Contenido

Existe cierta incoherencia en la terminología utilizada en el sector de los coches autónomos. Varias organizaciones han propuesto definir un vocabulario preciso y coherente.

SAE International, un organismo de normalización de la automoción, publicó en 2014 un sistema de clasificación con seis niveles -que van desde los sistemas totalmente manuales a los totalmente automatizados- con el nombre de J3016 Taxonomy and Definitions for Terms Related to On-Road Motor Vehicle Automated Driving Systems. Esta clasificación se basa en la cantidad de intervención del conductor y en la atención requerida, más que en las capacidades del vehículo, aunque están vagamente relacionadas. En Estados Unidos, en 2013, la Administración Nacional de Seguridad del Tráfico en Carretera (NHTSA) publicó un sistema de clasificación formal, pero lo abandonó en favor de la norma SAE en 2016. También en 2016, SAE actualizó su clasificación, denominada J3016_201609. La clasificación es objeto de controversia,[16]​ con diferentes enfoques propuestos para una expansión.[17]​.

Driving automation levels

Level 0: The automated system issues warnings and can intervene momentarily, but does not have sustained control of the vehicle.

Level 1: The driver and the automated system share control of the vehicle. Examples of this are systems in which the driver controls the steering and the automated system controls the engine power to maintain a certain speed (cruise control) or the engine and brake power to maintain and vary the speed (adaptive cruise control or ACC); and park assist, in which steering is automated while speed is under manual control. The driver must be prepared to regain full control at any time.

Level 2: The automated system takes full control of the vehicle: acceleration, braking and steering. The driver must supervise driving and be prepared to intervene immediately at any time if the automated system does not respond correctly. Contact between hand and steering wheel is usually mandatory during SAE 2 driving, to confirm that the driver is ready to intervene. The driver's eyes can be monitored by cameras to confirm that the driver is maintaining their attention in traffic.

Level 3: The driver can safely divert his or her attention from driving tasks, for example, the driver can text or watch a movie. The vehicle will handle situations that require an immediate response, such as emergency braking. The driver must continue to be prepared to intervene within a limited time, specified by the manufacturer, when the vehicle asks him to do so. This automated system is sometimes defined as a co-pilot that warns in an orderly manner when manual driving is necessary.

Level 4: As level 3, but driver's attention is never required for safety, for example, the driver can safely go to sleep or leave the driver's seat. However, self-driving is only supported in limited spatial areas or under special circumstances. Outside of these areas or circumstances, the vehicle must be able to safely abort the trip, for example, slowing down and parking the car, if the driver does not regain control. An example would be a robot taxi or a robot delivery service that covers selected places in an area, at a specific time and in specific quantities.

Level 5: (Also called "optional steering wheel"): No human intervention required. An example would be a robotic vehicle that works on all types of surfaces, all over the world, all year round and in all weather conditions.

Lucia Errandonea, creator of Algoritmic Societies,[18]​ assures that to know how a car moves and decides, we first have to understand how we move. Humans mix our senses through the knowledge we acquire to react to our environment, and we combine all of this with the motor system, which gives way to mobility. It says, "All of these systems are coordinated with the nervous system, which gives orders and makes decisions," which may be more difficult for an intelligent automated system to combine.

Furthermore, Errandonea points out that the fact that a machine cannot interpret the environment leads to creating different dilemmas and controversies. As could be the tram dilemma, which depicts what an autonomous car should do if it finds itself in the situation of saving one person or another. These autonomous cars are referred to as commercial cars, corresponding to the first level of automation. Another dilemma that I mentioned is "false marketing" about the level of automation.

Therefore, with this we conclude that the artificial intelligence that makes cars autonomous is not mature enough. As Julián Estévez, a researcher in Robotics and artificial intelligence, says, he talks about how difficult it is to get cars with artificial intelligence to be accurate enough for correct driving. Above all, because it is necessary to insert these autonomous systems into society, Estévez assures, "there is a problem between the coexistence of autonomous and non-autonomous cars" and hence the proposals to separate the circulation environments of each one, proposing ideas such as: having them in different lanes, different areas of the city or different schedules. Therefore, apart from being a technological problem, it is also the approach to cities and how we have to move.

The earliest known presentation of an autonomous vehicle was by Norman Bel Geddes at the Futurama trade show "Futurama (exhibition)") sponsored by General Motors for the 1939 World's Fair, which consisted of an electric vehicle that was controlled by an electrical circuit embedded in the road pavement.

In 1980, a vision-guided van from Mercedes-Benz, designed by Ernst Dickmanns") and his team from the University of Munich reached 100 km/h on streets without traffic. The European Commission was interested in the project and made an investment of €800 million for the EUREKA Prometheus project that sought to develop an autonomous vehicle.

In 1980, the Defense Advanced Research Projects Agency (DARPA) achieved the first vehicle that operated using radar, laser, and computer vision. In 1987, HRL laboratories demonstrated that a vehicle could be built that could design its own route once it left the map. The vehicle was able to move more than 600 meters through complex terrain such as slopes, large rocks and vegetation.

In 1994, two twin robot vehicles VaMP and Vita-2") from Daimler-Benz and Ernst Dickmans drove alone for more than a thousand kilometers on a 3-lane Paris motorway, on regular days and in heavy traffic at speeds of 130 km/h. The vehicles operated autonomously with little human intervention. They demonstrated that driving in free lanes, in convoy, and changing lanes to the right and left could be performed autonomously.

In 1995, Dickmanns' team modified a Mercedes-Benz S-Class to make a trip between Munich and Copenhagen and back, using computer vision with saccades and a computer to react in real time. The robot managed to reach speeds greater than 175 km/h on the German Autobahn, with an average time between human interventions of 9 km, which represented 95% autonomous driving. Again the vehicle drove in traffic, executing maneuvers to overtake other vehicles.

The pioneering work in Spain was developed by the CSIC.[19][20] Subsequently, the first autonomous driving tests in Spain were carried out in 2012 on a 100km route from El Escorial to Madrid carried out by the AUTOPIA group of the Center for Automation and Robotics. A manual vehicle dynamically generated a high-precision map to be followed by the fully autonomous tracking vehicle. The trip covered a wide range of driving scenarios, including urban areas, secondary roads and motorways, in standard traffic conditions.[21]​.

For its part, Audi announced in 2014 that its autonomous RS7 model reached 240 km/h on the Hockenheim circuit in Germany. The car completed one lap of the track in just over two minutes. Audi also placed a human behind the wheel for a comparison lap, this being five seconds slower than the unmanned car.[22]​.

In April 2015, a self-driving Audi SQ5, using an autonomous driving system developed by Audi and Delphi, managed to travel 5,400 kilometers in 9 days without the engineers inside the car touching the steering wheel except on a couple of occasions. Weeks later, a fleet of 25 autonomous vehicles designed by Google left the test tracks to be evaluated in the daily traffic of Mountain View (California), California, without passing 40 km/h. Google's prototype has the navigation software that it already used in its fleet of Lexus cars, which traveled more than a million kilometers autonomously and under the supervision of a human at the wheel.[24]​.

In the summer of 2015, the University of Michigan launched a prop village, called MCity (in:)"), in whose streets autonomous vehicles can be tested. The place has a street more than a mile long, curves of different radii, roundabouts, traffic lights, pavements of different surfaces, etc.[25]​ On the other hand, there are other projects that will complement MCity. One of them is the installation of 9,000 interconnected vehicles in the large Ann Arbor area, as well as another 20,000 on roads in the southeast of the state.[26] During the same week, in Buenos Aires (Argentina), representatives of the local government tested two units of the first prototype of an electric autonomous vehicle manufactured in that country. It is a vehicle without a steering wheel or pedals.[27]​.

The first thing we have to take into account when talking about autonomous driving is the fact that we are dealing with a machine, a large computer with wheels that makes decisions based on unquestionable guidelines.

In this way, on a road with a speed limit of 80 km/h, the autonomous car will never exceed that limit unless its predefined operating parameters are altered. Ambiguity does not exist in machine decision making.

If we stick to the theory, the autonomous car is the perfect figure as a transportation system, where its operation is based on compliance with imposed laws and guidelines. There is no interpretation, much less violation of the rules unless there is a ruling. That said, the figure of the autonomous car is understood as the best advance in safety that has occurred in recent years, since its operation results in a control and prudence that is impossible for a human being to achieve.

In 2021 and 2022, the terminologies “automated driving system” and “automated driving” appear in article 1 and new article 34 bis of the 1968 Convention on Road Traffic").[28]

The invention, development and dissemination of the autonomous vehicle is producing enormous positive effects (road safety, accessibility, energy efficiency, etc.) and at the same time causes multiple challenges.

Among the challenges caused by this technology are the changes that the automobile industry will face (new companies enter the sector), insurance (fewer accidents could mean a lower cost of "Premium (insurance)" premiums and a new definition of liability in the event of an accident), the labor market (many professional drivers such as taxi drivers or truck drivers and employees of mechanical workshops will lose their jobs while more robotics experts will be needed) or the supply of organs for donation (a significant decrease in the number of deaths could mean a big drop in potential donors).

Software Components

To achieve truly automatic driving in an urban situation with unpredictable traffic, many real-time systems are necessary, which must interoperate. For example, a system of location, perception of the environment, planning and, logically, a control system is necessary. In addition, a set of sensors are necessary that collect and provide the necessary information to make decisions.

When we think about a control system for an autonomous car, it is easy to fall into the temptation of imagining that the algorithms that control it resemble classic programming rules, such as “if (…) then…else…”, for example “If a pedestrian crosses the middle of the street, then the speed will be reduced, braking if necessary.” It is evident that this way of thinking is not appropriate since the number of possible situations is very high and therefore a system capable of generalizing is necessary.

Instead of implementing countless rules to recognize these objects, it is much more practical to implement a machine learning algorithm that is “trained” with different images that exemplify all possible situations.

Each of the images is associated with the type of vehicle it contains. The algorithm begins to process the images. Initially you try to guess which vehicle is in each image and you will be wrong very often at first. Since it knows “the solution”, that is, which vehicle is actually in each image, it modifies and adapts internal parameters and tries again. The process continues to iteratively reduce the failure rate. Later, when you are presented with new images you will be able to classify them correctly. We can, then, affirm that the algorithm has learned.

That same approach can be used for decision making and evaluation. Instead of providing a list of rules with which to evaluate the action to take for each situation, an algorithm is trained with traffic situations in which the correct action to take is specified. As before, the algorithm tries to guess the correct action and modifies internal parameters depending on whether it is wrong or right.

Another technique used in autonomous driving algorithms is probabilistic models. To make decisions, the car needs to predict its own position and also that of all the objects it identifies, such as pedestrians or other vehicles, since they are moving. Since it does not know the position they will be in, it tries to predict it using probability distributions.

Tesla and Nissan are some brands that promote this innovation.

Hardware Components

In order to make good decisions, the car needs to collect all the information available in its environment. To do this, autonomous cars have a series of sensors and cameras that allow the collection of information. More specifically, each vehicle is usually equipped with a GPS unit, an inertial navigation system, and a series of sensors: laser meters, a radar, a lidar and video.

Data from each sensor is filtered to remove noise and combined to increase awareness of the environment and update the map of your environment for obstacle avoidance. Data is collected periodically because the process of localization, environment mapping, and obstacle detection occurs continuously.

One of the most important processes is to create a map of the environment and locate the position of the car itself in the environment. Cameras and laser meters are used for this task. The laser meter scans the environment by making successive passes and calculates the distance to nearby objects based on the time it takes for the light beam to return. Combining this information with that collected by the camera makes it possible to build a three-dimensional model of the environment.

To locate yourself on the model and know your relative position on the map you must use GPS and the inertial navigation system. The GPS position may be wrong by several meters due to a delay in transmission, so it is necessary to use the inertial navigation system.

The inertial navigation system (INS) is a navigation aid system that uses accelerometers and rotation sensors to continuously calculate an estimate of the position, orientation and speed of the car's movement. Combining it with the GPS position it is necessary to obtain a more accurate position.

Uno de los principales obstáculos para la implementación de los vehículos autónomos es la regulación.[36]​.

La Convención de Viena sobre la Circulación Vial de 1968"), suscrita por más de 70 países de todo el mundo, establece los principios que rigen las leyes de tráfico. Uno de los principios fundamentales de la convención ha sido el concepto de que un conductor tiene siempre el control y la responsabilidad totales del comportamiento de un vehículo en el tráfico. En 2016, una reforma de la convención abrió las posibilidades a las funciones automatizadas.

En 2020 más de 50 países firmaron en la ONU un reglamento que acotará el uso y el funcionamiento de los vehículos de conducción automática. De acuerdo con la normativa, los sistemas de mantenimiento de carril podrán ser activados en aquellas carreteras en las que peatones y ciclistas no estén permitidos y que cuenten con una separación física que divida el tráfico que circula en sentidos opuestos. Por el momento el Reglamento limita la velocidad operacional de este tipo de sistemas a un máximo de 60 km/h.[37]​[38]​.

European Union

At the level of the Union, autonomous driving timidly began to be legislated at the beginning of the century with the approval of Directive 2010/40/EU, of the European Parliament and of the Council, of July 7, 2010, which establishes the framework for the implementation of intelligent transport systems in the road transport sector and for interfaces with other modes of transport Text relevant for the purposes of the EEA. A common framework was developed in European territory whose purpose is to achieve a coordinated and effective implementation of these new technologies applicable to road traffic. This directive was transposed into the Spanish legal system through Royal Decree 662/2010, of April 13, which establishes the framework for the implementation of intelligent transport systems (ITS) in the road transport sector and for interfaces with other modes of transport.

In the European Union for a few days in October 2015, several autonomous vehicles were integrated into the traffic flow of Bordeaux (France), during the Intelligent Transportation World Congress (ITS). The vehicles presented by five companies were identifiable thanks to a specific registration to warn users of their presence. To have access to the roads open to the public, the organizers had to lift a series of restrictions. For example, a professional driver was present in the vehicles during the tests. Other prototypes were presented only in closed circuit.[40] Likewise, the British government has indicated that there are no legal barriers to testing autonomous vehicles on public streets in the country. However, he clarified that road and vehicle control regulations must be modified to allow autonomous vehicles to travel freely.[41]​.

United States of America

In the United States, around twenty companies have obtained permission to carry out tests with autonomous vehicles[43]​ and their widespread implementation is expected before the end of this decade.[44]​.

California, Nevada and Florida have authorized the use of autonomous vehicles on public roads.[41] However, most states in the United States do not have any laws regarding autonomous driving. The few that do have adjusted their laws to allow research and testing.[45]​.

In June 2011, Nevada was the first to enact laws regarding driverless car operations.[46][47][48] Nevada's law went into effect on March 1, 2012, and the Nevada Department of Motor Vehicles granted the first license to a driverless car in May 2012. It is a Toyota Prius modified with experimental technology from Google.[49]​.

For its part, in 2015, Daimler AG announced that its Freightliner Inspiration prototype became the first autonomous truck to receive authorization to travel through Nevada alongside current traffic.[50]​ The vehicle had already been previously tested in Germany.[51]​.

In March 2022, the U.S. National Highway Traffic Safety Administration, the federal agency in charge of safety regulation in the United States, approved that cars could be built and sold without a steering wheel or pedals.[52]​.

Other regions

Other countries that are regulating the autonomous vehicle are the United Kingdom,[53]​ South Korea,[54]​ China[55]​ or Australia.[56]​.

Contenido

Existe cierta incoherencia en la terminología utilizada en el sector de los coches autónomos. Varias organizaciones han propuesto definir un vocabulario preciso y coherente.

SAE International, un organismo de normalización de la automoción, publicó en 2014 un sistema de clasificación con seis niveles -que van desde los sistemas totalmente manuales a los totalmente automatizados- con el nombre de J3016 Taxonomy and Definitions for Terms Related to On-Road Motor Vehicle Automated Driving Systems. Esta clasificación se basa en la cantidad de intervención del conductor y en la atención requerida, más que en las capacidades del vehículo, aunque están vagamente relacionadas. En Estados Unidos, en 2013, la Administración Nacional de Seguridad del Tráfico en Carretera (NHTSA) publicó un sistema de clasificación formal, pero lo abandonó en favor de la norma SAE en 2016. También en 2016, SAE actualizó su clasificación, denominada J3016_201609. La clasificación es objeto de controversia,[16]​ con diferentes enfoques propuestos para una expansión.[17]​.

Driving automation levels

Level 0: The automated system issues warnings and can intervene momentarily, but does not have sustained control of the vehicle.

Level 1: The driver and the automated system share control of the vehicle. Examples of this are systems in which the driver controls the steering and the automated system controls the engine power to maintain a certain speed (cruise control) or the engine and brake power to maintain and vary the speed (adaptive cruise control or ACC); and park assist, in which steering is automated while speed is under manual control. The driver must be prepared to regain full control at any time.

Level 2: The automated system takes full control of the vehicle: acceleration, braking and steering. The driver must supervise driving and be prepared to intervene immediately at any time if the automated system does not respond correctly. Contact between hand and steering wheel is usually mandatory during SAE 2 driving, to confirm that the driver is ready to intervene. The driver's eyes can be monitored by cameras to confirm that the driver is maintaining their attention in traffic.

Level 3: The driver can safely divert his or her attention from driving tasks, for example, the driver can text or watch a movie. The vehicle will handle situations that require an immediate response, such as emergency braking. The driver must continue to be prepared to intervene within a limited time, specified by the manufacturer, when the vehicle asks him to do so. This automated system is sometimes defined as a co-pilot that warns in an orderly manner when manual driving is necessary.

Level 4: As level 3, but driver's attention is never required for safety, for example, the driver can safely go to sleep or leave the driver's seat. However, self-driving is only supported in limited spatial areas or under special circumstances. Outside of these areas or circumstances, the vehicle must be able to safely abort the trip, for example, slowing down and parking the car, if the driver does not regain control. An example would be a robot taxi or a robot delivery service that covers selected places in an area, at a specific time and in specific quantities.

Level 5: (Also called "optional steering wheel"): No human intervention required. An example would be a robotic vehicle that works on all types of surfaces, all over the world, all year round and in all weather conditions.

Lucia Errandonea, creator of Algoritmic Societies,[18]​ assures that to know how a car moves and decides, we first have to understand how we move. Humans mix our senses through the knowledge we acquire to react to our environment, and we combine all of this with the motor system, which gives way to mobility. It says, "All of these systems are coordinated with the nervous system, which gives orders and makes decisions," which may be more difficult for an intelligent automated system to combine.

Furthermore, Errandonea points out that the fact that a machine cannot interpret the environment leads to creating different dilemmas and controversies. As could be the tram dilemma, which depicts what an autonomous car should do if it finds itself in the situation of saving one person or another. These autonomous cars are referred to as commercial cars, corresponding to the first level of automation. Another dilemma that I mentioned is "false marketing" about the level of automation.

Therefore, with this we conclude that the artificial intelligence that makes cars autonomous is not mature enough. As Julián Estévez, a researcher in Robotics and artificial intelligence, says, he talks about how difficult it is to get cars with artificial intelligence to be accurate enough for correct driving. Above all, because it is necessary to insert these autonomous systems into society, Estévez assures, "there is a problem between the coexistence of autonomous and non-autonomous cars" and hence the proposals to separate the circulation environments of each one, proposing ideas such as: having them in different lanes, different areas of the city or different schedules. Therefore, apart from being a technological problem, it is also the approach to cities and how we have to move.

The earliest known presentation of an autonomous vehicle was by Norman Bel Geddes at the Futurama trade show "Futurama (exhibition)") sponsored by General Motors for the 1939 World's Fair, which consisted of an electric vehicle that was controlled by an electrical circuit embedded in the road pavement.

In 1980, a vision-guided van from Mercedes-Benz, designed by Ernst Dickmanns") and his team from the University of Munich reached 100 km/h on streets without traffic. The European Commission was interested in the project and made an investment of €800 million for the EUREKA Prometheus project that sought to develop an autonomous vehicle.

In 1980, the Defense Advanced Research Projects Agency (DARPA) achieved the first vehicle that operated using radar, laser, and computer vision. In 1987, HRL laboratories demonstrated that a vehicle could be built that could design its own route once it left the map. The vehicle was able to move more than 600 meters through complex terrain such as slopes, large rocks and vegetation.

In 1994, two twin robot vehicles VaMP and Vita-2") from Daimler-Benz and Ernst Dickmans drove alone for more than a thousand kilometers on a 3-lane Paris motorway, on regular days and in heavy traffic at speeds of 130 km/h. The vehicles operated autonomously with little human intervention. They demonstrated that driving in free lanes, in convoy, and changing lanes to the right and left could be performed autonomously.

In 1995, Dickmanns' team modified a Mercedes-Benz S-Class to make a trip between Munich and Copenhagen and back, using computer vision with saccades and a computer to react in real time. The robot managed to reach speeds greater than 175 km/h on the German Autobahn, with an average time between human interventions of 9 km, which represented 95% autonomous driving. Again the vehicle drove in traffic, executing maneuvers to overtake other vehicles.

The pioneering work in Spain was developed by the CSIC.[19][20] Subsequently, the first autonomous driving tests in Spain were carried out in 2012 on a 100km route from El Escorial to Madrid carried out by the AUTOPIA group of the Center for Automation and Robotics. A manual vehicle dynamically generated a high-precision map to be followed by the fully autonomous tracking vehicle. The trip covered a wide range of driving scenarios, including urban areas, secondary roads and motorways, in standard traffic conditions.[21]​.

For its part, Audi announced in 2014 that its autonomous RS7 model reached 240 km/h on the Hockenheim circuit in Germany. The car completed one lap of the track in just over two minutes. Audi also placed a human behind the wheel for a comparison lap, this being five seconds slower than the unmanned car.[22]​.

In April 2015, a self-driving Audi SQ5, using an autonomous driving system developed by Audi and Delphi, managed to travel 5,400 kilometers in 9 days without the engineers inside the car touching the steering wheel except on a couple of occasions. Weeks later, a fleet of 25 autonomous vehicles designed by Google left the test tracks to be evaluated in the daily traffic of Mountain View (California), California, without passing 40 km/h. Google's prototype has the navigation software that it already used in its fleet of Lexus cars, which traveled more than a million kilometers autonomously and under the supervision of a human at the wheel.[24]​.

In the summer of 2015, the University of Michigan launched a prop village, called MCity (in:)"), in whose streets autonomous vehicles can be tested. The place has a street more than a mile long, curves of different radii, roundabouts, traffic lights, pavements of different surfaces, etc.[25]​ On the other hand, there are other projects that will complement MCity. One of them is the installation of 9,000 interconnected vehicles in the large Ann Arbor area, as well as another 20,000 on roads in the southeast of the state.[26] During the same week, in Buenos Aires (Argentina), representatives of the local government tested two units of the first prototype of an electric autonomous vehicle manufactured in that country. It is a vehicle without a steering wheel or pedals.[27]​.

The first thing we have to take into account when talking about autonomous driving is the fact that we are dealing with a machine, a large computer with wheels that makes decisions based on unquestionable guidelines.

In this way, on a road with a speed limit of 80 km/h, the autonomous car will never exceed that limit unless its predefined operating parameters are altered. Ambiguity does not exist in machine decision making.

If we stick to the theory, the autonomous car is the perfect figure as a transportation system, where its operation is based on compliance with imposed laws and guidelines. There is no interpretation, much less violation of the rules unless there is a ruling. That said, the figure of the autonomous car is understood as the best advance in safety that has occurred in recent years, since its operation results in a control and prudence that is impossible for a human being to achieve.

In 2021 and 2022, the terminologies “automated driving system” and “automated driving” appear in article 1 and new article 34 bis of the 1968 Convention on Road Traffic").[28]

The invention, development and dissemination of the autonomous vehicle is producing enormous positive effects (road safety, accessibility, energy efficiency, etc.) and at the same time causes multiple challenges.

Among the challenges caused by this technology are the changes that the automobile industry will face (new companies enter the sector), insurance (fewer accidents could mean a lower cost of "Premium (insurance)" premiums and a new definition of liability in the event of an accident), the labor market (many professional drivers such as taxi drivers or truck drivers and employees of mechanical workshops will lose their jobs while more robotics experts will be needed) or the supply of organs for donation (a significant decrease in the number of deaths could mean a big drop in potential donors).

Software Components

To achieve truly automatic driving in an urban situation with unpredictable traffic, many real-time systems are necessary, which must interoperate. For example, a system of location, perception of the environment, planning and, logically, a control system is necessary. In addition, a set of sensors are necessary that collect and provide the necessary information to make decisions.

When we think about a control system for an autonomous car, it is easy to fall into the temptation of imagining that the algorithms that control it resemble classic programming rules, such as “if (…) then…else…”, for example “If a pedestrian crosses the middle of the street, then the speed will be reduced, braking if necessary.” It is evident that this way of thinking is not appropriate since the number of possible situations is very high and therefore a system capable of generalizing is necessary.

Instead of implementing countless rules to recognize these objects, it is much more practical to implement a machine learning algorithm that is “trained” with different images that exemplify all possible situations.

Each of the images is associated with the type of vehicle it contains. The algorithm begins to process the images. Initially you try to guess which vehicle is in each image and you will be wrong very often at first. Since it knows “the solution”, that is, which vehicle is actually in each image, it modifies and adapts internal parameters and tries again. The process continues to iteratively reduce the failure rate. Later, when you are presented with new images you will be able to classify them correctly. We can, then, affirm that the algorithm has learned.

That same approach can be used for decision making and evaluation. Instead of providing a list of rules with which to evaluate the action to take for each situation, an algorithm is trained with traffic situations in which the correct action to take is specified. As before, the algorithm tries to guess the correct action and modifies internal parameters depending on whether it is wrong or right.

Another technique used in autonomous driving algorithms is probabilistic models. To make decisions, the car needs to predict its own position and also that of all the objects it identifies, such as pedestrians or other vehicles, since they are moving. Since it does not know the position they will be in, it tries to predict it using probability distributions.

Tesla and Nissan are some brands that promote this innovation.

Hardware Components

In order to make good decisions, the car needs to collect all the information available in its environment. To do this, autonomous cars have a series of sensors and cameras that allow the collection of information. More specifically, each vehicle is usually equipped with a GPS unit, an inertial navigation system, and a series of sensors: laser meters, a radar, a lidar and video.

Data from each sensor is filtered to remove noise and combined to increase awareness of the environment and update the map of your environment for obstacle avoidance. Data is collected periodically because the process of localization, environment mapping, and obstacle detection occurs continuously.

One of the most important processes is to create a map of the environment and locate the position of the car itself in the environment. Cameras and laser meters are used for this task. The laser meter scans the environment by making successive passes and calculates the distance to nearby objects based on the time it takes for the light beam to return. Combining this information with that collected by the camera makes it possible to build a three-dimensional model of the environment.

To locate yourself on the model and know your relative position on the map you must use GPS and the inertial navigation system. The GPS position may be wrong by several meters due to a delay in transmission, so it is necessary to use the inertial navigation system.

The inertial navigation system (INS) is a navigation aid system that uses accelerometers and rotation sensors to continuously calculate an estimate of the position, orientation and speed of the car's movement. Combining it with the GPS position it is necessary to obtain a more accurate position.

Uno de los principales obstáculos para la implementación de los vehículos autónomos es la regulación.[36]​.

La Convención de Viena sobre la Circulación Vial de 1968"), suscrita por más de 70 países de todo el mundo, establece los principios que rigen las leyes de tráfico. Uno de los principios fundamentales de la convención ha sido el concepto de que un conductor tiene siempre el control y la responsabilidad totales del comportamiento de un vehículo en el tráfico. En 2016, una reforma de la convención abrió las posibilidades a las funciones automatizadas.

En 2020 más de 50 países firmaron en la ONU un reglamento que acotará el uso y el funcionamiento de los vehículos de conducción automática. De acuerdo con la normativa, los sistemas de mantenimiento de carril podrán ser activados en aquellas carreteras en las que peatones y ciclistas no estén permitidos y que cuenten con una separación física que divida el tráfico que circula en sentidos opuestos. Por el momento el Reglamento limita la velocidad operacional de este tipo de sistemas a un máximo de 60 km/h.[37]​[38]​.

European Union

At the level of the Union, autonomous driving timidly began to be legislated at the beginning of the century with the approval of Directive 2010/40/EU, of the European Parliament and of the Council, of July 7, 2010, which establishes the framework for the implementation of intelligent transport systems in the road transport sector and for interfaces with other modes of transport Text relevant for the purposes of the EEA. A common framework was developed in European territory whose purpose is to achieve a coordinated and effective implementation of these new technologies applicable to road traffic. This directive was transposed into the Spanish legal system through Royal Decree 662/2010, of April 13, which establishes the framework for the implementation of intelligent transport systems (ITS) in the road transport sector and for interfaces with other modes of transport.

In the European Union for a few days in October 2015, several autonomous vehicles were integrated into the traffic flow of Bordeaux (France), during the Intelligent Transportation World Congress (ITS). The vehicles presented by five companies were identifiable thanks to a specific registration to warn users of their presence. To have access to the roads open to the public, the organizers had to lift a series of restrictions. For example, a professional driver was present in the vehicles during the tests. Other prototypes were presented only in closed circuit.[40] Likewise, the British government has indicated that there are no legal barriers to testing autonomous vehicles on public streets in the country. However, he clarified that road and vehicle control regulations must be modified to allow autonomous vehicles to travel freely.[41]​.

United States of America

In the United States, around twenty companies have obtained permission to carry out tests with autonomous vehicles[43]​ and their widespread implementation is expected before the end of this decade.[44]​.

California, Nevada and Florida have authorized the use of autonomous vehicles on public roads.[41] However, most states in the United States do not have any laws regarding autonomous driving. The few that do have adjusted their laws to allow research and testing.[45]​.

In June 2011, Nevada was the first to enact laws regarding driverless car operations.[46][47][48] Nevada's law went into effect on March 1, 2012, and the Nevada Department of Motor Vehicles granted the first license to a driverless car in May 2012. It is a Toyota Prius modified with experimental technology from Google.[49]​.

For its part, in 2015, Daimler AG announced that its Freightliner Inspiration prototype became the first autonomous truck to receive authorization to travel through Nevada alongside current traffic.[50]​ The vehicle had already been previously tested in Germany.[51]​.

In March 2022, the U.S. National Highway Traffic Safety Administration, the federal agency in charge of safety regulation in the United States, approved that cars could be built and sold without a steering wheel or pedals.[52]​.

Other regions

Other countries that are regulating the autonomous vehicle are the United Kingdom,[53]​ South Korea,[54]​ China[55]​ or Australia.[56]​.