Cars are the most ubiquitous transport method nowadays. They offer comfort, freedom, availability, various expenses and a slim-but-not-insignificant probability of being involved in an accident.

Casualties are diminishing, yet cars remain as one of the deadliest transport systems, only surpassed mainly by motorcycles. They are also an untenable economic burden, costing between 1% and 3% of each country’s GDP according to the World Health Organization [1].

Joint efforts by governments and car manufacturers are striving to reduce this rate significantly: the European Commission, for instance, has developed several strategies, a Cooperative Intelligent Transport Systems among others, which focuses on reducing human error and creating environmentally friendly solutions.

According to the data in the preliminary 2017 road safety statistics, there were 20% less fatalities in 2017 compared to 2010 and 2% less than 2016. The Commission has issued the objective of halving the road fatalities between 2010 and 2020, setting some targets [2] which are unfortunately not being met, meaning that reaching the goal will be challenging.

How can we increase road safety?

Bounteous efforts and resources are vital to increase road safety, and the industry has proposed solutions such as big data and autonomous vehicles. Since most car accidents are due to the human factor and not so much technical failure, it sounds logical to replace the human factor by software.

Possible solution: self-driving cars

The first challenge is to build a self-driving system that can perceive the road better than the best human driver, which won’t happen soon. Current maps are also not accurate enough [3][4]. But what does autonomous/self-driving mean?

Briefly stated, there are 5 levels of car autonomy:

• a 0-level car has no automated features,
• a level 3 means a trade-off between human and machine and, according to studies, it’s the deadliest level [3].
• The vehicle drives itself entirely in every condition in the level 5 [4].
• As of 2018, the most advanced cars in the market are halfway through the level 2.

Current self-driving cars are equipped with internal and external sensors, Lidar units, cameras and powerful software including driver’s speech and gesture recognition, language translation and reinforcement learning algorithms [5]. The cars continuously render the close environment by capturing data about the size and speed of nearby objects and forecasting possible changes. The main tasks of the algorithm can be divided into object detection, identification, localization and movement prediction.

In the reinforcement learning algorithm, rewards are assigned to certain outcomes, pushing the algorithm to learn to behave accordingly. The more data, the more precisely the algorithm updates its parameters to achieve the rewards.

The causation is thus as follows: the more data a self-driving car collects, the more robust and more adaptable will the algorithm be, and the fewer the accidents.

Yet just gathering data is not enough; roads, environments and the behavior of the human drivers can vastly differ depending if the area is urban or rural, the customs of the particular country and non-written laws.

It is therefore indispensable that these cars are trained with as much diverse data as possible. The following dilemma arises: how much data is necessary? When can self-driving cars be allowed to be on the road?

At which point in the development and testing can we allow cars on the road?

Autonomous vehicles need to drive a huge amount of distance before manufacturers can affirm that they are statistically as safe as human drivers, senior RAND Institute researchers argue [6].

They pose the question: “What’s important to people, do they want them to be safer than human drivers? Will that answer change over time?”. The amount of data that a manufacturer needs depends on how risk-averse the company is, and how trusting the general public.

The optimal point for entry is a controversial topic [7]:

people are more accepting of mistakes that other people make than mistakes that machines make, even if the machine overall performs better than people [8][9].

On the quest of reducing fatalities, self-driving cars have to face the challenges of creating and maintaining maps, mastering complex social interactions, responding to bad weather conditions, and showing robustness against cyberattacks [10].

Data privacy

The matter of privacy and data protection is not to be neglected, considering that cars and therefore private companies will know the driver’s most frequented places and may (and will) build a profile of the customer’s preferences and habits.

Nowadays, theoretically no one knows where a car is. If the passengers take their phone with them and activate location services, the OS can track their location, know the speed at which they move and when/where they stop. But not much else.

In cars where more software and Internet connectivity is involved, where users can give voice commands, there’s already some software company behind it processing the voice command. Surely these companies take privacy very seriously, but there is already a risk of data leakage, even if it’s minimal. The Amazon Echo device has had some data privacy issues already.

And in self-driving cars, the users put their lives and privacy on the hands of a software company, which is a big leap from traditional car manufacturers who have been around for decades. Uber, as instance, is merely 10 years old. Perhaps it is advisable to be wary about whom do we trust to drive us.

Other issues such as the regulation of the technology behind self-driving cars are contentious but beyond the extent of this essay.

Which companies have the lead?

When entering the competitive industry of self-driving cars, three companies are at the forefront, namely, Google, Tesla, and recently also Uber.

Google collects 15.000 autonomous miles per week and has accumulated 1.7 million miles in autonomous mode to this day in the USA [11], where fatalities, unlike in Europe, are increasing [12]. Americans drive 100 million miles before a fatal accident happens; hence, at the current rate Google would require years to drive the same distance.

Tesla, on the other hand, has a different approach. With the use of their “autopilot mode”, Tesla has gathered 1.3 billion miles [13], where the car was driven by a human with the Autopilot mode on. Nonetheless, using customers as test drivers has drawbacks, as the few but controversial fatal crashes show.

Uber, who started as a ridesharing company, has quickly pivoted into food delivery and bicycle sharing, as well as their own self-driving cars [14] and even self-driving trucks for a while. After a number of accidents, Uber continues to test their cars (at lower speed) in Pittsburgh [15][16] and Toronto [17]. Not much public information is available as to the nature and amount of data they are gathering. There is also no due date for their cars expand to other cities [18].

Conclusion

As a summary, autonomous cars are gradually joining the roads, a process which can be delayed depending on the reluctance of the general public, but not halted.

Many companies are investing in the development and enhancement of their self-driving algorithms and already in the USA roads in major cities serve as their playground.

As drivers get accustomed to being on the road with autonomous cars, the latters’ relevance is bound increase and the maybe-not-so-distant utopia of a fully autonomous and connected network of cars might not be so distant.

References

[1]: WHO World report (2013)
[2]: Road fatalities in EU since 2001
[3]: Crowdsourced maps should help driverless cars navigate our cities more safely (2019)
[4]: Mapping the world in 3D will let us paint streets with augmented reality (2019)
[5]: Future autonomous vehicle driver study (2016)
[6]: Levels of autonomy in cars
[7]: Navigating occluded intersections with autonomous vehicles using Deep Reinforcement Learning (2018)
[8]: Driving to safety: how many miles of driving would it take to demonstrate autonomous vehicle reliability (2016)
[9]: Why waiting for perfect autonomous vehicles may cost lives (2017)
[10]: Consumer interest in automation (2017)
[11]: CARAVAN Public Opinion Poll: Driverless Cars (2018)
[12]: Challenges and obstacles of self-driving cars (2016)
[13]: Waymo, on the road
[14]: U.S. road safety laws lag, while fatalities climb (2018)
[15]: Tesla, autopilot data (2016)
[16]: Our road to self-driving vehicles (2017)
[17]: Uber’s self-driving cars return to public roads for the first time since fatal crash (2018)
[18]: Uber was just approved to resume self-driving tests in Pittsburgh and the rest of the state (2018)
[19]: Uber’s self-driving cars back on the road in Toronto after 9-month hiatus (2018)
[20]: Uber expects a long wait before self-driving cars dominate (2019)