Autonomous Cars (En)

General and Cases

In May 2016, the first fatality in an accident involving self-driving cars occurred. The then Tesla Model S was traveling in Williston, Florida and failed to identify a white-colored truck. Consequently, it miscalculated its trajectory without detecting the presence of the vehicle, leading to an accident that resulted in the death of the driver, Joshua Brown. Even though the car was being guided by the "AutoPilot" system, it was traveling at an approximate speed of 120 kilometers per hour. Using this case as an example, it is possible to initiate a process of reflection on the contrasts between technological advancement and the safety of individuals who interact with these innovations in the automotive sector.

How these systems work and why errors contribute to improvement

It is important to emphasize how self-driving cars systems function and how these types of errors can contribute to the enhancement of technology, increasing its efficiency and safety. Firstly, it is necessary to understand the algorithm as a logic developed based on "if-then-else" rules, which guide the machine's decision-making process. In this way, the vehicle analyzes its surroundings, identifying all elements and applying them to its programming. The complexity of the machine's function is so high that it is impossible for a human to single-handedly program all the nearly infinite possibilities that the computer will encounter. Hence, deep neural networks are employed to teach the machine how to make its decisions.

For many years, the automotive sector has aimed to enhance advanced driver assistance mechanisms, such as airbags, sensors, or cruise control. However, what began to happen in 2016 with the implementation of Tesla's AutoPilot, although following the same logic already adopted by the industry regarding technology, distinguishes itself from what was done before as it represents a significant advancement. The fundamental difference between driver assistance mechanisms and AutoPilot lies in the level of interdependence each establishes with the vehicle's driver. In the first case, each feature is designed to optimize the driver's driving process, facilitating decision-making and preventing errors with alerts. In the latter case, instead of providing the driver with information and alerts to assist in decision-making, it gathers all this data, processes it, and, through software, makes decisions, significantly reducing the importance of the human factor in the process.

The foundational structure for rapid parallel processing, necessary for the implementation of machine learning, was initially created with the purpose of serving the video game market, using Graphics Processing Units (GPUs). With the adoption of this new tool, it became possible to develop computational capabilities for deep neural networks. Supported by this technology, digital image recognition and speech recognition became feasible. Thus, it became part of the vehicle's processing.

There is an interesting difference between human and machine perception regarding certain types of activities. As a result, there are tasks that are easy for humans to perform but represent a much greater complexity for machines, and vice versa. The act of driving also falls within this logic, being far more complex for computers. To enable these machines to carry out the task correctly, a highly robust real or simulated driving dataset is required. By processing this extensive information, vehicles learn collectively as a fleet rather than individually.

How advanced are these vehicles, and how far are they from becoming more common?

Despite advancements in computing technology, with vehicles like Tesla's Model S and Model X featuring Autopilot as a tool since 2016, even as these models have existed and improved since their launch, with various limitations being addressed through updates, there are still challenges to overcome for the widespread adoption and popularization of autonomous cars in the market. In each new market, there are unique challenges to be faced by automakers and their algorithms due to national legislation and cultural differences. So, despite the growing presence and expectations surrounding these cars on the internet and in the media, their accessibility remains somewhat distant and far less developed than the technology itself.

To better understand and facilitate the governance of autonomous cars, the United States' National Highway Traffic Safety Administration (NHTSA) uses six levels to classify the degree of vehicle automation concerning human drivers independence. These levels align with the standards set by the SAE International (Society of Automotive Engineers), which serves as the foundation for several regions around the world.

In summary, Level 0 is marked by no automation, with at most driver assistance mechanisms, at this level, the vehicle is entirely dependent on the driver.

Level 1 introduces tools such as Adaptive Cruise Control, which already performs certain driving functions for the driver, such as staying within the lane or maintaining a constant speed relative to the vehicle in front. However, at this level, the car still relies on constant driver interaction.

At Level 2, known as partial automation, the driver becomes less important, although they must be attentive with their hands on the wheel, ready to take control. The car can combine various automated functions like braking and steering.

In Level 3, with the combination of existing technologies such as artificial intelligence and edge computing, it becomes possible for cars to effectively drive themselves under ideal conditions, with limitations. Nonetheless, driver attention is still required to make decisions in unpredictable situations. This level aligns with the current reality of more advanced cars in the market.

Level 4, high automation, eliminates the need for human decision-making while driving, requiring only destination inputs from the human. Cars at this level do not yet exist in the market and are still in testing.

Finally, Level 5 represents full automation, where the car is so independent that it does not require steering wheels and pedals as basic equipment, and it is fully prepared to handle adverse driving conditions.

Tesla AutoPilot

When it comes to autonomous cars, the first thing that comes to mind is a Tesla, which has managed to convey to the general public the impression of being “almost” an autonomous car. However, Tesla’s AutoPilot is only a semi-autonomous mode, operating at level 2 of automation, just like several other traditional brands with their respective semi-autonomous modes.

This perception has a reason. Despite being at the same level of automation, Tesla is ahead due to its pioneering approach, but there are significant differences in operation. Each brand also has a different vision, which means some brands do not go beyond a certain point in their cars.

Let’s start by talking about Tesla, specifically AutoPilot, which is Tesla’s semi-autonomous mode. As mentioned earlier, this mode, which was introduced in 2016, continuously receives updates, bringing improvements and new functions, many of which are still in beta versions. Tesla uses its owners to test and improve the system. This was only possible because Tesla markets its cars as modern and technological, targeting people who are looking for precisely that, seeing a futuristic and ahead of its time car. This explains why owners used beta versions extensively, providing Tesla with valuable user data. This is why Tesla is ahead of its competitors, even though they are at the same level as defined by the SAE, the brand has an advantage.

From 2020, all Tesla cars became equipped with AutoPilot, but you need to purchase it separately for $10,000 or add it on after buying the car. Another thing to note is that AutoPilot may easily become a subscription in the future, as the car is directly connected to Tesla. Additionally, Tesla is developing Full Self Driving (FSD), which is in beta version. FSD is an expansion of AutoPilot, bringing more features such as traffic light control and stop signs. In the future, it is expected to merge with AutoPilot in the coming years, potentially advancing to the next level.

Another point of significant discussion is the number of accidents involving Teslas.

According to the NHTSA (National Highway Traffic Safety Administration), between July 2021 and May 2022, 69,6% of accidents involving autonomous systems featured Teslas. However, it’s important to note that data for cars from other brands in semi-autonomous modes are not available in real-time, so it is plausible to consider that this number is lower. Nevertheless, these numbers are relatively small when considering the total number of accidents that typically occur without involving this technology.

Furthermore, there is misuse of the mode, as mentioned earlier. AutoPilot is not fully autonomous, so drivers need to remain vigilant for potential issues and use it under appropriate conditions, such as on highways and roads. However, it’s common for drivers to activate AutoPilot and mistakenly believe they are in a 100% autonomous car, leading to accidents that could have been avoided.

There are also differences in how it functions. Starting in 2021, only top models like the Model S and X are equipped with radar, while simpler models use cameras to execute AutoPilot, as the brand has a different vision.

This vision comes from CEO Elon Musk. According to him, cars should learn to drive in the same way as humans, that is to say, using vision. This is why the brand is using only cameras in some models. This approach goes against the direction of the rest of the industry, as the removal of radar limits certain decisions, especially in adverse conditions like fog and storms, for example.

Traditional Brands

The “traditional” brands have fallen behind not only due to the delay in developing and implementing semi-autonomous modes, but also because they cannot have the freedom that Tesla enjoys. Being historical brands with a legacy, they are reluctant to offer a feature that hasn’t been thoroughly tested and approved for the public. In other words, these brands cannot release a beta version of something on the market because their customers are not the same as Tesla’s. This alone puts them far behind, even though numerous tests are conducted, the data from people using the modes in the real world is much more comprehensive, aiding in faster refinement.

But those who think they are lagging behind are mistaken. The investment in this area is massive and it’s becoming a reality. Brands like Volvo, Audi, BMW, Mercedes, Honda, Toyota, Ford, GM, and others are advancing their semi-autonomous modes, which are becoming increasingly better and more complete.

Level 3 of Automation

The major challenge, however, is not reaching level 3 of automation, brands like Mercedes, Audi and Honda have already achieved it with their models the S-Class, A8 and Legend respectively. The issue lies in legal concerns, causing vehicles with such a level of automation to be available only in very limited regions.

The fact that in many countries, brands can be held liable for potential accidents caused by vehicles with this level of automation hinders the introduction of more models. Additionally, there is a lack of trust in these vehicles by politicians, leading to bans in various regions. For these reasons, this level is only present in high-end models and is consequently not available in every region.

In the USA, there are already states where level 3 cars can be officially operated. Nevada was the pioneer, followed by California, which recently authorized vehicles of this level. So far, only Mercedes has obtained approval in both states, but it is expected that more brands will achieve the same level soon.

Currently, in the EU, there are already new regulations regarding autonomous vehicles from 2022. However, each country has the freedom to authorize them or not. Germany was the first country in the world to establish a legal framework for fully autonomous driving, allowing autonomous driving functions on defined routes, we are talking about level 4!

France is another country where level 3 vehicles are possible, with the Mercedes S-Class already approved for use.

In other EU countries, like Sweden, the responsibility in case of possible accidents has not yet been defined. There is no specific rule regarding liability in European legislation, which prevents brands from making these cars available in those regions.

The United Kingdom, on the other hand, has plans to release level 3 cars this year, but there are no authorized cars at the moment. As part of this plan, the UK aims to have vehicles at levels 4 and 5 certified by 2025.

Conclusion

There is much room for improvement in this regard, reaching this level goes far beyond the technological capability, but it’s also a matter of trust. After all, at this level, the driver can do other things while the vehicle is in autonomous mode, without needing to pay attention to the road. Of course, this is limited to specific moments and appropriate locations, but the fact that the driver is not attentive raises many questions about legislation and liability.

Furthermore, vehicles at this level need to contend with other risks, particularly cybersecurity, as they cannot be vulnerable to attacks that might result in the loss of control over crucial vehicle functions.

All of these issues are already on the table, and it’s a matter for the future. Level 4 and 5 are considered fully autonomous, and the decisions and regulations for the release of level 3 will serve as the foundation to determine when we can achieve full automation.

References

Sage Journals. Machine Learning, social learning and the governance of self-driving cars 2017. Available at: https://journals.sagepub.com/doi/abs/10.1177/0306312717741687

Uol. Carros elétricos da Tesla ganharam um piloto automático - eis como ele funciona 2015. Available at: https://gizmodo.uol.com.br/tesla-model-s-autopilot/

Próximo Nível. Você conhece os 6 níveis de um carro aut?nomo? 2020. Available at: https://proximonivel.embratel.com.br/voce-conhece-os-6-niveis-de-um-carro-autonomo/

Iot for all. Artificial Intelligence (AI) and Autonomous Vehicles 2022. Available at: https://www.iotforall.com/artificial-intelligence-and-autonomous-vehicles

Tremend. How Data Science is Used in Autonomous Driving? 2022. Available at: https://tremend.com/insights/data-science-in-autonomous-driving/

The Economic Times. Tesla Autopilot: What is it and how does it work? Here’s everything you may want to know 2023. Available at: https://economictimes.indiatimes.com/news/international/us/tesla-autopilot-what-is-it-and-how-does-it-work-heres-everything-you-may-want-to-know/articleshow/101601035.cms

U.S.News. What Does Tesla’s Full-Self Driving Mode Do? 2023. Available at: https://cars.usnews.com/cars-trucks/advice/tesla-full-self-driving

Car and Driver. How Capable Is Tesla’s Autopilot Driver-Assist System? We It to the Test 2021. Available at: https://www.caranddriver.com/news/a35839385/tesla-autopilot-full-self-driving-autonomous-capabilities-tested-explained/

Tecnoblog. Tesla libera polêmico Full Self-Driving em beta para todos nos EUA e Canadá 2022. Available at: https://tecnoblog.net/noticias/2022/11/25/tesla-libera-polemico-full-self-driving-em-beta-para-todos-nos-eua-e-canada/

Autocrypt. The State of Level 3 Autonomous Driving in 2023: Ready for the Mass Market? 2023. Available at: https://autocrypt.io/the-state-of-level-3-autonomous-driving-in-2023/#:~:text=At%20CES%202023%2C%20Mercedes%2DBenz,L3%20in%20Nevada%20for%20now

Business Going Digital. The legal framework for autonomous vehicles in the European Union 2020. Available at: https://www.businessgoing.digital/the-legal-framework-for-autonomous-vehicles-in-the-european-union/

motor1.com. Mercedes Drive Pilot Level 3 autonomous tech officially on sale in Germany 2022. Available at: https://uk.motor1.com/news/584163/mercedes-autonomous-drive-pilot-on-sale/

electrek. California certified as next US state to allow Mercedes’ Level 3 autonomous DRIVE PILOT 2023. Available at: https://electrek.co/2023/06/08/california-certified-us-mercedes-level-3-autonomous-drive-pilot-driving/