Those of us with the Wisconsin AV Proving Grounds regularly find ourselves questioning claims and allaying fears in the same breath. The following article captures this well and is re-posted with permission from Robert Poole, Searle Freedom Trust Transportation Fellow and Director of Transportation Policy, Reason Foundation. (Original Post)
Many Autonomous Vehicle Assumptions Need to be Questioned
As a transportation professional, I’m a bit overwhelmed by proliferating articles and technical papers dealing with autonomous vehicles. And I’m increasingly distressed by the growing disparity between what appears in the popular media and what’s in the technical literature, because the former is what seems to motivate legislators and planners. So here are five challenges to the mass-media version of our AV future.
AVs will save millions of lives. Certainly, we all hope this will prove to be the case. But it’s not as simple as many people seem to think. As Washington Post reporters Michael Laris and Ashley Halsey III explained in a long feature article (Oct. 18, 2016), how will we know how much safer AVs are? Only about half of all crashes are reported to police (as opposed to insurance companies), and some people avoid doing the latter to avoid a possible premium increase. The former head of the National Transportation Safety Board, Mark Rosekind, said that widespread use of AVs should not proceed until they were demonstrated be “much safer” than conventional vehicles, but without comprehensive data on actual accident rates, it’s difficult to make such a comparison. Also, researcher Tom Dingus of the Virginia Tech Transportation Institute points out that alert, attentive, sober drivers are very low risk. It’s drunk drivers plus very young and very old drivers that drive up the averages. In addition, the most promising form of vehicle automation relies on machine learning—but even the experts in that field have no idea what or how the machine actually learns. Matthieu Roy of the National Center for Scientific Research in France says that, “You would never put [a machine learning] algorithm into an airplane because you cannot prove [to regulators] the system is correct.”
AVs Will All Be Connected Autonomous Vehicles (CAVs). A January 3, 2017, Wall Street Journal news article called “Wiring Streets for Driverless Cars” presented this conventional wisdom, which is being promoted by the National Highway Transportation Safety Administration with its Vehicle to Infrastructure (V2I) efforts. Reporter Paul Page touted new digital signs on a freeway near Washington, DC “as a first step toward what highway planners say is a future in which self-driving cars will travel on technology-aided roads lined with fiber optics, cameras, and connected signaling devices.” But unlike lower-brow media, Page went on to note that many billions of dollars would be needed to wire more than 4 million miles of paved roads and 250,000 intersections. If AVs depend on that kind of infrastructure investment, don’t bet on an AV future. Fortunately, most AV researchers don’t think anything like that is necessary. AVs, especially those with full (Level 5) autonomy, will need to be self-sufficient even in alleys, on gravel roads, and on countryside dirt roads.
Full, Level 5 Autonomy Will Be Here within a Decade. In addition to academic researchers such as UC Berkeley’s Steven Shladover who projects Level 5 (all types of roadway, all weather conditions, no driver needed ever) as a 2075 phenomenon, a number of technology-literate commentators have begun throwing out caution flags. For example, telematics blogger Michael L. Sena headlined a recent issue of his The Dispatcher: “SAE Level 5 Driverless Cars Are Not Just Around the Corner.” He took issue with recent reports claiming that Transport as a Service (requiring Level 5) will be ubiquitous by 2030 nationwide. He also cited a thoughtful analysis by The Economist, headlined “Forget hype about autonomous vehicles being around the corner—real driverless cars will take a good deal longer to arrive.” Wired’s Aarian Marshall had a piece in February explaining “Why Self-Driving Cars *Can’t Even* with Construction Zones,” discussing very real problems with machine learning. He also noted an announcement by Nissan that its current plans don’t include Level 5; instead, they assume a human occupant who can take over control when the AI cannot cope, and the human can contact a Nissan call center for help.
Fleets Are the Future, Not Owned AVs. A recent “analysis” by Governing magazine, summarized in the August issue, sounds the alarm that cities’ budgets are seriously at risk from the impacts of the transition to AVs. For the largest 25 cities, the magazine’s team collected data on parking revenues, fines and citations, traffic camera fines, gas taxes, vehicle licensing fees, etc. Many of these cities each year generate several hundred dollars per capita from these vehicular revenues, much of which could disappear in the AV future, warns the article. Except—much of the impact stems from the assumption that shared fleets of robo-taxis replace individually owned cars, thereby eliminating most urban parking requirements (and hence parking revenue). Another built-in assumption is that most or all AVs will be electric, which is hardly a given. Robotaxis and individually owned AVs that can go elsewhere after dropping the owner off require Level 5 automation, which is hardly a near-term phenomenon. So city officials should not begin losing sleep over plummeting parking revenues.
AVs Will Reduce Congestion. As I’ve pointed out in previous issues, the majority view among AV researchers is that the transition to AVs will increase vehicle miles of travel (VMT), for a variety of reasons including bringing personal vehicle autonomy to millions who are unable to drive today, as well as reducing the time cost of commuting. But a related idea remains—that due to reduced distance between AVs on roadways, existing roads will be able to handle greater volume with less congestion. But even popular media are starting to consult experts who disagree. Business Insider recently interviewed Lew Fulton, co-director of UC Davis’ Institute of Transportation Studies. He expressed particular concern about zero-occupant (Level 5) vehicles being a new source of increased congestion. Such vehicles, programmed to run errands, deliver packages, etc. will lead to far more cars on the road. Fulton calls them “zombie cars.”
Before leaving these points, I want to recommend a more thoughtful article. Despite its misleading title, “The Road Ahead for Connected Vehicles” (when the article actually discusses AVs, not CAVs), is a sober discussion by industry experts—established auto industry people, high-tech AV pioneers, and consultants. It’s a product of Wharton’s Program on Vehicle and Mobility Innovation, which has absorbed the former MIT International Motor Vehicle Program. If you have time to read just one article to get a more balanced view of this challenging area, this one is hard to beat. (http://knowledge.wharton.upenn.edu/article/road-ahead-connected-vehicles)