Four Ways Self-Driving Cars May Emerge
In 2015, companies said self-driving cars would be “road ready” by 2020. Clearly, we are still a few years away, and we haven’t even settled on how these cars should operate.
A Medium post by Stefan Seltz-Axmacher, the founder of self-driving truck startup Starsky Robotics, provided yet another nail in the temporary coffin for autonomous vehicles. In it, Axmacher documented various issues that led to him to close Starsky in March and said we are still another decade away from a real-world self-driving car.
Axmacher is certainly on trend — we’ve seen a substantial decrease in self-driving investment over the past two years — but there are still reasons to be optimistic.
There are two major hurdles to surpass:
- AI sophistication: In Axmacher’s post, he said artificial intelligence self-driving vehicles use — a pattern matching algorithm — is not sophisticated enough for the real world. While the algorithm excels in controlled simulations, it struggles with edge cases (as seen 1, 2).
- Safety: 90 to 95 percent of road accidents are caused by human error. An assertion made by operators, including Waymo, is self-driving will eliminate these accidents. However, public opinion on self-driving vehicles is very low, as an autonomous incident receives far more news coverage than a regular accident. No federal or state authority will likely approve of self-driving until it can be proved to be significantly less dangerous than humans because of this.
Neither of these are impossible to surpass. Waymo and other operators reported fewer disengagements in 2019, according to California’s DMV. Waymo is ahead, at 0.076 disengagements for every 1,000 driverless miles. Cruise Automation (owned by General Motors) came second, with 0.19 disengagements per 1,000. Zoom, Nuro, and Pony.AI all reported less than one disengagement per 1,000.
If the argument is autonomous vehicle AI will never be mature enough for the real world, Waymo is proving that theory wrong. A disengagement rate that low, for a technology that is still adapting to new environments and adding new features, is impressive.
Waymo does have an arsenal of advantages however, borrowing Google’s massive data center infrastructure to simulate billions of miles and DeepMind’s neural network to improve AI decision making. It may be difficult for smaller operators to meet Waymo’s standard, although most are already achieving less than one disengagement every 1,000 miles. Disengagement is a loaded-term as well; the autonomous vehicle is not always the cause, so the accident per mile may be much lower.
Safety is, in my opinion, more of a communication and perception problem. Like AI, people with less knowledge tend to extrapolate technology and ruminate its extremes. For example, critics have argued a self-driving vehicle would not be able to make an ethical decision on whether to plow into an elderly woman or swerve off and injure a child, as if a human would make the correct decision in this absurd scenario.
Positions are also hardened from never experiencing the technology. When all mainstream coverage of self-driving cars is negative, it’s challenging to perceive the value. Once services roll out to more cities and towns, I expect many of the people that vowed to never step inside an autonomous vehicle will, probably because someone at work did and said it was fun.
It’s going to be difficult, far more difficult than manufacturers, developers, and analysts predicted when the technology was at its zenith in 2015. However, I suspect given time to improve and trial in more cities, the first two obstacles are surmountable.
Once we’re there, the only thing left is the deployment strategy. Here are four ways that could occur:
Passenger taxi services
Uber, Lyft, Didi, and Grab are all working on self-driving vehicles. Taxi services, which are almost all unprofitable ventures, believe the shift to autonomy could be a breakthrough, removing the main expense (the taxi driver) from the system.
It seems like a natural transition, although taxi services will need their own infrastructure (garages) in every city to clean, repair, and house the cars if there aren’t any customers. They could also partner with local garage owners.
In a ‘The Economist Explains series’, they said: “Rather than work everywhere, these [vehicles] will initially operate within geographically limited and well-mapped urban areas. And they will be shared robotaxis, summoned when needed using a ride-hailing app. The first self-driving vehicle you ride in will be shared, for a combination of technological and economic reasons.”
There is a worry that this will lead to more congestion and clogged streets. If the only cost is fuel, taxi services may deploy thousands of autonomous vehicles that roam the streets endlessly or park in strategic locations, to reduce the wait time.
Most technology firms are approaching deployment this way, uninterested in supplying the technology to car brands or manufacturing the cars themselves.
Shuttle services
It’s not definite that autonomous vehicles will ever be granted the same privileges as normal vehicles. We don’t live in a world defined by statistics and facts, but one where people fear technology and enough of a protest or media storm could severely damage the industry. In a scenario like that, shuttles may be plan B.
Shuttle or bus services run the same route every day, picking up people at strategic stops. To avoid edge cases, shuttles may operate routes away from places the AI is likely to struggle or be perceived as dangerous, like schools and town centers.
There’s definitely a world in which taxi services provide faster and cheaper shuttle services for multiple riders, while they wait for regulators to approve taxis. They could also be provided alongside ride-hailing services. Local regulators may also have an easier time approving and policing a single route, than an entire city of robotaxis.
Rent a car
Many do not want to rely on a taxi service, no matter how efficient, for their daily commute. However, giving ownership entirely to the driver comes with its own problems. Instead, car manufacturers may allow customers to rent the vehicle, retaining full control.
Keeping ownership in the hands of the car manufacturer may seem counter-intuitive, but it allows the manufacturer to set the terms of use. This could include no changes to the vehicle, checking it in for tests every few months, and consenting to the collection of real-time data. If the customer owns the vehicle, the manufacturer would be required to obtain consent for tracking, especially in the European Union.
As Daniel Araya said in Forbes: “The current legal assumption is that autonomous vehicles will be purchased and owned by customers. But what is more likely is that AVs will simply accelerate the shift to transportation-as-a-service.” That as-a-service industry is likely to include rental or subscription, for it to become a ubiquitous product.
Full ownership
The scope of self-driving makes it hard not to see industry-wide changes coming. However, it’s not impossible to see car manufacturers acquiring the software and hardware, installing it on their latest models, and selling them through dealerships.
General Motors, which owns Cruise Automation, is aiming to sell autonomous vehicles. Mike Abelson, GM vice president of global strategy, said in 2018 that ownership would be necessary for rural areas of the U.S., where taxi services are unavailable.
It’s unlikely these self-driving cars will be hybrids, in which the rider can become the driver. Most car manufacturers and technology suppliers have said they intend to skip Level 3 autonomy, the step in which the AI hands over to a human in certain situations. In a research paper, Ford proved once a driver’s focus is off the road, it takes a while to re-adjust, which increases the likelihood of an accident. Some of Ford’s test drivers also fell asleep once they turned on autonomous mode.
While full ownership has been the default model for vehicle sales for the past 100 years, it may become too costly to own the vehicle outright, with cheaper taxi and rental services available. That said, convenience does trump cost in the transportation world, so there may be a large market of customers willing to pay over the odds to own their car.
