Self-driving cars are starting to evolve

Today was the third day of Ars UNITE, our virtual conference, and the topic of the day was the advent of the self-driving car.

Our self-driving car feature this morning looked at the technological solutions that will soon allow our cars to drive themselves under certain circumstances, assuming the regulations and other policy issues are in place. That piece has seen a lively discussion covering a number of different areas. Ars reader mexaly suggested that “[t]o succeed, robots need only drive better than average humans. That’s not a high bar.”

Some were skeptical that self-driving cars would be safer. caldron writes, “I think it is a big leap to assume a self-drive is better than a human at driving. Sure in certain conditions and in terms of reaction time, but no computers have been able to reach our level of decision making and ability to react in abstract and unpredictable situations, and there is none in the foreseeable future. We make constant micro-decisions all the time. When there is a grey-area situation that requires deduction I am not so sure a computer will be able to react properly.”

However, others pushed back against this line of thinking. Says fdbyrant: “Why do you trust humans to react “properly”? Most people won’t because we are not even trained to and lot of us don’t have regular experience with such situations. If I was to encounter your proposed situation (and hey I can travel and take a vacation like everyone else)—quite frankly somebody is going to probably end severely injured, if not dead.”

jcsparks agreed: “I still maintain that there isn’t enough physical time within a crash event window for human reason, as you call it, to confer an advantage over computer reaction speed. Unexpected situations are exactly that, unexpected, so you’re reliant upon your reactions, not cognition, to deal with them. If some people deal with those better than other people, I would argue that’s probably down to experience rather than reason.”

We followed the feature with a live discussion this afternoon. Joining us were Ogi Redzic, senior vice president for Connected Driving at HERE, and Catherine McCullough, executive director of the Intelligent Car Coalition. Our panelists answered questions on a range of topics, from looking where the technology currently is to the many open policy questions.

The complete transcript

For more in-depth details from the discussion, here’s a complete transcript of today’s event. It has been lightly edited for clarity, and questions from Ars and Ars readers are bolded for easier browsing.

Jonathan Gitlin: Today’s discussion will be about the self-driving car, from the technology that will enable our vehicles to drive us around to the policy questions that have to be answered before that happens. I’ve got some prepared questions for our guests, who will be joining us shortly. But if you have any questions about self-driving vehicles, please submit them as comments. I’ll be pushing those questions into the discussion throughout this session. Joining us today will be: Ogi Redzic, Senior Vice President for Connected Driving at HERE, and Catherine McCullough, Executive Director at the Intelligent Car Coalition.

Ogi Redzic: Hello everybody. It’s great to be here, fun topic today.

Jonathan Gitlin: And thanks to you as well, Catherine

Catherine McCullough: Glad to be here, Jonathan

Jonathan Gitlin: Recent self-driving car demonstrations from Audi, Tesla, Volvo, Mercedes-Benz, and others suggest that the technology for self-driving cars—at least ones that could drive themselves on the freeway—is either here or almost complete. Are there still significant technical challenges that need to be solved before autonomous cars are ready for the showroom?

Ogi Redzic: Most folks working on self-driving cars today acknowledge that there are two key components: 1. the sensors and the “brain” in the car that is executing the self-driving function 2. powerful backend/cloud solution that provides the car with highly accurate map data to help plan next driving maneuvers, beyond the sensor reach. The sensor technology in the vehicle still has some challenges to make it cost effective for larger scale rollout. And the back end will rely on automated crowd validation and event creation. But we believe there are no blocking hurdles. We are 1-3 years away from early use cases.

Catherine McCullough: I think many would say that semi-autonomous driving is already here in the form of crash-avoidance systems such as auto-steering/lane correction, blind-spot monitoring, forward collision avoidance, pedestrian detection, etc. All of these systems take some action instead of the driver. If you we are talking about a more advanced technology than that – where a car can take the place of the driver’s hands and feet, certainly the automakers you’ve mentioned – along with others like GM – seem set to bring them to market quickly (in GM’s case, “Super Cruise” will be on a Cadillac model by 2017). And as Ogi points out, the sensor and mapping technologies seem there as well.

Chris W: Will cars know when they’re incapable of handling a situation and let me know so I can put my slice of pizza down and take over?

Ogi Redzic: @Chris, this is exactly why this technology is being deployed in increments. Today we live in a so called “Partially Automated” era, where you have your adaptive cruise control and such features. Where we are going next is “Highly Automated Driving” where you will still need to be AWAKE and SOBER so that when the car asks you to take over, with few seconds. so yes, you should be able to leave your pizza on passenger seat and take over within those few seconds.. “Fully automated” is a few years after that still.

Catherine McCullough: In general, I think the legal/regulatory/policy field has had to keep up with the technology, which has led the way

@JasonPReagan: Catherine makes an excellent point — self-driving tech will evolve incrementally rather than suddenly. Each year, I would think automakers will add new autonomous functionality to each model. Kind of a “boiling a frog in a kettle” paradigm maybe?

Jonathan Gitlin: That certainly seems to be a theme I’m hearing whomever I’m talking to, Catherine, and one that our readers have raised too. Are there some policy questions that are going to be harder than others to solve before autonomous cars can go on sale? And Brian (from Philly) has a related question:

Brian (from Philly) says: Its my understanding that the technological path is reasonably understood, but the true challenge will be the liability related issues (i.e. what happens when two self driving cars get into an accident). How are we doing on that front? how involved have insurance companies been to date?

Ogi Redzic: Jonathan, you are absolutely correct. Technical discussion in the industry seems to me to be ahead of where our legislation is, and not just in this country but elsewhere as well. Only this year, for example, select few countries in Europe have developed laws to allow limited exploration of self-driving vehicles on public roads. A lot more needs to be done and public and private sector need to work much closer together in years to come. @Brian – Insurance industry is certainly getting engaged. Obviously self-driving cars will have some of liability move to OEMs/Software/Service developers. Industry needs to adjust for that. They know this is coming, and I see signs that they have started preparing themselves for this. But it is really early.

Catherine McCullough: I think the policy questions around autonomous vehicles are very solvable. If, for instance, we are talking about questions of liability, factors such as insurance will play a role. For instance, where semi-autonomous (crash avoidance) technologies are available, some insurance companies are already providing discounts. I think the main challenge when it comes to policy is to refrain from making overly broad or preemptive law or regulations that stifle the development of these technologies – because these technologies save lives, reduce carbon footprints, and have other societal benefits that we should value.

Ogi Redzic: Fully agree with Catherine – so many benefits to encourage these activities.

Jonathan Gitlin: As navigation moves from on-board (unconnected) systems to those provided via the cloud, are we about to revolutionize traffic management? Will navigation system providers be able to work with each individual local authority or will regional/national coordination be necessary? Some of the audience have similar questions too:

James C. says: Is there any potential that self-driving vehicles could make the roads more congested? I hate driving, but I would gladly dump my bus ride if I had a self-driving car that would allow me to mostly not pay attention. I wonder how much such autonomous cars might challenge public transportation?

Michael Kanz says: What worries me most about self-driving cars is that they are programmed to follow the speed limit, which means that in Chicago, you’ll be irritating everyone on the planet because no one drives the speed limit there. Can autonomous cars be made that can “break” the law?

Alex Bellus says: Michael, they’re not necessarily programmed to follow the speed limit. They’re programmed to drive safely with the flow of traffic.

Catherine McCullough: I think the traffic-management aspects are some of the most exciting. The potentials to reduce the time we spend at stoplights, or looking for parking spaces, or optimizing traffic routes — all have the ability to save time, money, and fuel. I suspect some of the answer to that question when folks learn how to properly value the data that makes it possible.

Ogi Redzic: @James – I believe you are touching on a very important subject. Car ownership in the future. You are referring to a fully autonomous (driverless) vehicle that comes and picks you up, drops you off and possibly picks up few other passengers heading in the same direction. Same number of vehicles will be able to transfer significantly more passengers and this will affect public transit system as well. Will you really want to own a car if you live in a city then? I believe fewer will, already shown in growing attraction of car sharing services.

Ryan says: I have a question, feel free to reformat if used: Many car enthusiasts lament new car technologies that remove driver input from the road. To them, driving is as much a passion and enjoyment as it is a mode of transportation. If fully autonomous cars were to become mainstream in the next few decades, do you feel that this will limit or remove a humans ability to drive manually on public roads?

Ogi Redzic: @Ryan, I ask myself this a lot. As our team works on building automated driving technology I find myself wanting to buy a stick shift vehicle next, as I really enjoy the fun of driving. But I do believe the problem of urban congestion trumps the joy of driving, and nobody is really enjoying being stuck in morning commute. But that country road outside of the city – I vote to save that for our fun drive on weekends.

Jonathan Gitlin: An intriguing solution to Ryan’s question was proposed in a recent competition held by Audi; the team from Seoul proposed having semi-permissible driving modes, where you’d be allowed to drive more enthusiastically if road conditions were suitable, as well as a roller-coaster mode for autonomous driving that drifts through the corners (safely).

Catherine McCullough: Ryan, I think this is a great question. Personally, I don’t see a time when driving will be so restricted that folks are unable to drive manually on public roads. I must admit that I am one of those who truly enjoy driving, and I don’t think I will ever purchase a vehicle that has no availability for human input. I might be OK with allowing the car to take over for a few minutes in stop-and-go traffic, but once I’m out of that situation, I want my vehicle back. In other words, while I value their innovative spirit, Google will have to pry my stick shift out of my cold, dead hand.

Another point on Ryan’s question – I think there is value in learning how to drive, in that it is one of society’s best ways to teach people how to responsibly handle power. (We aren’t always as good at this as we should be, but I think there is inherent value in teaching people how to take and handle calculated risk, generally).

Jonathan Gitlin: Improvements to safety, congestion, and energy use are all factors driving the development of self-driving cars. How rapidly do you think we might go from autonomous cars being on sale to them being mandated? Will (for lack of a better term) ‘dumb’ cars and intelligent cars be able to coexist on our roads?

Ogi Redzic: It is very unlikely that there will be a mandate for self-driving cars any time soon. After all, cars are not cheap and replacing the entire population of cars can’t be achieved in few years. So we need to prepare ourselves for the world where both self-driving and human-driving cars will share the same roads. Which also means they will need to adjust their behavior for each other.

Catherine McCullough: On the mandate question, I agree with Ogi – I don’t see any reason for a mandate. There are many benefits of these technologies, but that does not mean it is necessary — or desirable — for the government to mandate them. First of all, I think widespread adoption will happen naturally because consumers will see the benefits for themselves, in terms of saved money, time, frustration, etc. Also, in general, broad mandates tend to stamp out the very investment and research you want companies to pursue.

@JasonPReagan: My worry is that the media will harpoon the emergence of the self-driving trend by making every self-driving accident seem like the Apocalypse. I’ve seen them do this already with commercial drones. How can those who favor autonomous driving systems get in front of this inevitable “frankenstein syndrome” type of overreaction?

John F: It easy (sort of) to imagine an elegant system wherein there are completely driverless cities and all vehicles function robotically as a part of the mass transit system, but then remain old-fashioned semi- or entirely-manual vehicles outside of such cities. IF something like that is the goal, does it really make sense to have to develop a vehicle even more advanced than necessary as an intermediate step?

Ogi Redzic: @John, the first implementations of Automated Driving will be more highway based (both high speed and congestion commute) than city center. City center is probably most challenging environment to develop for, as there are so many variables – bikes, pedestrians, parked vehicles…

Michael says: Should we expect a backlash when an accident occurs? You can imagine that car accidents will look more like airplane accidents where on average it is much safer but failures are catastrophic. Will people be able to accept such accidents?

Ogi Redzic: Great questions guys. There is no question that media will overplay accidents early on. But this technology is intended to save lives, first and foremost. So it will take some time for people to realize that for each accident that happened, many more will be avoided. But media will also force the industry and government to keep improving, rapidly. There will be teething issues, and they will be addressed.

Jonathan Gitlin: We’ve also had several questions about the maps self-driving cars will need:

Petrut M: How will the mapping business be transformed by self-driving and road updating cars? What will be the main characteristics of those maps.

Ogi Redzic: @Petrut – maps for automated driving are very distinct from maps used for navigation today. They are much more precise as you need to keep vehicles within lanes and you need things like slope, curvature, height, bank of the road. You also need to keep these maps fresh, meaning as soon as one vehicle notices a change on the road our cloud infrastructure needs to process it and make it available to other cars. So map becomes a very living enabler for automated driving.

Chris W: It was recently written that Google cars drive themselves so well around their campus because their campus is so well mapped. How far off is something like a 90% automated cross-country drive where the car can’t be as confident of the exact road state (construction, new roads etc)

Ogi Redzic: @Chris W regarding your question about maps – this is where we spend a lot of our time and effort. We plan to have key roads mapped to the precision necessary for Automated Driving, and we will do it in time for the launch, roughly 2018-2020 timeframe.

John R: With all of this discussion relating to cloud processing and traffic analysis, what about personal security in the midst of all of this? If my (and many other persons) car is constantly reporting where it is over some wireless data connection, then malicious actors have a potential treasure trove of information. Many might be worried about this data being sold to marketing departments without our knowledge, and separately the risk to high-profile individuals should this data be compromised is not trivial.

Catherine McCullough: Re: the security questions, these issues are being examined now. NHTSA just closed a public comment period on V2V, there are other inquiries and initiatives going on re: cyber as well. For instance, automakers just announced a cyber security consortium: http://www.autonews.com/article/20141021/OEM11/141029957/auto-industry-forming-consortium-to-fight-hackers. John R, I think you pose a combined privacy/security question. Re: security, see the answer below. Re: privacy, I think automakers are learning from the tech experience about the importance of establishing a trust relationship with consumers, and I think you will see more news in the near future as automakers work through this issue.

Jonathan Gitlin: Several people in the audience have questions about how autonomous cars will handle ethical dilemmas:

Dave: Some have claimed that self-drive cars might be faced with contingencies where one humans safety must come at the cost of another (like the railway dilemma). Are such situations apparent among current engineering problems/constraints or merely theorized?

Kyle: At some point, due to humans’ inability to consistently perform a complex task like driving errorfree, it seems inevitable that self driving cars will be able to perform this task better than humans. Do you ever see a legal framework evolving which outlaws the current status quo in the name of safety and saving lives? Would it ever be a crime to accidentally cause serious injury to something or someone by driving manually if you ignored the option to use the self-driving functionality?

Duncan A: One popular dilemma associated with autonomous vehicles is the ethics surrounding accidents. How does the vehicle choose between harming pedestrians and harming the driver (in a hypothetical situation where harm is unavoidable).

Ogi Redzic: @Duncan – great question. Not properly answered yet. Work in progress in my opinion

Alex Bellus: @Ogi, what sort of precision is needed with mapping software for autonomous cars? Is it down to the nearest foot? The inch?

Ogi Redzic: @Alex, we are talking about roughly 10cm relative precision.

Petrut M.: Can Open Street Maps be a part, if not the backbone, of this mapping infrastructure?

Ogi Redzic: @Petrut – I believe not. The level of precision required for automated driving can today only be achieved by professional mapping vehicles, like ones Nokia HERE and Google have. These cars use very expensive LIDAR equipment collecting Terabytes of data every day, that needs to be professionally processed. There will be role for community but not in the building of the initial layer, at least not in the beginning.