Self-Driving Cars: The Concept and Levels, and the Self-Driving Future • Slash

Autonomous vehicle, or self-driving car, is a computer-controlled vehicle that drives itself. It should not be viewed simply as an upgraded car, however. The implications of the AV vision go much deeper, painting a picture of a full smart-city infrastructure. AV tech is predicted by most experts to reach maturity around and after year 2040, and if we do not increase awareness about this new technology and agree on solutions to possible problems, we will not be ready by that time the technology enter our daily lives. Let’s consider what ethical, practical, and legal issues AV may raise, and what possible solutions we could find for those.

Driving is a complex task that involves perception, navigation, planning, inference, decision-making, and controlling the vehicle. Humans use brain, hands and feet, and the AV uses neural networks and mechanical components. Cameras and radar make the AV’s perception function possible, inference is secured through algorithms to predict the intent of cars or pedestrians, etc. Another component of driving is the unpredictability – from sudden weather changes to unexpected human actions and reactions. How well can AV deal with it?

Let’s first consider the types of AV and the evolution it has traveled so far.

Currently, the Society of Automotive Engineers classifies AVs as “Level 0” to “Level 5.” Levels of 0 to 3 are more of extra functions, AI tools that the human driver uses. Level 4 is more autonomous, but it drives on a fixed route. Finally, Level 5 (L5) is fully autonomous, for instance, a robo-bus controlled by an app.

Levels 0 through 3 are available in commercial vehicles now, L4 has been deployed experimentally in some cities in 2018, but L5 is still far out of reach. One of the main problems is that AI needs to be trained on large amount of data representative of real-life driving, but there is no way no to collect that much information on all possible movements of the objects on the road in all possible scenarios. We could synthesize data and program rules, but synthesized data is not as good as the real one, and the rules may be inapplicable or contradicting in certain situations.

The promise of AV is greater convenience, lower cost, and improved safety. Let’s consider the possibilities.

AI algorithms can move AVs be closer to people who might need a ride soon (e.g. when an event is about to conclude, or when an airplane lands in the airport). The routes of AVs routes can be optimized depending on the total amount of time users wait and the total amount of time AVs are empty. Carpooling will cost much less with AVs, because 75 percent of the fare currently goes to the driver.

In the post-2041 world, where AVs are as much part of our lives as smartphones are now, people would gain the hours spent on driving back for their lives. As more data is collected, the AI technology for AV would improve, and the efficiency of AV would increase with more automation. Subsequently, AVs would be used more, driving down the cost, and with the many hours freed from driving, greater productivity would follow.

AVs will be able to communicate instant messages with one another to increase efficiency. For instance, if an AV is experiencing issues or needs to get ahead on the road, it can signal its needs or intentions, or a warning to other AVs. With the efficient AVs, fewer people will drive and buy cars. A lot of parking garages and spaces will fall out of use, freeing up huge amounts of space, and the reduced traffic congestion will decrease air pollution and consumption of fossil fuels.

Concerning approval of AVs by the governments, it is possible only when AVs are deemed “safer than people.” Several ethical issues arise, such as the concept of entrusting human lives into the hands of AI or the dilemma of “the trolley problem.” And if an accident occurs and includes a fatality, what is the judicial process going to be like? Who is going to be held accountable: car manufacturer, AI algorithm provider, or the engineer?

Widespread introduction of AVs will also raise the issue of employment. Millions of people now work as drivers, at gas stations and car dealership, parking garages, and AVs will put most of them out of work. Mechanical repairs will be far less demanded than software expertise, also affecting the employment rate.

There are problems that can potentially confuse the AI, such as natural disasters or acts of terrorism. For such cases, expert human drivers can take over from a tele-operations center, using high-fidelity video with minimal latency. The transmission of it would require a bandwidth of 6G, which is set to be available by about 2030.

Plus, there is the danger of sensationalism. Every error by AVs or accidents including them are extensively reported in the media, and while such attention is justified, it can potentially create an environment of mistrust and hostility toward the technology, which would impede its progress. A lot depends on the way the technology is introduced. Currently, there are two approaches: the Waymo way, which is to collect data slowly in safe environments and avoid fatalities before launching AV, and the Tesla way, which is to launch the product as soon as the AI is reasonably safe, knowing that in time the system will save many lives, although some may be lost in the beginning.

The development of L5 can be accelerated with removal of one assumption: that L5 will be functioning in the cities the way we have them now. The future city can be imagined as having separate layers for cars and pedestrians, with wireless communication between the road and the AVs, and the AVs on such augmented roads operating as trains on “virtual” rails.

The L5 technologies will mature through application of L0-L4 technologies over the years, continuous data collection, and iterative improvement. We have already started using autonomous mobile robots, and autonomous transport trucks can be next, and afterwards, we may have the 2041-and-beyond future with Level 5 AVs.