What’s the deal with self-driving cars? Demystifying the slow-burning tech

There was a time when we as a society couldn’t even fathom the idea of self-driving cars existing, let alone owning one. How did all this change? Over a decade ago, the US Congress decided to invest in driverless vehicles—not with the intention for us hard-working citizens to get some shut-eye on our way to work, but to protect its soldiers on the battlefield. 

To do so, the US paired up with the Defense Advanced Research Projects Agency (DARPA), which, as you may know, had a hand in helping create the Internet and GPS technology. But progress was slow on the self-driving cars front, so DARPA opened up the project to other sharp minds around the world with the 2004 DARPA Grand Challenge. The competition attracted top universities like Stanford, and participants were eventually recruited by leading tech and automotive companies in order to kickstart the autonomous vehicle (AV) industry. 

How do robocars work?

In terms of design, while there are slight variations from company to company, the main idea behind each self-driving system is the same. Each aims to create an internal map of their surroundings using various sensor technologies such as radars or lidars (a technology that gained popularity during the DARPA competitions). 

Google’s Waymo, for example, has used sonar, radar, lasers, and high-powered cameras at different stages of prototype development, while Uber’s operates using 64 laser beams, along with other sensors. 

Of course, these aren’t the only technologies acting as pieces of the self-driving puzzle. Some other pieces include: machine learning; responsibility-sensitive framework (RSS), which is the development of mathematical models that help identify the cause of a dangerous situation and how the AV should consequently respond; and network infrastructure, such as 5G and cloud systems, in order to facilitate vehicle-to-infrastructure and vehicle-to-vehicle communication. 

What makes a car considered «automated»?

Not all self-driving cars are created equal. To assess each vehicle’s intelligence and automation capability, the Society of Automotive Engineers divides them into levels.

• Level 1 has been widely used since the 1950s, with features such as the anti-lock braking system (ABS). 
• Level 2 has been in use since the year 2000, when we were all amazed by automatic emergency braking. 
• Level 3 is in play today. Here, the machine handles two or more driving functions at once, such as lane keeping and cruise control. 
• Level 4 is said to be in testing and only operational under human supervision. 
• Level 5 is fully controlled by computers, with no human intervention whatsoever. It is predicted to be achieved and widely used by 2060.

The race to complete automation

Is there a way to find out which producer of self-driving cars is paving the way to level 5? Two figures are examined when determining how advanced a company’s program is. The first is the number of miles the car has driven. The more an AV has driven, the more training data the company has on it (this also tends to indicate how much money they’ve invested in it). The second factor is defined by the number of disengagements—or the number of times a human driver has had to intervene in order to prevent an accident or handle a perilous situation—per mile driven.

So who’s got the best figures? It isn’t an AV company who’s leading the race, but the all-powerful Google’s sister company, Waymo. They’ve managed to clock a total of 20 million miles, with 1.2 million miles being driven in 2018 alone in the state of California. They registered 0.09 disengagements per 1000 miles—the best stat of the year. Waymo was followed by GM’s Cruise, clocking around half a million miles with 0.19 disengagements.

An important point to note here is that not all companies are open to sharing these statistics—and they don’t have to. In fact, California is actually the only state in the US that requires companies to report such stats by law. And while many argue that these factors give a good idea of a vehicle’s readiness for commercial deployment, others hold that they can “create perverse incentives” and is not the best way to build trust and showcase credibility, due to the lack of detail and context.

A hiccup in the road

Where there’s experimentation, there’s bound to be failure, and Uber knows that well. In March 2018, the major industry player was met with a fatal accident, killing a 49-year-old woman despite having a human safety driver in the vehicle. This called into question whether the number of miles safely driven is really proof enough of a vehicle’s readiness. After all, human drivers are statistically responsible for one severe accident for every 100 million miles driven, and no AV player has reached that many miles yet.

Another factor in the public’s distrust of self-driving cars is the famous trolley problem. Will an AV prioritize its passengers over pedestrians if one or the other will be killed in an inevitable split-second accident? Luckily, we won’t have to mull over this ethical dilemma for long. According to a researcher from Stanford, it won’t be engineers who make these decisions when encoding responses in the design of AVs, but lawmakers. For example, Tesla’s Autopilot highway feature has been programmed to follow the legal speed limit, even if you set it slower to feel safer. This way, the road transportation authority is liable.

The journey ahead

After a decade of heavy investment in the industry, we’ve seen giant leaps of progress—but also some technological constraints. This reality pushes back Elon Musk’s prediction of having “fully self-driving cars” by the end of 2020. For now, companies will be taking one baby step at a time, cruising through the levels of automation carefully. 

An interesting case study to watch is unfolding in Phoenix, Arizona, where Waymo has started offering a self-driving taxi service to the most daring cab-goers. The service has a very limited area of operation, and takes into consideration the complexity of road traffic and weather conditions. For now, Waymo has humans constantly watching the live feed of these vehicles, so if the cars get into a tricky situation, humans get step in. For instance, say a van stops in the middle of a road to unload furniture. This may baffle the AV, so the human overseeing it will send it a signal so it can make the best decision based on the information provided.

In conclusion, there’s still a long way to go and a lot of questions that need to be answered—be it the cost of production or the industries that would most benefit from self-driving cars. The question isn’t really when autonomous vehicles will come into play, but where and how they’ll be best put to use.

 

Being born and raised in Dubai hasn’t stopped Shameek Aswani from feeling 100% Indian at heart. He is a dual-degree student currently pursuing the Master in Digital Marketing at HST after successfully completing his MBA at IE Business School. He believes in learning from everything and everyone, a realization which resulted from having lived in 3 different cities in 3 different countries. A truly global citizen. Connect with him here.