Firefly 1 is Waymo's (Google) fully self-driving reference vehicle. Credit: Waymo.

  • Tungsteno

How does an autonomous car think?


The idea that in the future we’ll be able to travel in cars that drive themselves either excites us or makes our heads spin. There are the optimists like Elon Musk, the founder of Tesla, who see this future as just around the corner, and other more cautious experts who don’t even consider it a realistic scenario or have serious doubts after several crashes have occurred. What is clear, however, is that car companies are incorporating more and more intelligent features into their new models. How does the artificial brain needed to get them driving actually work?

The state of the art of automation in automobiles

According to the most common international classification, there are six levels of automation of a car: level 0 indicates vehicles without any autonomy, while level 5 refers to cars able to move about without human intervention under all circumstances. Between the two extremes we find different intermediate points.

Companies like Waymo, belonging to Google, already carry out tests with level 4 vehicles (complete autonomous driving in restricted areas) and also level 5. In general, experts agree that in the market today you can find consolidated intelligent systems corresponding to level 2, which autonomously fulfil more functions at the same time, although they require the driver's control at all times. One example is the Nissan Propilot, incorporated into some recent models, an assistant to highway driving that is responsible for maintaining a set speed and safe distance or avoiding lane shifts in certain traffic conditions.

There is also talk of level 3 systems (capable of driving the car autonomously in certain contexts). Under this category we find the Audi AI traffic jam pilot, designed to be initially integrated into the new A8, presented as the first system that "can take care of the tasks that driving requires in situations of heavy highway traffic at speeds of up to 60 km/h".

The Audi AI traffic jam pilot is the world’s first system that enables highly automated driving at Level 3. Credit: Audi.

The brain of the cars

Antonio Masegosa, a researcher at Ikerbarsque and the University of Deusto, argues that the automation of a vehicle is achieved thanks to four main elements: actuators that control components such as the steering wheel, the brake or the accelerator; sensors that allow the environment to be perceived (especially cameras and radars); maps with a high degree of detail; and a control system for all the above.

Especially with regard to this last aspect, explains Masegosa, artificial intelligence plays a "fundamental" role. This technology allows the car to know "which objects are around it and if they can represent an obstacle", as well as recognize "a sign or a traffic light", through the information it receives from the sensors.

Secondly, adds the researcher, AI ​​is concerned with another essential issue: predicting what might happen. "When we drive, we make small predictions, for example if another vehicle is going to turn left or not. The AI ​​does the same." Thanks to these skills, the central system of an autonomous car will be able to make decisions for us, he concludes.

A few steps further along thanks to deep learning

Another push towards autonomous driving, says Masegosa, comes thanks to deep learning. "We used to have to program each one of the possible outcomes that could happen and the decisions that the car would have to make," he explains. "With deep learning technology, the vehicle is able to learn by itself how it should behave in each situation based on the information it receives."

Some companies take advantage of real driving situations to collect information about the behaviour of vehicles and users. Two examples are Waymo, whose autonomous cars have already travelled more than ten billion miles on US roads, and Tesla, whose cars come equipped with an "autopilot" system (which currently does not reach level 5) whose software is able to improve its own performance based on the data collected. Last November the Musk company announced that its vehicles had already covered one billion miles with this system activated.

The new Tesla incorporates the Autopilot feature that enables your car to steer, accelerate and brake automatically within its lane. Credit: Tesla.

A great potential to be explored

Masegosa believes that the main advantage of automating the car is the potential to ensure greater safety. "Autonomous vehicles can observe at 360º what their environment is, perceive it at all times, and anticipate what other road users are going to do. Also, they never get distracted," he maintains. The experience of travelling will also improve. "They will allow time to be employed on more productive tasks than driving."

Another expected benefit is greater efficiency in energy and environment terms. "Artificial intelligence has complete control of the car," says Masegosa. "It will be able to squeeze the most out of the available energy and, therefore, reduce consumption and emissions." Vehicles without a driver, in his opinion, can also represent a new opportunity for mobility for vulnerable groups, such as the elderly or the disabled, or those of limited resources. "With regard to taxi services, without the cost of the driver, the price could be cheaper by up to half," he argues.

We still have some time before we’ll see fully autonomous cars on our roads. How much time is the subject of a broad debate in the sector. According to Masegosa, among the most complex problems to solve is the ability of AI systems ​​to operate in environments without clear signage or detailed maps, such as rural areas, and to predict sporadic events, for example when interacting with cars driven by humans.

In any case, it doesn’t appear that this race towards autonomous cars is going to be interrupted, at least according to the road map established in March by the European Road Transport Research Advisory Council. The organization, promoted by the European Commission, foresees that between this year and 2022, models with different level 3 systems will arrive on the market and, starting in 2024, level 4 models will also appear. However, it prefers not to get too specific about predicting when there will be cars that no longer need us for anything when it’s time to hit the road.

· — —
Tungsteno is a journalism laboratory to scan the essence of innovation. Devised by Materia Publicaciones Científicas for Sacyr’s blog.

  • Artificial intelligence
  • Autonomous vehicles
  • Waymo
  • Antonio Masegosa
  • Driving

We use our own and third party cookies for analytical purposes. Click on HERE for more information. You can accept all cookies by clicking the "Accept" button or set them up or refuse their use by clicking .

Cookie declaration

These cookies are necessary for the website to function and cannot be disabled in our systems. These cookies do not store any personally identifiable information.

Name Provider Purpose Expiration Type
LFR_Sesión_STATE_* Liferay Manage your session as a registered user Session HTTP
GUEST_LANGUAGE_ID Liferay Determines the language with which it accesses, to show the same in the next session 1 year HTTP
ANONYMOUS_USER_ID Liferay Manage your session as an unregistered user 1 year HTTP
COOKIE_SUPPORT Liferay Identifies that the use of cookies is necessary for the operation of the portal 1 year HTTP
JSesiónID Liferay Manages login and indicates you are using the site Session HTTP
SACYRGDPR Sacyr Used to manage the cookie policy Session HTTP

These cookies allow us to count visits and sources of circulation in order to measure and improve the performance of our site. They help us know which pages are the most or least popular, and see how many people visit the site. All information collected by these cookies is aggregated and therefore anonymous.

Name Provider Purpose Expiration Type
_gat Google It is used to throttle the request rate - limiting the collection of data on high traffic sites Session HTTP
_gid Google It is used to store and update a unique value for each page visited Session HTTP
_ga Google This is used for statistical and analytical purposes for increasing performance of our Services Session HTTP