WINTER, 2014 -- On July 25, 2012, Governor Jerry Brown visited Google’s Mountain View headquarters where he signed into law legislation requiring the Department of Motor Vehicles to adopt regulations governing the testing and operation of automated vehicles. The law stipulated the new regulations be completed by January 1, 2015.
Those regulations will be based in large part on the work of a handful of ITS research engineers, who recently delivered their recommendations to the DMV.
The team of research engineers included Christopher Nowakowski, Han-Shue Tan, Ching-Yao Chan, Wei-bin Zhang, Professor Roberto Horowitz, co-director of PATH, and was led by Steven Shladover. The Berkeley Transportation Letter sat down with Shladover to discuss how his team approached such a large and complex project.
BTL: What did the Department of Motor Vehicles task you to do?
SS: The Legislature instructed the DMV to develop regulations to cover three stages of automated vehicle development. The first is on-road testing that the developers or manufacturers of these systems want to do in order to prove the technology. The DMV has already done most of the work on that itself.
The second set of rules applies to the sale and operation of these vehicles by the general public. So imagine that someone is ready to sell a commercial product that uses automation technology. What are the rules for registering the vehicles? What are the rules for licensing the drivers? What are the rules, in short, to govern the operation of those vehicles and the approval of those vehicles for use on public roads? That’s where most of our effort has been focused. But in addition, the Legislature has asked for rules for when the technology advances to the point that a driver is not necessary.
BTL: Are you talking about fully automated vehicles?
SS: Not necessarily, and here there is great confusion among the public. There are many different levels of automation that have been developed in recent years. The first is automation of a single driving function, such as speed or steering control of the vehicle. For example, adaptive cruise control systems that are already available on various cars will adjust the speed in response to that of the vehicle traveling directly ahead of it. The second level refers to a vehicle that combines automated functions such as speed and steering control. Mercedes has recently introduced a new car that has that capability, but it’s very limited in that it permits the driver to take his hands off the wheel for only a short time. After five or ten seconds, the system will deactivate and warn the driver to re-engage. The idea is to force the driver to be continually paying attention and engaged in the driving. Both levels one and two are already commercially available, on the road now, and not part of what we’ve had to deal with.
Our job was to deal with those vehicles that allow the driver to disengage from the driving task. These are level three and higher systems. A level three vehicle would allow the driver to do something else but still be prepared to take control if the system should get into trouble.
Despite all the mythology out there, the most sophisticated system in existence today is not nearly as capable as any licensed driver. There may be any number of situations or conditions when the system will cease to be useable, and the driver has to be able to re-engage.
BTL: What would be an example of that sort of situation?
SS: Perhaps your vehicle system recognizes lane markings and other vehicles, but can’t compute when faced with flaggers or cones in a work zone. Another challenging situation might be a rainy night when lane markings are not very good and there is glare off the water on the road. In this situation, the system can’t see the lane markings. If you, the driver, can’t see them, neither can one of these systems.
BTL: So the ability of the computerized system to see the roadway is no better than a human’s?
SS: No, not if it depends on the use of machine vision technology. Despite all the mythology out there, the most sophisticated system in existence today is not nearly as capable as any licensed driver. There may be any number of situations or conditions when the system will cease to be useable, and the driver has to be able to re-engage. One of the open questions is how much time is needed for the driver to re-engage. NHTSA (National Highway Traffic Safety Administration) has sponsored research on this subject but those results are not yet available for us to review and consider.
BTL: Are you talking seconds or minutes?
SS: Somewhere in the range of five or ten seconds. For example, say you’re texting a message or checking your tablet and suddenly the alarms go off in the car and you must take control or you’ll crash. Now you are at a stage that is subject to the regulations we’ve proposed because you’re permitted to do something other than drive the car part of the time. I will add, too, that based on what’s going on in the vehicle industry around the world, these applications are really only for freeway driving. We’re not talking about a driverless car ferrying you around local residential streets or downtown Berkeley or San Francisco. That’s way too complicated. This is for a relatively benign freeway environment where the traffic is not too complicated, where you don’t have pedestrians or people opening car doors or backing out of driveways. There’s a lot of work going on within the vehicle industry to develop these kinds of capabilities but none of this is commercially available yet. Still, we need to have the regulations, first for the testing and then for when the vehicles do become commercially available.
BTL: So that is level three. What comes next?
SS: Level four adds a system capability to bring the vehicle to a safe state if the driver fails to re-engage. For example, if you’re reading a tablet or a book and you fall asleep and don’t hear the vehicle’s alarms, you’ve got a serious problem in a level three system. But a level four system will be designed to bring the vehicle to a safe state, even if the driver doesn’t respond. That might mean bringing a vehicle over to a shoulder and stopping, but if there’s no shoulder—perhaps you’re in the Caldecott Tunnel—does that mean the vehicle needs to drive you all the way to the end of the tunnel until it gets to a shoulder? Or does it stop in a traffic lane inside the tunnel, which isn’t generally a safe thing to do? Those are still unanswered questions. So what it means to go to a so-called minimal risk condition is still largely undefined. It’s a concept but it hasn’t been made explicit.
BTL: What about a completely automated vehicle that will act like a driverless cab?
SS: Level five would be a fully automated system that can basically go anywhere and do anything under automatic control, without driver involvement. That’s what gets most of the hype, but in my opinion it is many decades away. Others think it is coming sooner, but many people in the vehicle industry believe it will be 2025 or 2030 before these vehicles are ready, and I think even that is on the optimistic side.
People often use the terms 'driverless cars' or 'autonomous vehicles' to describe these systems, but those are serious misnomers. In most cases these vehicles still have a driver, but the role of that driver could be significantly reduced compared to today’s drivers.
BTL: I’ve noticed that some people, including those at the DMV, refer to “autonomous” or “self-driving” cars, but you use the term “automated vehicles.” What is the difference?
SS: People often use the terms “driverless cars” or “autonomous vehicles” to describe these systems, but those are serious misnomers. In most cases these vehicles still have a driver, but the role of that driver could be significantly reduced compared to today’s drivers. The word “autonomous” is often used by people in the robotics world to signify complete automation, but that is a serious abuse of the English language. “Autonomous” means independent or self-sufficient, so it refers to a vehicle that obtains its information by itself, using only its own onboard sensors, and it operates independently of all other vehicles and the infrastructure. By contrast, cooperative vehicles also obtain information by communication with other vehicles and with the roadway infrastructure and can negotiate with those other entities to coordinate their actions, thereby enhancing traffic smoothness, efficiency and safety. The existing collision warning and control assistance systems on the market are autonomous, but they have very low levels of automation. In the future, we are likely to see fully automated systems that are cooperative, but less likely to be autonomous, since that would severely limit their ability to collect and exchange vital information.
BTL: How much standardization is there among automobile manufacturers developing these new systems and vehicles?
SS: Even within levels three and four there will be a lot of diversity. Manufacturers are not all going to develop the same thing, and that is one way in which this gets complicated. It’s difficult to define a common test protocol that will apply to vehicles that are designed to do significantly different things.
We will need to think about what type of driver training may be needed. And because one system may not be like another, we will also need to think about what type of education drivers will need in order to understand what their vehicles can and cannot do. Will there be any special type of driver’s license endorsement required? What would the requirements be for obtaining that license endorsement?
BTL: Different states are developing their own rules and regulations. How standardized will they be?
The DMV has been pushed into a situation of having to do something more quickly and not as well supported as what NHTSA will eventually do, but the State Legislature believes it is necessary to protect public safety in California.
SS: Establishing regulations for the safe design and construction of vehicles is normally the responsibility of NHTSA at the federal level. But NHTSA has a very slow, methodical, and careful process for developing regulations, so it will take them a long time before they finish their work. When they do, however, their regulations will supersede those of the state. The DMV has been pushed into a situation of having to do something more quickly and not as well supported as what NHTSA will eventually do, but the State Legislature believes it is necessary to protect public safety in California.
BTL: Haven’t Nevada and Florida passed legislation like California’s?
SS: Nevada, Florida, Michigan and the District of Columbia have passed legislation, but none has gotten into this in nearly the depth that California has. Nevada issued a set of rules that are pretty basic and they’re not focused on public safety issues.
BTL: How do you develop regulations for vehicle systems that do not yet exist?
Keep in mind that the automobile industry has been dealing with safety issues for many, many decades. They have well-established processes for designing and testing the systems to make sure they solve safety problems. They know that the consequences are severe if something gets on the market that is not safe, and they go to great lengths to avoid that.
SS: We rely on engineering judgment based on experience with similar systems.
We’ve also spent a lot of time meeting with people from companies developing the vehicle systems, talking to them about their development processes and their assumptions about how to provide for safety with their systems. We’re trying to make sure when we develop the recommendations for these regulations that they reflect the best practices of the industry. Keep in mind that the automobile industry has been dealing with safety issues for many, many decades. They have well-established processes for designing and testing the systems to make sure they solve safety problems. They know that the consequences are severe if something gets on the market that is not safe, and they go to great lengths to avoid that. What we hope regulations will eliminate is the possibility of, say, Joe’s Garage hacking together something that’s crude and unsafe and then sending it out on the road. The big vehicle companies understand the necessity for careful management of risks and so do their first tier suppliers.
BTL: Speaking of safety, when you read articles in the popular press the big question revolves around liability. Who is responsible if the car crashes? The driver or the manufacturer?
SS: That’s not within the scope of our project. But I think the people who have looked at the issue thoughtfully and carefully have reached several conclusions. Perhaps the most important is, if the tech people have designed the systems carefully enough such that these vehicles are actually safer than what we’re driving today—which is not at all easy and it’s a big “if”— all the liability concerns will fade away. In addition, automated systems that go out on the market will be equipped with event data recorders (“black boxes”). So if there’s a crash it will usually be pretty clear who or what is at fault.
BTL: Your deadline of Jan. 1, 2014 fell a year ahead of the state’s deadline, which is a year from now. What happens during the next year?
SS: There is an extensive process the DMV must go through in the coming year, including translating our technical recommendations into regulatory language, releasing the new regulations for public comment for a certain number of days, holding public hearings, receiving comments from those hearings and responding to the comments they receive. Then it goes through more layers of approval within the state. That’s all needed to meet the legislative requirement that the regulations be in place by Jan. 1, 2015. Everything had to work back from that deadline.
BTL: What did you all find to be the biggest challenge in this project?
SS: The schedule and the lack of data. There was so much that had to be done in such a short period of time, and in many cases the data simply don’t exist. We’re required to make recommendations on how to handle various aspects of these vehicles, yet there was often no data available that could be used to produce a well-supported scientific answer. In addition, I would say that a big challenge is striking the right balance between having rules that protect public safety, yet won’t be so burdensome that it drives the industry away. Most of the major automotive companies already have research labs in Silicon Valley. This is a situation in which California has the opportunity to be the world center for this type of development activity, and we want to encourage that to happen.
For more information:
See the website for the TRB summer workshop on automation.
Read S.E. Shladover's article, “Automated Vehicles: Dreaming with Deadlines” in Thinking Highways, Vol. 7, No. 3, October 2012, pp. 20 – 22.
-- Christine Cosgrove