely used by ships and submarines to detect shoals of fish or to avoid running aground on the seabed. However, this method is not without controversy. The loud sounds have been responsible for scaring and disorienting many cetaceans, even causing their death by pushing them into the depths in their flight. Therefore, although this is a well-established system, other options are being explored.
On land, long-distance sonar is also not feasible as it could also disturb wildlife and human well-being. It is, however, used in some cases for autonomous parking systems. But for long distances, the automotive industry has focused on developing a similar system that, instead of emitting sound, emits different wavelengths of light. These detection systems are called «active» and include radar, if it uses microwaves, or LiDAR.[5], if it uses the infrared spectrum. In both cases, the sensors measure how long it takes photons to reach all points in the environment and how long it takes for each photon to return, thus creating a 3D map of the vehicle's surroundings. Both systems are also able to calculate the direction and trajectory of moving objects in the environment by making these measurements several times per second and applying vector calculations.
The difference between radar and LiDAR lies mainly in their resolution and detection distance. LiDAR can detect objects at close range with an accuracy of centimetres, even millimetres in some cases, whereas radar, due to its longer wavelength, can detect objects further away, but much less accurately. Radar can detect an approaching bulky object, but cannot distinguish between a car, truck or motorbike, whereas LiDAR will be able to distinguish them accurately. It is also particularly useful in anticipating stationary obstacles and irregularities in the road at close range. Finally, LiDAR is not constrained to a single frequency spectrum, but can use several ranges, thus combining resolution with detection distance. Thus, so far, it has established itself as the most attractive system for autonomous vehicles to sense their environment.
A final advantage of these systems has to do with the data protection of the people around us. Cameras are constantly taking images of buildings and people. This could prevent autonomous driving in sensitive locations, or violate data protection laws especially in the European Union. In contrast, radar and LiDAR create 3D point arrays that cannot be reconstructed into a precise image. Therefore, in this type of detection system it is impossible to distinguish the detail of a person's characteristic features, thus respecting their privacy.
The sensors we have discussed so far only detect information from the environment and translate it into the binary system. Ones and zeros. In order for this data to be useful, the on-board computers must analyse it and transform it into measurable figures with which to produce a response. To do this, vehicles contain so-called electronic control units, or ECUs, scattered throughout the bodywork. They are usually located near the sensors or the mechanisms they control. There, they process the information and send it directly to the response systems or to another ECU that specialises in response control.
The functionality of ECUs, and hence of sensors and feedback systems, depends in turn on the software installed in them. This software has become increasingly complex and sophisticated in order to anticipate all possible driving scenarios. Before a vehicle is put on the road, it is tested for thousands of kilometres both in closed spaces and in open traffic under the watchful eye of test personnel. The greater the vehicle's autonomy, the tougher and more severe the tests will be to ensure that it does not endanger other road users.
But in addition to the computing power of each vehicle, the near ubiquitous internet connection offers a greater ability to enjoy a safe and efficient journey. By connecting to the cloud, an autonomous vehicle can obtain road information instantly. It is then able to establish the fastest routes between two points taking into account the overall situation. The autonomous driver can access a wealth of information, as well as updates for their ECUs, without having to store it in their limited systems or visit a specialist workshop.
But internet connectivity is a two-way street. In addition to receiving information, an autonomous vehicle can also send geolocation data, as well as alarm data in the event of an accident, allowing authorities to respond as quickly as possible. These connections have their risks, as there may be vulnerabilities whereby hackers are able to activate or deactivate both sensors and response systems, creating a dangerous situation for the vehicle's occupants or other road users.
With this in mind, the majority of vehicles in the vehicle fleet are of class L0, L1 or L2. These vehicles offer protective systems for drivers, such as emergency braking, the ability to maintain a cruising speed, or to keep a certain distance from a vehicle and change lanes if necessary. But there are certain places where more advanced driving systems are taking to the streets.
As will be discussed in more detail in the next section, there are fully autonomous taxi companies in both the United States and China. The best known are those of the company Waymo, which has been operating a Jaguar I-Pace vehicle for years in California (Los Angeles and San Francisco), Arizona (Phoenix) and Texas (Austin). Also in the United States, in Las Vegas, the company Motional, which has a fleet of half a thousand Hyundai Ioniq 5 vehicles, has driven more than two million kilometres without an accident in which the vehicle was at fault. In these cities, the autonomous taxi service started out with a supervising driver, as the law required the vehicles to have a person at the wheel at all times. But following changes in the law, they are now unsupervised, and carry passengers around the city.
Due to these more permissive laws, more and more companies, such as Tesla, are joining the autonomous taxi business in the US. The automotive giant began its journey on 22 June 2025 with 10 taxis, albeit with limited success, as they have been involved in several accidents that are under investigation by the National Highway Traffic Safety Administration (NHTSA).[6]. This statement of intent is a clear pulse on China, where the companies Baidu, AutoX or WeRide have fleets of autonomous taxis in cities such as Wuhan, Shanghai, Beijing or Guangzhou.[7]. In addition, Xin Jinping has shown a clear interest in further expanding the fleet of autonomous vehicles to cars, trucks and buses in his major cities.
On the other hand, some cities have opted for the creation of specialised areas for testing these vehicles. This is the case of Ottawa, in Canada, where they have created L5[8], The new test zone is an industrial zone where autonomous vehicle companies can test their systems in driving situations as close to reality as possible. These tests are crucial to refine the software that controls the vehicles and reduce the number of accidents, which is already low.
According to Waymo, in the 35 million kilometres its vehicles have travelled in the last five years, the vehicles have been involved in 192 collisions.[9]. 18 of these collisions resulted in injuries, although the reports do not mention responsibility. These figures are better than those of their flesh-and-blood counterparts, which over the same distance travelled suffered more than 1,000 collisions, of which around 60 resulted in injuries. With the following improvements in technology, autonomous vehicles are expected to become even safer[10].
In Europe, regulations are also favouring the integration of autonomous vehicles into the vehicle fleet, especially focused on mobility in cities. The pioneers are those of countries such as Norway and Finland, where these vehicles, called e-Jest and e-ATAK, are already a regular part of their roads. Although they are not the only ones, there are several autonomous bus projects that have been developed in various EU countries and among them, also at the forefront, are projects developed in Spain.
In short, over the last few years, autonomous vehicles have burst onto the automotive fleet, ready to transform both mobility and public transport globally. This integration on our roads has been made possible by nations in the Americas, Asia and Europe, which have committed to both innovation and business models that were once only part of science fiction stories. Today, the technological race is the product of large investments, advances in different scientific fields and a risky but solid legislative commitment, the result of which is that the streets are seeing more and more driverless vehicles.
For the time being, to ensure that the technology continues to move in the right direction, the vehicles (with the engineering minds behind them) continue to collect and analyse real travel and incident data. In this way, they are getting better and better at consolidating the software that moves and steers vehicles and industries in the right direction. In addition, legislators and official bodies are also redoubling their efforts to bring regulatory frameworks into line with this new reality, where drivers have less of a role and therefore less responsibility in the event of an accident.
But all progress requires caution. The adoption of these technologies is not a sprint, but a transformation that, over the long term and with commitment from institutions and centres, will shape a new era of mobility.
IN ACTION. New business models inside and outside the car
While the end of the last decade predicted an unstoppable proliferation of autonomous vehicles over the next few years, as is often the case in the technology sector, the general feeling today is that numerous technical, regulatory and economic challenges and complexities will slow down their deployment. The market is demonstrating that the demand is there and the technology race, in any case, is not going to stop, especially in software and machine vision. So as the new artificial intelligence (AI) cycle unfolds its full capabilities, the industry's reaction could change. But experts agree right now that L2 and L2+ systems will dominate until 2035 due to their cost-effectiveness and regulatory compliance, while L3 adoption will remain limited, and L4 deployment will account for only around 4% of new personal vehicles in 10 years' time.[11].
Robotaxis have already demonstrated their technological feasibility and, as mentioned in the previous section, are being deployed on a large scale in cities in the United States and China. Figure 3 shows the growth forecasts by region. Both countries have more than 30 cities with testing, more than the rest of the world combined. The Asian giant is expected to adopt L2+ and L3/L4 vehicles more quickly, due to strong consumer demand, increased regulatory agility and the development of the innovative ecosystem. The Chinese government considers autonomous driving a strategic priority and by the end of 2024, 15% of new vehicles would incorporate the most advanced forms of ADAS (Advanced Driver Assistance Systems). Commercial operation of autonomous vehicles is regulated at the national level, but implementation is at the municipal level. Baidu's Apollo Go offers a fully autonomous payment service in Wuhan and, together with companies such as Pony.ai and WeRide, operates with some restrictions in limited areas of cities such as Beijing, Guangzhou, Shenzhen and Chongqing.
Meanwhile, by mid-2025, more than 1,500 robotaxis were operating commercially in five US cities, and that number was expected to grow to 35,000 by 2030.[12]. If these expectations are realised, they would generate $7 billion in annual revenues and capture approximately 8% of the US ridesharing market, up from less than 1% today.
Other countries are not standing still, and the map of actors compiled in Figure 2 shows the scale of the ecosystem. By 2035, there could be a large number of robotaxis in between 40 and 80 cities around the world. International expansion includes the United Arab Emirates, Japan and even Mongolia, but the challenge remains to move from highly controlled and supervised testing to commercial launch. The United Arab Emirates has set a target of 4,000 robotaxis in Dubai by 2030, Germany has a framework for the implementation of autonomous vehicles and the Automated Vehicles Act.[13] in the UK will enable commercial deployment from 2026. In the UK, this will unlock the potential of an industry with an estimated value of up to £42 billion and the capacity to create 38,000 additional skilled jobs by 2035.[14]. The UK law will require autonomous vehicles to meet safety standards at least as high as those of careful and competent human drivers, and to pass rigorous safety checks before being allowed on the roads.
The impetus comes not only from governments, but primarily from the very start-up companies that are emerging in the sector. Beyond General Motors' decision to withdraw funding for its Cruise autonomous taxi subsidiary at the end of 2025[15], With the launch of the first autonomous ride-hailing service in China, Chinese taxi companies are expected to be keen to expand outside their home country. With Baidu testing in Hong Kong and WeRide launching through Uber in Abu Dhabi, the race to dominate autonomous transport services globally has kicked off.[16].
Europe's wariness of this new market is striking, considering that the first vehicle sold to the public with L3 autonomy worldwide was introduced in Germany in 2022. This was made possible by the country's adoption of UNECE R157, a regulation already applied in more than 50 countries, which initially allowed the activation of artificial pilots for traffic jams up to 60 km/h and has extended it up to 130 km/h. In March 2025, legislation came into force in Switzerland that will allow autonomous driving on motorways, as well as driverless robotaxis under certain conditions.
Today in the EU, however, there is still a need to harmonise testing and policies for autonomous vehicles so that they do not stop at country borders. Indeed, the current disparity of regulations and different levels of progress pose challenges that require EU-wide harmonisation.[17]. The 29 signatories of a Letter of Intent at Digital Day 2017 agreed to launch 5G cross-border corridors.[18]. The European Commission's ambition is to build on them to deploy automated driving projects. 5G tests have been carried out on more than 1,000 km of motorways, including four cross-border corridors: Metz-Merzig-Luxembourg, Munich-Bologna via the Brenner Pass, and Porto-Vigo and Evora-Merida, both between Spain and Portugal. The unified EU rules, such as those set out in the General Security Regulation[19] 2019 and ADS specifications[20] 2022, lay the foundations for harmonisation. However, more needs to be done. Unlike the fragmented state-level approach of the US or the hierarchical mandates of China, Europe could turn its ability to align and cooperate into a competitive advantage.
Goldman Sachs Research's forecast indicates a compound annual growth rate for the sector of around 90% between 2025 and 2030. Delaney estimates that gross margins for an autonomous vehicle operator could reach 40-50% in the next three to five years, which would raise gross profits in the US to approximately $3.5bn by 2030. However, scaling fleets involves overcoming so many hurdles that euphoria has given way to a more realistic and moderate view. Extensive physical infrastructure is needed, with depots, maintenance centres and high-speed connectivity. So the number of cities suitable for mass deployment remains limited at present. Moreover, developing the software for robotaxis requires billions of dollars of R&D investment, in contrast to the costs associated with ADAS/AD (autonomous driving) technology, which many large OEMs are opting for today, and which are clearly more affordable.
At the 2025 Consumer Electronics Show (CES) in Las Vegas, Waymo CEO Tekedra N. Mawakana was one of the keynote speakers, and Volvo showed off technological pyrotechnics to promote its autonomous truck, which uses Aurora Innovation's driving system. But the general perception was that, rather than repeating the ambitious claims of yesteryear, both technology companies and traditional automakers are prioritising practical, market-ready solutions.[21]. Nearly all McKinsey respondents (96%) believe that strategic alliances between startups and ridesharing companies, ICT giants, large manufacturers and major users such as carriers will be crucial to the development of autonomous vehicles. The majority also believe that the North American market will be the most fragmented, with only 15% expecting it to be dominated by one or two players, in contrast to 38% who believe that this is precisely what will happen in the European case.[22]. Therefore, it is unlikely that car manufacturers, in general, will engage in a comprehensive development of the new sector, as this would drain the resources they urgently need for the transition to electric propulsion systems.[23].
On the other hand, as the industry scales up, costs could fall as the technology itself evolves, in key areas such as cameras. Each of Waymo's current vehicles is estimated to cost $150,000 to produce.[24]. The next generation of hardware will offer higher performance at significantly lower cost. Tesla, Wayve and other companies are taking the approach of developing their systems directly with more affordable hardware, using only cameras and mimicking humans. As a result, driving per kilometre is getting cheaper, and could drop from around 22 euro cents in 2025 to nine euro cents in 2040. Insurance costs could also fall from 31 to 14 cents per kilometre over the same period. Finally, autonomous vehicle companies will need fewer and fewer remote operators, who act as a safety net and provide virtual assistance: each could manage 35 vehicles in 2040, compared to 10 in 2030 and just three today.
For autonomous trucks, the evolution will be similar. The cost per kilometre could fall from €3.8 to €1.18 in 2030, while the cost of human-driven trucks will rise from €1.63 to €1.75 per kilometre, driven by higher wages for drivers. The EU seems, in fact, more focused on the integration of robotic shuttles (roboshuttles), autonomous buses and trucks with public transport systems than in the launch of commercial or private autonomous vehicles, although international borders between Member States still pose challenges for long-distance applications.
Freight transport is the most promising area for automation. Autonomous truck sales could account for up to 30% of total new truck sales in the US by 2035. The US is best positioned to lead adoption, especially on long- and medium-haul routes, due to its total cost of ownership (TCO) advantages and the pressing need to address the driver shortage. Currently, only a few autonomous trucks are being deployed in the Permian Basin and the US state of Texas, but their number could reach 25,000 vehicles by 2030, which would still represent less than 1% of the total existing commercial truck fleet. The turnover for autonomous freight transport would then be about $18 billion out of a total market for the road freight sector of about $660 billion.[25].
Globally, an additional half a million autonomous L4 truck drivers are expected to be reached, while the remaining drivers will take on supervisory roles or continue to work in less developed regions. Europe could also reap strong benefits via TCO for long and medium haul, but needs a pan-European approach to quickly scale operations and reach the 26% in new truck sales in 2035 that experts forecast. The focus on the continent is on high-quality L4 fixed routes for autonomous short-haul (300-700 km) transport. Sweden's Einride is planning to implement them in the UK, Norway and Sweden, requiring companies to set up ancillary facilities along the way.
China could follow an even slower pace of autonomous truck adoption, because its TCO is lower, unless its government decides to intervene more clearly. It has companies such as Inceptio and DeepWay that have already sold more than 1,000 units. The former has established long-term partnerships with more than 13 entities and had reached 100 million kilometres by May 2025. However, it has achieved this after a strategic shift in its operations from two L4 drivers to a single driver (L2+/L3) and expanded its routes to 800-1,000 km. The result has been a reduction in driver labour costs of up to 40%, and an increase in revenue of approximately 8%.
The development of robotic shuttles and autonomous buses, which also offered very encouraging prospects, has stalled. The former were once seen as the ideal solution for last-mile transport, with more than 25 companies vying for primacy in the sector. However, insufficient funding is preventing many companies from moving beyond small-scale trials, and has revealed an unexpected lack of public interest, essential for establishing a viable business model in this sector. Companies that continue to invest in the technology include WeRide in Singapore, QCraft in 10 cities in mainland China, Pix Moving in 16 cities around the world, and Hyundai in South Korea and the UK. In May 2024, the Renault Group presented its autonomous vehicle strategy.[26] and launched, for the first time in Europe, together with WeRide, a road experiment with two autonomous minibuses during the Roland-Garros tennis tournament. They covered 1,000 kilometres and transported almost 700 people. Further experiments are already underway across Europe, including in Zurich (Switzerland), Valence (France) and Barcelona.
As we discussed in the first chapter, the expansion of autonomous vehicles is supported by good accident figures, which claim 1.2 million lives on the roads. Social media shows accidents involving freak robotaxis, such as the one involving a trumpet orchestra in Waymo's car park in August 2024. But the California Department of Motor Vehicles (DMV) keeps records of all collisions with autonomous vehicles and the reality is that between January 2019 and July 2024, 600 reports were filed and only 29 times could blame be attributed to an artificial driver. In 95% of the cases[27], In the case of the 571 human-caused crashes, the responsibility was human, a widely accepted percentage for human-caused traffic accidents, while the remaining 5% is attributed to other causes, such as mechanical failure, wild animals jumping onto the road and unintended accidents. Furthermore, in many of the 571 human-caused crashes, the autonomous vehicle was not even driving.
Robotaxis in California drove 5.3 million kilometres without a safety driver in 2023 and can only be held liable for one collision every 482,000 kilometres, compared to 322,000 kilometres for US drivers nationwide, rising to once every 161,000 kilometres in San Francisco. A study by Swiss Re of more than 40 million kilometres of self-driving Waymo vehicles has revealed that they were subject to approximately 90% fewer insurance claims compared to human-driven vehicles, even the most advanced ones.[28]. It is estimated that when the market penetration rate of vehicles is 10%, they can reduce the risks of vehicle accidents and injuries by 50%; when the penetration rate is improved to 50%, the risks can be reduced by 90%.[29].
Sensor data sets are also useful for determining liability.[30]. Bosch, Continental AG, Denso Corporation, Veoneer, Valeo, Hella, Aptiv, Panasonic, ZF Friedrichshafen AG, Hitachi, Velodyne, Shenzhen Anzhijie Technology, Ibeo Automotive Systems, Ouster, Quanergy Systems, LeddarTech, Luminar, Hesai Tech or Leishen.[31], are some of the technology leaders in the field of sensors for autonomous vehicles, and the list is growing. As mentioned in the first section of the chapter, LiDAR, which uses light to measure distances between objects, has long been considered the most accurate of the sensing technologies, providing high-resolution three-dimensional maps in virtually all weather conditions.[32]. Its balance in emergency braking, pedestrian detection and collision avoidance is second to none. However, it is priced three to five times higher than radar, which is increasingly positioning itself as an alternative.
The recent change has to do with radar resolution issues. Now, it is almost on a par thanks to the larger aperture of the latest radar, which also features echolocation and applies the principle of time-of-flight measurement, as does LiDAR. In doing so, it creates point cloud images of the vehicle's surroundings. To enable the new functions, the radar systems switched to multiple-input multiple-output (MIMO) antenna arrays that enable high-resolution mapping.
However, beyond the physical risks, there are other types of cyber threats that could also affect the expansion of autonomous vehicles. If they are attacked by hackers, normal driving behaviour could be altered and cause security problems.[33]. The most serious potential attacks are expected to be those targeting vehicles' global navigation satellite systems and those that seek to confuse vehicles' operating systems with false information, to direct them to incorrect destinations or along unstable driving routes. To a lesser extent, there are concerns about possible data leakage. By recording everything that happens in and around them, in order to provide personal shared mobility services, autonomous vehicles have an impact on privacy. They record and evaluate user data, including frequency of service use, average travel time or frequently visited locations, and identify users through mobile phone numbers, facial recognition and fingerprints.
The technological advances that will be hitting the market in the coming years are also working in favour of autonomous driving. In the case of the new generative AI (GenAI) cycle, end-to-end (E2E) models are replacing traditional rule-based systems that struggle to handle the complexity of real-world driving. By creating synthetic data, GenAI significantly aids the training of autonomous systems, as real-world data collection is costly and incomplete. AI is also strengthening human-machine collaboration through improved driver monitoring systems (DMS) and in-vehicle interfaces (HMI), which enable more intuitive voice commands and adaptive controls.
Developing the right AI algorithms, however, requires billions of euros of R&D investment and must be planned and adapted to each city's unique driving environment. In addition, in the European case, AI systems implemented in or in connection with autonomous vehicles are classified as high-risk AI systems under EU AI law if they affect driving and passenger safety. This requirement will, in any case, be implemented through the Type Approval Framework Regulation (TAFR).[34] and the General Security Regulations (GSR)[35], Both rules form part of the sectoral legislation, which is amended for this purpose by the IA Act.[36].
In the US, there are no national standards or guidelines, allowing states to determine their own. Product liability laws now come into play when accidents occur and their interpretation depends on multiple factors, especially whether the vehicle was operated properly for its level of automation. China has issued Draft Proposed Amendments to the Road Safety Law that regulate testing and approval requirements, as well as the allocation of liability when accidents or violations occur.[37].
Current autonomous driving technologies, especially those related to perception and prediction, have benefited greatly from advances in machine vision.[38]. However, challenges remain in many dimensions, such as managing complex and rapidly dynamic environments, explaining decisions and following human instructions. Artificial eyes often fail to understand context and this limits progress towards more advanced autonomous driving. The emergence of LLMs (extended language models) and VLMs (vision-language models) offer possible solutions. For example, current large-scale vision models often contain billions of parameters, making both fine-tuning and inference resource-intensive and not compatible with real-time requirements. Parameter-efficient fine-tuning technology (PEFT) could reduce the number of trainable parameters.
In terms of autonomous vehicle decision management, the optimal formula will be a hybrid: the vehicle should take critical actions in real time, while data analysis and model updates should remain in the cloud. Issues such as software architecture and hardware requirements for the implementation of large VLMs still need to be solved, considering that it is impossible to determine the specific cause of a traffic accident from a single image. In addition, the hallucination phenomenon of autoregressive large language models poses serious challenges for practical applications.[39] and fuels the ethical debate on the attribution of liability in the event of an accident.
The degree of sophistication in the use of AI will have an impact on the impact of autonomous vehicles on the environment. Green driving is expected to reduce energy consumption by up to 20%. Improvements in collision avoidance, due to increased safety features, can lighten the weight and size of autonomous vehicles, which could reduce fuel consumption by 5 to 23%.[40]. Right-sizing the vehicle, i.e. adapting it to certain needs and uses, could also add another 21-45% of savings. Paradoxical as it may sound, autonomous driving could increase vehicle flow by approximately 15%, while reducing traffic congestion by 30%.[41], emissions, with a consequent reduction in fuel consumption of up to 4%. Some believe, however, that this will become an incentive to increase the number of vehicle kilometres travelled, bringing us back to square one. Finally, smart parking models will also contribute to fuel savings by minimising the time it takes to search for free spaces.
In the wake of the UK's Automated Vehicles Act coming into force in spring 2026, the collateral aspects of the expansion of a technology that forces a rethink of many of our current paradigms have been analysed. For example, the elimination of the need for human drivers.[42] and the corresponding jobs could significantly reduce labour costs.[43]. Also, autonomous vehicles could operate 24/7, which transforms the entire logistics landscape and could lead to, among other things, an increase in demand for e-commerce.[44].
On a social level, autonomous vehicles offer substantial potential for improvement for people who are underserved by the current public transport system or who suffer from mobility barriers. In the UK, 25% of the population have mobility difficulties, are afraid to use the tube and do not travel by car so as not to drive alone, and are virtually condemned to stay at home. Many cities, even those with more than 20 million inhabitants, cannot afford city-wide rail systems and rely on small public service buses. In Germany, it is estimated that 80,000 bus drivers will be needed by 2030. The old technology of rail systems (metro, tram, light rail, etc.) could be replaced by 24/7 services at a base cost of only half that of current public transport systems.[45].
Shared mobility allows consumers to access different modes of transport according to their needs.[46]. As opportunity costs decline, people may choose to live further away from urban centres, which could increase the value of peripheral properties and drive urban sprawl. Businesses and individuals will find it easier to interact between locations and benefit from the spatial concentration of agglomeration economies.[47]. This will require thinking about changes to existing urban planning and infrastructure, particularly digital infrastructure that must enable designated lanes, intelligent traffic systems and charging structures.
In the future, during autonomous driving, passengers are expected to be able to perform a wide variety of activities that are not currently possible in the vehicle, such as watching movies, drinking, using laptops and shopping. Relevant companies in the ICT sector, the entertainment industry, retail or the means of payment sector online are already analysing in depth the opportunities opened up by this new platform, which has been in place for long periods of time. captives to potential users. At the same time, the design of vehicle interiors is undergoing a major overhaul: new immersive possibilities for virtual and augmented reality integrated into windows and dashboards are converging at the operating table with previously unthinkable furnishings such as work tables, new access systems and the disappearance of components that may become meaningless, from pedals to rear-view mirrors.
Technology companies' penetration of the shared mobility market[48], The European Commission is promoting an alliance for manufacturers to share technologies to foster the development of autonomous vehicles, in parallel to their support for electric vehicles, with solutions for sensing, artificial intelligence, communication, navigation and high-definition geospatial data, as well as software systems, chips and graphics processing units (GPUs) for vehicles, and entertainment applications. The European Commission is promoting an alliance for manufacturers to share technologies to foster the development of autonomous vehicles, in parallel to its support for electric vehicles.[49], and aims to strengthen supply chains, especially in the area of batteries for electric vehicles.[50]. It wants to avoid relying on existing suppliers, especially after the bankruptcy of the main national battery developer, Northvolt.
IN SPAIN. An industry in trials, pending regulation
The Road Safety Strategy 2030 defines the intention for Spain to become a testing laboratory for autonomous vehicles, “and to achieve significant progress in regulation and the necessary basic conditions, both in vehicles and on the roads, in order to progressively implement safe automated driving”, as stated in the draft Royal Decree for fully automated vehicles, which is currently being approved. The government envisages the creation of an office or one-stop shop for the management of test applications, as well as a certification system for autonomous vehicles, based on the accreditation of compliance with the precepts of the General Traffic Regulations.
In June 2025, the Automated Vehicle Safety and Technology Assessment Framework Programme (ES-AV Programme) was unveiled.[51], which establishes a national code for tests and operations with automated or remotely driven vehicles, from prototypes to pre-homologation. The objective is to advance in the definition of the safe circulation and certification policy, and the ES-AV Programme Management Centre (CG-ESAV), belonging to the Subdirectorate General of Mobility Management and Technology of the Directorate General of Traffic, has been configured as the body in charge of managing authorisations and admissions.
Proposals can be addressed to the Office for the Facilitation of Automated Vehicle Testing on Public Roads (OFVA). The main data of the tests and operations that have received authorisation are already available: E-BUSKAR, for example, tested an autonomous bus in Leganés between January and February 2025 and travelled 279.20 km; Renault tested two shuttle in Barcelona in March and accumulated 141.65 km; ALSA is carrying out ongoing tests with a shuttle at the Cantoblanco campus (UAM), with a planned duration of two years; and the CTAG began trials in Seville in May, also with a shuttle, which are still ongoing.
However, autonomous driving experiments have been going on for years in our country. In 2022, the Galician CTAG had participated in tests in the framework of the European 5G-MOBIX project, focused on 5G technology for advanced Connected and Automated Mobility. The experiments took place on the Tui-Valença corridor, involved a shuttle autonomous electric 100% developed by the Spanish centre and addressed speed adaptation (to avoid a pedestrian, for example) and remote driving in a critical situation.
Through the first call of the Electric and Connected Vehicle (EV) grant, three projects have been financed.[52] focused on autonomous driving and connected mobility, led by Renault Group, Ficosa and another by Avanza Zaragoza. However, the total amount is only 14.5 million euros, which makes it difficult to mobilise a change in the production model. Spain has five corridors where these technologies can be tested. Applus+ IDIADA in the province of Tarragona, which specialises in the validation of communication technologies and control systems in the ADAS/CAV circuit, including protection against cyber-attacks, also stands out among the companies that are positioned as benchmarks in the development of autonomous vehicles in our country. Germany's Bosch division in Spain is also actively participating in projects focusing on advanced sensors, radars, cameras and AI, with a view to achieving Level 3 autonomy. GMV and Indra are developing key technologies in the R3CAV project, while Abertis, through the Future World project, and MASERMIC, a company from Guipuzcoa, are also involved in this sector.
Currently, the main barrier to the circulation of autonomous vehicles in Spain is the lack of a regulatory framework that allows the operation of Level 4 or higher, despite technological advances, according to ANFAC.[53]. Regulatory compliance is not only a matter for our country, but must be addressed at European level and with technological integration strategies that reduce the impact of borders. The majority of the commercial offer of light vehicles, both passenger and goods vehicles, is concentrated in automation level 2. It is present in 81% of the passenger car models available and in 57% of the light commercial vehicle models. The collision warning system for pedestrians and cyclists (97.9%), although not mandatory for light vehicles, the emergency lane keeping system (ELKS) (96.1%) and the intelligent speed assistant (ISA) (94.6%) are particularly noteworthy. For freight transport, 71% of the commercial vehicle models on offer have a level 2 autonomy, while for buses, the majority of the commercial offer is still concentrated on automation level 1.
[1] Martínez, M. (2022) The Horse Dung That Redefined Architecture - Ferrovial blog, Ferrovial. Disponible en:: https://blog.ferrovial.com/es/2022/10/estiercol-caballo-redefinio-arquitectura/#:~:text=Y%20lo%20mismo%20ocurr%C3%ADa%20en,caballos%20de%201300%20establos%20diferentes. (Consultado el 20/06/2025).
[2] Litman, T. (2025) Autonomous Vehicle Implementation Predictions, Victoria Transport Policy Institution. Available at: https://www.vtpi.org/avip.pdf (Accessed 20/06/2025).
[3] Autonomous vehicles: Timeline and roadmap ahead (2025) World Economic Forum. Available at: https://www.weforum.org/publications/autonomous-vehicles-timeline-and-roadmap-ahead/ (Accessed 20/06/2025).
[4] Rangwala, S. (2022) Thermal cameras gain acceptance for Adas and autonomous cars, Forbes. Available at: https://www.forbes.com/sites/sabbirrangwala/2022/10/28/thermal-cameras-gain-acceptance-for-adas-and-autonomous-cars/ (Accessed 20/06/2025).
[5] Saveliev, V. (2025) ‘New tendencies in LIDAR technology for Autonomous Vehicles and other applications’, Autonomous Systems: Sensors, Processing, and Security for Ground, Air, Sea, and Space Vehicles and Infrastructure 2025., p. 2. doi:10.1117/12.3053382. (Accessed 20/06/2025).
[6] Edwards, C. (2025) US safety regulators contact Tesla over erratic robotaxis, BBC News. Available at: https://www.bbc.com/news/articles/cg75zv4gny2o (Accessed 25/06/2025).
[7] Rose, I. (2025) The slow but steady advance of driverless vehicles, BBC News. Available at: https://www.bbc.com/news/articles/cly41yx9w88o (Accessed 20/06/2025).
[8] Symonds, D. (2019) Mayor launches Ottawa L5 autonomous vehicle test track, ADAS & Autonomous Vehicle International. Available at: https://www.autonomousvehicleinternational.com/news/testing/mayor-launches-ottawa-l5-autonomous-vehicle-test-track.html (Accessed on 20/06/2025).
[9] Martínez, P. (2024) Driverless Taxis Face Full Spread Across the US, and the ‘culprit’ is a Spaniard, elconfidencial.com. Available at: https://www.elconfidencial.com/motor/nueva-movilidad/2024-10-10/hyundai-ioniq-waymo-taxi-autonomo-robotaxi-jaguar-zeekr-jose-munoz_3979422/ (Accessed on 20/06/2025).
[10] Abdel-Aty, M., Ding, S. A matched case-control analysis of autonomous vs human-driven vehicle accidents. Nat Commun 15, 4931 (2024). DOI: 10.1038/s41467-024-48526-4
[11] “Autonomous Vehicles: Timeline and Roadmap Ahead”, World Economic Forum / Boston Consulting Group, 24 April 2025.
[12] “The Autonomous Vehicle Market Is Forecast to Grow and Boost Ridesharing Presence”, Goldman Sachs, 3 July 2025.
[13] www.legislation.gov.uk/ukpga/2024/10/contents/enacted
[14] “Self-driving vehicles set to be on roads by 2026 as Automated Vehicles Act becomes law”, UK Government, 20 May 2024.
[15] news.gm.com/home.detail.html/Pages/news/us/en/2024/dec/1210-gm.html
[16] Graham Hope, “Autonomous Vehicles: The Outlook for 2025”, IoT World Today, 27 December 2024.
[17] “Mastering mobility: autonomous vehicles and EU-alignment”, EIT Urban Mobility, 6 December 2024.
[18] digital-strategy.ec.europa.eu/en/policies/cross-border-corridors
[19] eur-lex.europa.eu/eli/reg/2019/2144/oj
[20] eur-lex.europa.eu/eli/reg_impl/2022/1426/oj
[21] Jeremy Carlson, Brock Walquist, “A More Practical Future for Autonomous Vehicles”, S&P Global, 31 January 2025.
[22] Derek Chiao et al. “Autonomous vehicles moving forward: Perspectives from industry leaders”, McKinsey, 5 January 2024.
[23] “Autonomous driving”, Roland Berger, 30 October 2024
[24] Clément Chamboulive, “AI-powered autonomous driving: global expansion and regulatory support”, Robeco, 5 March 2025.
[25] Goldman Sachs, 2025
[26] media.renaultgroup.com/autonomous-vehicles-for-public-transportation-the-growing-interest-of-local-authorities-and-mobility-operators-in-renault-groups-approach/
[27] James Jeffs, “Autonomous Vehicle Industries Now and in 10 Years”, IDTechEx, 20 November 2024.
[28] Di Lillo, L. et al., Do Autonomous Vehicles Outperform Latest-Generation Human-Driven Vehicles? A Comparison to Waymo's Auto Liability Insurance Claims at 25.3M Miles, Swiss Re / Waymo, December 2024.
[29] Lin Tu, Min Xu, An Analysis of the Use of Autonomous Vehicles in the Shared Mobility Market: Opportunities and Challenges, Sustainability, 8 August 2024, doi.org/10.3390/su16166795
[30] Amolika Sinha et al. Crash and disengagement data of autonomous vehicles on public roads in California, Sci Data, 23 November 2021, 10.1038/s41597-021-01083-7.
[31] www.linkedin.com/pulse/global-autonomous-vehicle-sensors-market-growth-future-outlook-nhdkf/
[32] Stefan Dobler, Victor Kondel, “LiDAR And Radar Battle For Autonomous Vehicle Turf”, Oliver Wyman, July 2023, accessed on 11/07/2023.
[33] Han Wu et al. Adversarial Driving: Attacking End-to-End Autonomous Driving, 2023 IEEE Intelligent Vehicles Symposium (IV), 27 July 2023, DOI: 10.1109/IV55152.2023.10186386
[34] eur-lex.europa.eu/legal-content/EN/TXT/PDF/?uri=CELEX:02018R0858-20230730
[35] eur-lex.europa.eu/legal-content/EN/TXT/PDF/?uri=CELEX:02019R2144-20220905
[36] Osman Gazi Güçlütürk, Bahadir Vural, “Driving Innovation: Navigating the EU AI Act's Impact on Autonomous Vehicles”, Holistic AI, 20 March 2024.
[37] www.china.org.cn/china/2021-04/04/content_77376009.htm
[38] Xingcheng Zhou et al. Vision Language Models in Autonomous Driving: A Survey and Outlook, IEEE, 16 May 2024, 10.1109/TIV.2024.3402136.
[39] L. Huang et al., A survey on hallucination in large language models: Principles, taxonomy, challenges, and open questions, ACM Transactions on Information Systems, January 2024, doi.org/10.1145/3703155
[40] css.umich.edu/sites/default/files/2024-10/Autonomous%20Vehicles_CSS16-18.pdf
[41] Xujun Wang, An Analysis of the Future Trends and Challenges of Autonomous Driving Technology, Proceedings of the 2nd International Conference on Machine Learning and Automation, 8 November 2024, doi.org/10.54254/2755-2721/104/20241087.
[42] Alexandros Nikitas, Alexandra-Elena Vitel, Corneliu Cotet, Autonomous vehicles and employment: An urban futures revolution or catastrophe?, Cities, July 2021, doi.org/10.1016/j.cities.2021.103203
[43] Jeremy Webb, The future of transport: Literature review and overview, Economic Analysis and Policy, March 2019, doi.org/10.1016/j.eap.2019.01.002
[44] Heleen Buldeo Rai, Sabrina Touami, Laetitia Dablanc, Autonomous e-commerce delivery in ordinary and exceptional circumstances. The French case, Research in Transportation Business & Management, December 2022, doi.org/10.1016/j.rtbm.2021.100774
[45] Anja Huber et al. “Autonomous vehicles: The future of European transport?”, McKinsey, 23 June 2025.
[46] Lin Tu, Min Xu, An Analysis of the Use of Autonomous Vehicles in the Shared Mobility Market: Opportunities and Challenges, Sustainability, 8 August 2024, doi.org/10.3390/su16166795
[47] Daniel J. Graham, Stephen Gibbons, Quantifying Wider Economic Impacts of agglomeration for transport appraisal: Existing evidence and future directions, Economics of Transportation, September 2019, doi.org/10.1016/j.ecotra.2019.100121
[48] León, L.F.A., Aoyama, Y. Industry emergence and market capture: The rise of autonomous vehicles. Technol. Forecast. Soc. Chang., July 2022, doi.org/10.1016/j.techfore.2022.121661
[49] Nick Flaherty, “EU alliance to push driverless vehicle technologies, large scale pilots”, eenewseurope.com, 3 March 2025.
[50] Chris Rosamond, “Europe proposes big push on self-driving vehicles, and measures to boost EV demand”, autoexpress, 3 March 2025, accessed 10/07/2025
[51] www.dgt.es/muevete-con-seguridad/vehiculos-seguros/conduccion-automatizada/marco-pruebas-vehiculos-automatizados/
[52] Diego Estebanez, “España acelera en la carrera por el coche autónomo a la espera de cambios en la normativa”, El Economista, 15 August 2025.
[53] “Barómetro Vehículo Autónomo y Conectado 2024”, ANFAC, November 2024.