The industry has reassessed the state of AV technology after several self-driving accidents and a shocking fatality
It’s time to have an honest discussion about self-driving cars. Nobody is likely to start getting a return on their investment in autonomous vehicle (AV) technology before 2025, and a fully self-driving vehicle is unlikely before 2030. Those are both optimistic estimates.
And those estimates are coming from companies in the automotive industry. A panel held at NXP’s event last week elicited some surprisingly candid views on the state of (AV) technology since an Uber fatality and several Tesla accidents.
One panel — “Self-Driving Cars: What’s the Payoff for Carmakers?” — featured two representatives from tech suppliers and two EV OEM executives. They covered ground ranging from the “safety versus cost” debate to “how safe AVs must be” and “how soon” before the public will trust an AV: Decades? Or sooner?
The panel’s assessment had an unusually high level of realism. Still fuzzy, however, were specifics on how tech companies and carmakers plan to recover from the billions of dollars they have poured into the development of AV technologies. Will they ever see a payoff?
The following are my nine takeaways after listening to their panels:
1. Why do we need highly automated vehicles?
We’ve been talked into AVs because, presumably, machine-driven cars can make the roads safer.
Officially, this is all about safety.
According to the World Health Organization (WHO), nearly 1.35 million people die in road crashes each year. Road-traffic injuries are now the leading killer of people aged 5–29 years.
Both automotive and tech industries agree: Reducing these numbers is the No. 1 goal.
Furthermore, by taking unreliable human drivers out of the equation, they promise that AVs will not only keep the streets safe but will enhance mobility for people unable to drive — children, the handicapped, and senior citizens. Robotaxi fleets will cut the cost of cab transportation and truck platooning will reduce air pollution.
In sum, the AV industry is pushing all of the right buttons in hopes of gaining the trust and support of governments and the public. It’s hard to argue against safety.
But this is 2019. The industry has been working on autonomous vehicles, at a staggering cost, for years. Has capitalism suddenly decided to put human safety ahead of profit?
2. This is about the automotive industry’s survival
Tech companies and automakers are jumping through hoops to develop highly automated technologies. Why?
Safety is a MacGuffin — a plot device. Its importance to the plot is not in safety for safety’s sake but in its impact on the car market and carmakers’ motives.
Last week, listening to the NXP forum, it was almost a relief when panelist Wesley Shao, director of Autonomous Driving at EV startup Byton, bluntly stated: “This isn’t about cost. It’s not about safety. This is about carmakers’ survival.”
This sort of honesty pervaded the panel.
Alexander Tan, NXP’s vice president and general manager of Automotive Ethernet Solutions, shared two scenarios of the future AV industry. One is the rise of mobility services with self-driving vehicle technology as the norm — particularly in urban settings. The other scenario is widespread self-driving car ownership.
Two competing factors, in either scenario, are “safety and cost,” Tan said.
If self-driving cars prove to be extremely safe, people will more swiftly adopt AVs, and regulations will be supportive. But right now, it’s not possible for the normal architecture used in consumer-owned cars to compete in cost with a car designed for mobility as a service.
“Unless car OEMs have the innovation and the vision to change the architecture to support these features at the right cost,” Tan predicted, carmakers will remain in the older model and find themselves outcompeted in this AV space.
Now, if that’s not a scary scenario for car OEMs, what is?
3. AVs are still decades away from driving a vehicle at the same level as humans
Vijay Albuquerque, senior vice president of Strategic Initiatives at Luminar, a LiDAR company, said, “The [AV] industry is still in the development stage.” He argued that where we are today with AVs is still nowhere close to the way that competent human drivers drive.
With just two eyes, human beings over the last 100 years have proven to be remarkable at driving cars, Albuquerque said. In contrast, it turns out that it’s not easy for computers to drive cars — especially if they are to drive “reliably not just at 99% but 10 nines after that.”
Set that reliability issue aside for the moment and consider disengagement data, he said.
(By law, people actively testing self-driving cars on public roads in California are required to disclose the number of miles driven and the frequency at which human drivers had to take control, a moment of crisis known as a “disengagement.”)
Albuquerque said, “Based on this publicly available information — the number of interventions per mile — AVs today are still a couple of decades away before they can drive vehicles on their own at the equivalent level of human drivers.”
This timing — whether a couple of decades, a decade, or even closer — is a matter for debate. However, there is no hardware today that guarantees the safety we seek. The software stack isn’t mature, either. Byton’s Shao said, “When it comes down to safety, there is absolutely no sensor that can be omitted from the AVs today. Especially among today’s sensors, none is a strict subset of the other. So with that, you’ve got to throw everything on it; you have to throw in the whole kitchen sink.”
4. The Uber fatality and Tesla accidents cast a long shadow
In discussing AV safety, the panelists referred to a fatal Uber accident last year and several Tesla accidents in recent years.
NXP’s Tan said that after the Tesla crashes, people argued, “Hey, in this particular case, there was an accident, but overall, this [autopilot] saves lives.” In Tan’s view, this is not a compelling argument. If a truck was crossing and a Tesla drove right into it, people would reasonably say, “I would never have done that.” Safety is not an abstract general concept. Consumers will judge AVs’ safety by individual cases, he said.
Byton’s Shao concurred: “Any accident — whether Uber or Tesla — coming out of the autonomous-driving vehicle will be viewed as an issue for the entire AV industry.”
5. To justify AVs on public roads, AVs must be 10 times safer
NXP’s Tan said that the conventional expectations among car OEMs and ride-share companies such as Uber and Waymo is that “AVs must be considerably safer — meaning several times safer than human drivers.”
Byton’s Shao believes that for the general public to accept the AV, it must be 10 times safer than human driving.” He explained that this is not based on scientific evidence. It has to do with human psychology.
Rachad Youssef, vice president of software product management at Nio, agreed that AV safety must be an order of magnitude better than human driving. Why? Psychology, he said.
However, he added two caveats: How do you prove safety, and how does a passenger tool around in a robocar? Youssef stressed, “It is not enough to claim that you have a system that’s 10 times safer. You need to provide a statistical proof that you can reach those numbers. Similarly, you need to know the state of mind of a user. Does he feel a lot safer inside the AV rather than driven by a human?”
Youssef concluded, “Absent that, this technology will simply not be accepted on a societal level.”
6. There will be no payoff for the AV industry before 2025
Everyone agrees that safety is paramount. But even the most optimistic panelist estimates that the earliest AV production in volume won’t happen before 2025.
So how does the industry plan to recover its tens of billions of dollars pumped into AV technologies? Will they ever see a payoff?
Patience, managing expectations, clever segmentation (by starting an AV rollout only in places where it makes practical sense), collecting more data, doing more development and learning, and reuse of automated technologies in other autonomous systems appear to be the strategies. Although none guarantee actual payoffs, they justify the additional investments in AV development over the next five to 10 years.
Nio’s Youssef believes that the payoff will be more intangible. He said that he hopes to “win mindshare” by investing early and being an early entrant to the AV market.
Youssef said that startups need to find specific use cases in which AVs can be applied in a pragmatic way. Without disclosing Nio’s specific plans, he said, “It’s still an early industry … but we think there will be any number of ways to amortize the investment.”
Nobody at this point is predicting a universal AV rollout. Rather, panelists suggested that AV launches will match varying needs in different geographical markets.
For example, Youssef said, “Using AVs makes sense in cities like Shanghai or driving 101 in the Bay Area.”
Byton’s Shao said that when it comes to payoff, it’s first about how much and second about how long. “I think it’s a long process. Everyone must be ready for a marathon.”
OEMs must figure out a way to calibrate the payoff and manage the pipeline, he added.
Calling himself an optimist among car OEMs, he said, “I don’t think anything gets paid off before 2025.” Given the current AV ecosystem, public acceptance of AVs, laws, regulations, and NCAP requirements, the economy of scale for AVs won’t start kicking in before 2025, he explained. Without scale, there’s no reward.
Luminar’s Albuquerque doesn’t see any big payoff before 2025 or even 2030. But he stressed that he is encouraged by the evident popularity of ride-sharing services.
But even for robotaxis, Byton’s Shao countered, “You can’t easily recoup money before 2025. If the scale is not there, software is not ready, safety is not there, and you need to deploy a safety driver — who will be by no means a regular driver but will be an engineer.” In his opinion, the benefits of robotaxis are in “collecting data and improving technologies” while limiting the driving range of AVs in certain areas such as airports and campuses.
7. There will be losers
NXP’s Tan said, “There is no doubt in my mind that money will get back” to the AV industry. But the way to do that is by reusing some of the fundamental AV technologies in other areas. “What’s being built here is autonomy technology that can be used way beyond cars — such as drones or other physical objects acting autonomously.”
But specifically, for carmakers not venturing into different fields, Tan cautioned, “There will be winners and losers in the next 10 to 15 years. And some of them will be pretty big.”
This is already happening, he said. “Look at consolidations between Renault and Nissan or other alliances” popping up in the industry.
Two factors drive these events: electrification and self-driving. “How you manage these two big double whammies” will decide your future as a carmaker, he explained.
All that the sensors needed, the growth of compute power, and what looks like an in-car data center all add cost to AVs. If car OEMs looking at safety and reliability are not managing cost by embracing a new car architecture (like the one being used by a robotaxi vendor), some might hit a dead end and never recoup their outlay, Tan said.
8. Level 5 cars will largely remain an academic concept
The panel was eager to paint the picture of AV progress as “gradual evolution,” rather than presenting a certain date set for Level 5 AV launch.
Nio’s Youssef maintained that Level 5 AVs — able to go anywhere at any time without constraints — will remain largely “an academic concept.” If the industry rolls out Level 4 cars with proven safety, well-defined use cases, and useful applications, it will effectively make the launch of Level 5 “unnecessary,” he said.
But pressed on when the fives will emerge, NXP’s Tan said that Level 5 cars will remain aspirational, with a few development efforts, until 2030. “Some will try out a concept of Level 5 vehicles in the next decade” but most likely in geographically constrained areas such as theme parks.
9. Is there an opportunity to try less safe AVs on the road?
Comparing the safety of the AV with a human driver may not be exactly fair. After all, not all human drivers are equal. Some are drunk, and many are unreliable or too old to drive anymore.
Even if the industry has not yet created an AV 10 times better than that little old lady from Pasadena, one audience member asked: “Isn’t there an opportunity for AVs that may not be 10 times better than a human?”
Byton’s Shao unequivocally said, “The answer is no.” Let’s say that somebody is drunk and wants a robotaxi to take him home. If the robotaxi is not safer than a human driver and ends up in an accident along the way, Shao asked, “What would be society’s response? I don’t think it’s acceptable. People would say, ‘Hey, he should have taken a taxi with a better driver — a human driver — to get home safely.’”
He added, “Regardless, it has to be 10 times better.”
Panel at NXP’s event last week: “Self-Driving Cars: What’s the Payoff for Carmakers?”