The Policy Trajectories of Autonomous Vehicles

University of Pennsylvania ScholarlyCommons

Wharton Public Policy Initiative Issue Briefs Wharton Public Policy Initiative

5-30-2018

John Paul MacDuffie University of Pennsylvania

Follow this and additional works at: https://repository.upenn.edu/pennwhartonppi

Part of the Economic Policy Commons, Infrastructure Commons, Policy Design, Analysis, and Evaluation Commons, Public Economics Commons, Public Policy Commons, Transportation Commons, and the Urban Studies Commons

Recommended Citation MacDuffie, John Paul, "The Policy Trajectories of Autonomous Vehicles" (2018). Wharton Public Policy Initiative Issue Briefs. 54. https://repository.upenn.edu/pennwhartonppi/54

This paper is posted at ScholarlyCommons. https://repository.upenn.edu/pennwhartonppi/54 For more information, please contact [email protected]. The Policy Trajectories of Autonomous Vehicles

Summary Autonomous Vehicle (AV) technology promises to dramatically reduce deaths and economic losses from crashes caused by human error, increase mobility for those with disabilities, and revolutionize the auto industry. Yet legislation to facilitate oversight of the development and deployment of AVs is stalling in Congress. Professor John Paul MacDuffie offers a primer on AV technology policy, and discusses strategies for addressing safety and other public concerns while still facilitating AV innovation in the private sector.

License

This work is licensed under a Creative Commons Attribution-Noncommercial 4.0 License

This brief is available at ScholarlyCommons: https://repository.upenn.edu/pennwhartonppi/54 publicpolicy.wharton.upenn.edu The Policy Trajectories of ISSUE BRIEF VOLUME 6 Autonomous Vehicles NUMBER 4 John Paul MacDuffie, PhD MAY 2018

In March, news broke that a self-driving car belonging to Uber accidentally struck and killed a pedestrian in Arizona.

It was the first non-passenger death caused by an autonomous vehicle (AV) in the United States. Two SUMMARY years earlier, in May 2016, Tesla made headlines when • As automobile crashes are tremendously costly both in terms one of its cars with automated capabilities collided of human fatalities and economic losses, and typically result with a truck on a Florida highway while in “Autopi- from human error, the development of autonomous vehicles lot” mode, killing the driver who had not responded (AVs) has become a priority at all levels of government. to the car’s sensors beckoning him to reassume full • But the development of AVs has not been casualty-free, thus control of the driving task. That incident will forever raising a key question: How do we best promote AV innovation 1 be remembered as the first “self-driving” car death. while ensuring public safety? Despite the daily tragedies that unfold across the • The current flexibility states enjoy in regulating AV technology U.S., which witnessed 37,461 driving-related deaths and safety, while desirable from the perspective of supporting in 2016, stories like these become breaking news exploration and experimentation for learning purposes, has 2 because of their intrigue. A future with self-driving created an inefficient patchwork of guidance across the country. cars dominating the streets and highways of America • H.R. 3388, or the SELF-DRIVE Act, would potentially address could mean far fewer driving-related deaths, and it’s this problem by enforcing a uniform standard for both technology a future sought by government (at all levels), industry, and safety, while also barring states from blocking the use of AVs without human controls within their borders. and many of the potential users of these vehicles. But when accidents inevitably occur during these years of • As a general rule, such a paradigm of regulatory consistency technological development, many people reasonably that boosts innovation would enhance safety too, given the iterative nature of software and technology development. ask, “What risks are we (society) willing to accept to advance this technology?” and “What are policy- • Beyond H.R. 3388, there are many tools available to government makers doing to mitigate these risks?” The American lawmakers and regulators to foster such innovation. This brief looks at several, including new voluntary federal policy guid- public will need to answer these questions many times ance, the development of a standardized AV “driver’s license over in the coming years, but the federal government test,” infrastructure investments, and geofencing and other has already made its preferences clear, at least for local policy initiatives. the moment. publicpolicy.wharton.upenn.edu

In September 2016, the U.S. government is trying to stay ahead of of AVs—and highlight how federal Department of Transportation and support this burgeoning tech- lawmakers and regulators can mitigate through its National Highway Traffic nology. The aforementioned 37,461 near-term safety risks while facilitat- Safety Administration (NHTSA) deaths in 2016 represent a 14% ing innovation. 6 published non-binding performance increase from 2014. According to This Issue Brief also serves as a guidance in order to facilitate the NHTSA, “94 percent of crashes can snapshot of the state of AVs in the development of AVs, offering a be tied to a human choice or error,” U.S. at a crucial moment, when the consistent regulatory regime to the so removing human judgment from costs (in terms of human lives) of carmakers and technology companies the driving equation could save many developing self-driving technology are 7 competing to bring self-driving cars lives. Not only that, AVs would allow already materializing but before any 3 to market. This document included increased mobility to those with dis- binding federal legislation establishes a 15-point safety assessment for abilities, revolutionize the auto indus- firm legal parameters. Regardless of developers and model state policy. try, and potentially decrease economic the fate of H.R. 3388, many policy One year later, NHTSA updated its losses from crashes, which may have challenges and opportunities lie ahead. 4 8 guidance, and in the same week, the been as high as $836 billion in 2010. U.S. House passed H.R. 3388—the How exactly we arrive at a future THE CURRENT STATE OF AV SELF-DRIVE Act—a standard set- with fully automated cars is still TECHNOLOGY ting piece of legislation that seeks to largely up for debate, and in this Issue further advance the innovation of this Brief we will examine the inflection In 2017, Waymo—Google’s self- 5 technology. Both the NHTSA guid- points in two sequential phases. In driving car company—filed a lawsuit ance and H.R. 3388 have important the first phase, we will address the against Uber, claiming that the ride- implications, as we will discuss, but benefits and costs of setting uniform sharing company stole trade secrets what these measures make immedi- standards—for both technology and when a former Waymo employee ately clear is that federal lawmakers safety—versus allowing for flexibility left to work for Uber. The case was and regulators are working alongside at the state or local levels in establish- intriguing, both for some of its odd private industry to boost innovation ing these standards. As part of that details, but also because of the trade in ways that, in other policy contexts, discussion, we will review the current secrets at the heart of the dispute. they so often do not. legislation proposed in Congress and Uber allegedly stole Waymo’s Lidar There are concerns from consumer the actions undertaken by several designs. Lidar is a highly specialized advocates and other parties that these states up to this point. In phase two, sensor that, in conjunction with cam- proposed measures put the safety of we assume that public support for eras, radar, and various other sensors, 9 Americans at undue risk, and we will developing AVs holds at least until helps a self-driving car “see.” Lidar address these in turn. But there is no 2021—the year many companies have estimates distances (from lane mark- disagreement as to why the federal promised to deliver large scale rollouts ings and road edges) by using illumi-

NOTES

1 Less than a week after the recent Uber accident, a Tesla mated Vehicles Policy, available at https://www.transporta- despite the addition of more safety features to passenger crash—again caused by an Autopilot error—resulted in tion.gov/AV/federal-automated-vehicles-policy-septem- cars, and despite the steady reduction in accident-related the death of another driver. Executives from other com- ber-2016. deaths in the decades prior to 2014. For data on U.S. panies were quick to point out that such incidents are not 4 U.S. Department of Transportation (2017), Automated Driv- accident death rates, see https://en.wikipedia.org/wiki/ representative of the AV industry as a whole, with some ing Systems 2.0, available at https://www.nhtsa.gov/sites/ Motor_vehicle_fatality_rate_in_U.S._by_year. (e.g., Waymo CEO John Krafcik) going so far as to say that nhtsa.dot.gov/files/documents/13069a-ads2.0_090617_ 7 Singh, S. (2015), Critical reasons for crashes investigated their company’s technology would have prevented these v9a_tag.pdf. in the National Motor Vehicle Crash Causation Survey. accidents. 5 H.R. 3388 available at https://www.congress.gov/115/ (Traffic Safety Facts Crash Stats. Report No. DOT HS 812 2 See https://crashstats.nhtsa.dot.gov/Api/Public/ViewPubli- bills/hr3388/BILLS-115hr3388eh.pdf. 115). Washington, DC: National Highway Traffic Safety cation/812451. 6 The uptick in accident-related deaths, which many attri- Administration. 3 U.S. Department of Transportation (2016), Federal Auto- bute to increased instances of distracted driving, occurred 8 See https://crashstats.nhtsa.dot.gov/Api/Public/ViewPubli- 2 publicpolicy.wharton.upenn.edu

nated, invisible lasers. Unlike the other they must accelerate, brake, and about how ultimately to attain full hardware upon which self-driving cars maneuver on the road without human autonomy. Toyota, for example, sub- rely, Lidar is expensive to mass-pro- intervention. All three steps pose scribes to the idea of “human in the duce, although great strides have been problems, technologically and policy- loop,” which involves human drivers made in the last few years to make it wise. The abruptness with which doing most of the driving most of the more affordable. In February 2018, just Waymo settled its case with Uber time, but the automated system (AS) five days into the trial, Uber settled out underscores the fact that hardware would kick in during a dangerous situ- of court, paying Waymo $245 million alone is not the greatest obstacle facing ation, like fishtailing. Tesla’s approach and granting its competitor oversight these companies. It is the algorithms is essentially the opposite: the AS will rights for reviewing the future devel- undergirding the software in self- handle all routine driving scenarios 10 opment of Uber’s Lidar technology. driving cars that will be an especially but will alert the human driver to Uber initially offered to settle for a significant focus of regulation in the retake control of the wheel (through a larger sum, but Waymo accepted the coming years. combination of bright lights, audi- relatively lower settlement amount in Since even a brief scan of news tory alerts, and physical sensors in the exchange for the ability to monitor the articles reveals that terms like “auto- seats) whenever unpredictable circum- way Uber develops Lidar on its own mated,” “autonomous,” and “self-driv- stances develop. Meanwhile, Waymo and integrates that hardware with ing” are used interchangeably—despite and Ford believe Level 3 automation its software. different people having different is not even feasible due to safety con- In general, AVs are the result of interpretations of each term—it is cerns. They argue that forcing a dis- successful hardware-software integra- important for there to be a consistent tracted rider to assume control of the tion and the melding of digital and means of describing the many levels of driving task and to monitor the road 12 automotive components. As Rahul automation. Accordingly, NHTSA environment in a matter of seconds is Mangharam of the University of has adopted SAE International’s defi- too risky, especially at higher speeds. Pennsylvania describes, AVs continu- nitions for distinguishing the different For this reason, Waymo cars will not 11 13 14 ously execute a three-step process. levels of automation. All companies even have steering wheels or pedals. First, they must perceive all of their competing to market AVs—whether Most carmakers, however, have been surroundings. Then, using all of the they are carmakers like Ford, GM, or silent on this debate, implying that data gathered by the hardware, they Daimler, or technology companies like most believe Level 3 automation is must plan their routes and how they Waymo, Uber, or nuTonomy—have achievable and marketable in incre- will navigate the changing landscape. accepted these definitions as their tar- ments. In theory, Level 3 cars could For this, AVs need advanced process- gets, despite there being disagreements reach the market faster, as they have ing computers and complex software about whether Level 3 automation is fewer NHTSA regulations with which to guide their decision-making. The even marketable (see Figure 1). to contend. The first challenge for final step is the act of driving itself: Carmakers have different beliefs either Level, however, is the current

NOTES

cation/812013. provides a concise list, which includes the following: auto- because of excessive liability and the lack of governmental 9 Cameras are best for object recognition; radar and LIDAR are matic emergency braking, adaptive cruise control, adaptive permission. See Michael Taylor, “The Level 3 Audi A8 Will both good at distance; but only LIDAR (short for light direction lighting, backup cameras, park assist, forward collision Almost Be the Most Important Car in the World,” Forbes, and ranging) provides sharp 3-D images, accurate distance, warning, lane keeping assist, lane departure warning, September 10, 2017. and high ability to recognize objects in all light conditions. blind spot detection, and automatic crash notification (e.g., 13 U.S. DOT, supra note 3. 10 See https://www.uber.com/newsroom/uber-waymo- OnStar). Automated cars have been available to consumers 14 Will Oremus, “The Big Problem with Self-Driving Cars,” settlement/. for a long time. The lab and road tests that companies are Slate, September 8, 2015. 11 Jason Laughlin, “Can driverless cars be safe? Grand Theft Auto conducting now, in fact, involve developing Level 3 and 15 For the source of Figure 2, see Gabriel Weiner and Bryant helps Penn scientists find out,”Philly.com , December 6, 2017. Level 4 capabilities. Audi has claimed to have even cre- Walker Smith, Automated Driving: Legislative and Regulatory 12 Today, there are numerous examples of Level 1 and Level ated the first car with Level 3 automation, but they assert Action, available at cyberlaw.stanford.edu/wiki/index.php/ 2 automation present in many cars on the market. NHTSA that they are unable to deploy it anywhere in the world Automated_Driving:_Legislative_and_Regulatory_Action. 3 publicpolicy.wharton.upenn.edu

FIGURE 1 THE FIVE LEVELS OF VEHICLE AUTONOMY is not. At the federal level, it is cur- Execution of Failback System Monitoring SAE Steering and Performance Capability rently illegal to operate AVs without Name of Driving level Acceleration of Dynamic (Driving Environment human controls on U.S. roads. The Deceleration Driving Task Modes) states noted in Figure 2, however, have Human driver monitors the driving environment passed a variety of different rules on No Human Human Human 0 n/a everything from what can be tested Automation Driver Driver Driver and where, to what data AV compa- Driver Human Driver Human Human Some driving nies are required to share with state 1 Assistance and system Driver Driver modes officials for safety monitoring pur- Partial Human Human Some driving poses. And most state laws do include 2 System Automation Driver Driver modes provisions requiring a human driver Automated driving system (“system”) monitors the driving environment “in the loop” to help protect against accidents, although the recent Uber Conditional Human Some driving 3 System System incident is evidence of the risks still Automation Driver modes inherent in road tests. States that High Some driving have higher emissions standards, such 4 System System System Automation modes as California and Massachusetts, also obligate AV companies to Full All driving 5 System System System account for that variable in any new Automation modes AV test designs. The current flexibility states enjoy in regulating AV technology lack of legal parameters. Which states cars on their roads. The map in Figure and safety, while desirable from the or localities will even allow them on 2 shows where legislation has and has perspective of allowing for exploration their roads? not been passed that would authorize and experimentation for learning pur- 15 AV testing. Arizona (red on the poses, has created an inefficient patch- PHASE 1: THE CHALLENGE map because state legislation failed) work of guidance across the country. OF LEGISLATING A UNIFORM has actually been the preferred testing States are free to disregard NHTSA’s STANDARD ground for companies like Waymo and model state policy, of course, and Uber because of active promotion by many have, but in publishing these States have differed widely as to their state and local officials. This is in keep- documents, NHTSA has reaffirmed interest in and approaches toward ing with the general presumption that its authority to oversee the changing 16 allowing the testing of Level 3 and 4 experimentation is allowed . . . until it auto industry. The non-binding guid-

NOTES

16 Specifically, NHTSA has reaffirmed its “authority to identify Are Both More and Less Than They Seem,” The Center for propriate at the state level for a vehicle with no human driver. safety defects, allowing the Agency to recall vehicles or Internet and Society, October 23, 2017. 24 Laughlin, supra note 10. equipment that pose an unreasonable risk to safety even 20 Sarah Light, “Autonomous Vehicle Bill Leaves Safety Gaps,” 25 There is more than one pathway to achieving V2V com- when there is no applicable Federal Motor Vehicle Safety The Regulatory Review, September 25, 2017. munication capability. Two leading contenders are DSRC Standard (FMVSS).” 21 S. 1885 available at https://www.congress.gov/bill/115th- (Dedicated Short Range Communications) technology and 17 U.S. DOT, supra note 3. congress/senate-bill/1885/text. 5G, the next telecom standard. 18 After a concerted lobbying effort from labor groups con- 22 Eric Kulisch, “Safety groups say Congress has given self- 26 See http://archive.darpa.mil/grandchallenge/. cerned about the economic impacts on their members driving companies too much leeway,” Automotive News, 27 See http://fortune.com/2018/02/06/volkswagen-vw- from automation, self-driving commercial trucks were not March 22, 2018. emissions-scandal-penalties/. included in this legislation. 23 Historically, federal law has covered vehicles and state law 28 For a good overview of how AVs will reshape the auto in- 19 Bryant Walker Smith, “Congress’s Automated Driving Bills has covered drivers – hence the confusion about what is ap- surance industry, see John Cusano and Michael Costonis,

4 publicpolicy.wharton.upenn.edu

ance offered by NHTSA in each of the FIGURE 2 STATUS OF STATE LAWS ON SELF-DRIVING CARS past two years has helped to provide some additional assurance to AV companies that the federal government, absent legislation, is eager to boost innovation and is being careful not to set too many boundaries. The language and content of these documents were favorable to companies developing AVs, and most companies will likely operate within the established bounds. Neither document picked winners or losers and both presented uniform standards for algorithms and technology that would be applicable across all states—something manufacturers will No State Laws ultimately need. But just as impor- Passed tant, they upheld currently mandated Under Consideration Failed federal safety standards and confirmed that states should retain their “tra- This map shows the status of all state laws concerning self-driving cars as of April 2018. Additionally, Arizona (2015) and Massachusetts ditional responsibilities for vehicle (2016) issued state executive orders facilitating the testing of AVs on public roadways. Source: Gabriel Weiner and Bryant Walker Smith, Automated Driving: Legislative and Regulatory Action, cyberlaw.stanford.edu/wiki/index.php/ licensing and registration, traffic laws Automated_Driving:_Legislative_and_Regulatory_Action and enforcement, and motor vehicle 17 insurance and liability regimes.” it would bar states from being able to number rises to 100,000 over the next 18 Despite the NHTSA guidance, block the use of AVs without human three years. For their part, manu- current legislative proposals would controls within their borders. States facturers of self-driving cars would restrict the flexibility that states cur- also would no longer be able to set be required to demonstrate that their rently enjoy. H.R. 3388, or the SELF- rules on AV production and testing AVs are at least as safe as traditional DRIVE Act, now awaiting a vote in standards. Under this bill, self-driving cars. They would have to submit the Senate after passing the House carmakers could seek exemptions from “safety evaluation reports” to NHTSA, with unanimous support in Septem- existing safety standards in the first including data related to crashes and ber, would enforce a uniform standard year for up to 25,000 cars—a measure cybersecurity, thereby formalizing the for both technology and safety, and meant to accelerate production. This earlier NHTSA guidance. But they

NOTES

“Driverless Cars Will Change Auto Insurance. Here’s How Insurers Can Adapt,” Harvard Business Review, December 5, 2017. 29 Corporate Partnership Board (2015), Automated and Autonomous Driving: Regulation under Uncertainty, Inter- national Transport Forum (OECD).

5 publicpolicy.wharton.upenn.edu

21 would not need NHTSA approval their preferences with H.R. 3388. If or a bill like it, becomes law, there are of their unique technologies before the Senate uses S. 1885 as a guide, the many opportunities for government bringing their cars to market. next iteration of this legislation could lawmakers and regulators at all levels There are various perspectives include language that preempts even to foster innovation in AV technology on how H.R. 3388 deals with safety. more state functions, including some without singling out automated cars Bryant Walker Smith of the Univer- of the same things NHTSA recently via dedicated federal legislation. State sity of South Carolina School of Law affirmed in its guidance: vehicle exemptions and executive actions, generally supports the bill’s approach registration and licensing, traffic law new voluntary federal policy guidance, to safety, as it follows what he calls enforcement, and regulating insur- state and federal appropriations, and 19 the “public safety case.” This is the ance. But several lawmakers are using local policy initiatives are all tools still idea that as long as “an automated Senate procedures to hold up the AV available to do this. We will discuss driving developer tells the rest of us START Act in committee because of some of them here. what they are doing, why they think it concerns about safety, cybersecurity, is reasonably safe, and why we should and data privacy, demanding explicit 1. JOINING THE LEVEL 3 VS. LEVEL 4 believe them,” this paradigm “encour- testing and access to micro-level data AUTOMATION DEBATE 22 ages innovation in safety assurance from AV companies. In the presence of new information, and regulation, informs regulators, Between the concerns from NHTSA performance and safety and—if disclosure is meaningful— consumer safety advocates and the guidance could change. Thus far, the helps educate the public at large.” He trouble faced by S. 1885, it may be U.S. government has shied away from emphasizes that H.R. 3388 increases that any Senate-passed version of making its own judgment on the the obligations of AV companies rela- the SELF-DRIVE Act would make debate over Level 3 viability. Car- tive to existing federal law. concessions on safety authority. In makers clearly prefer the incremental Sarah Light of the Wharton general, however, regulatory consis- pathway to market offered by Level School, however, elucidates a serious tency across U.S. states that boosts 3 automation, but if this proves too concern shared by many consumer innovation will enhance safety, given risky, the voluntary federal guidance advocates. Specifically, she notes that the iterative nature of software and could shift towards directly favoring in addition to preempting all state technology development. the Waymo approach of bypassing safety standards, H.R. 3388 actually “human in the loop” and shooting for mandates none at the federal level, PHASE 2: NON-LEGISLATIVE Level 4 automation as the next mile- leaving a significant gap in regulation. POLICY OPPORTUNITIES FOR stone. The current guidance is pru- Acknowledging the benefits to inno- PROMOTING AV INNOVATION dent given the information available, vation and economies of scale from because should Level 3 automation uniform technology standards, Light The SELF-DRIVE Act is most likely prove to be safe and marketable, it states, “[P]reempting state action not the final word on how regulators would be an overreach for the govern- even in the absence of federal safety will seek to balance the facilitation ment to already be picking winners. standards fails to take into account of innovation with the ensuring of This debate may end up being one of the significant concerns that states public and consumer safety. Even if the most pivotal strategic, technical, have in protecting their citizens from it becomes law and preempts state and regulatory issues. Then again, from 20 harm.” She proposes that a state safety-setting authority, states and a regulatory perspective at least, it may exemption for safety-setting authority municipalities will continue work- not. Level-specific evaluation may be be added to the final measure. ing to fill the regulatory gaps, as they possible in the future (see #2, below), The bill now moves to the Senate, still have a responsibility for protect- and the government may never need to 23 which had been considering S. 1885 ing their citizens. Conflicts may be interject itself in this industry debate. (the bipartisan AV START Act) in inevitable, but they will be workable. committee, to decide how to reconcile Yet regardless of whether H.R. 3388,

6 publicpolicy.wharton.upenn.edu

2. ENFORCING A SINGLE mas, which is clearly the domain technology companies do. And it is STANDARD FOR PERFORMANCE of public policy when the decision technology companies and small AV EVALUATION AND ETHICAL maker is a piece of software mar- start-ups leading the other camp. DILEMMAS keted to consumers. How society— The original breakthrough Rahul Mangharam’s team of scien- and eventually policymakers—judges moment for self-driving cars was tists at the University of Pennsyl- the ethical soundness of any line of when a research team from Carnegie vania is developing what they call code is, again, still up in the air. Mellon entered and won DARPA’s a “driver’s license test” for self- third and final competition of self- driving cars. Using “mathematical 3. INVESTING IN INFRASTRUCTURE driving cars in 2007 (“Urban Chal- 26 diagnostics and simulated reality,” There are two camps on the ques- lenge”). Identifying the inherent Mangharam’s test seeks to evalu- tion of whether significant infra- software advantages that technology ate the safety of AVs before they structure investment—in the form companies had in turning non-auto- 24 ever reach public roads —a clear of repainted lane markings, repaired mated cars into AVs, Google entered distinction from H.R. 3388’s reli- and increased safety signage, and the fray. Their philosophy involved ance on company disclosures of newly paved roads—would be a turning an AV into an independent “safety evaluation reports” and its boon or a red herring to the devel- cell that would not be reliant upon automatic green light to market opment of self-driving cars. These direct communication from other cars without NHTSA pre-approval. camps are divided on their answers cars or upon upgraded or smart He has argued that independent to the question of how much infrastructure. This approach has federal evaluation of technological connectivity is required to make greatly accelerated the development performance and safety offers the AVs safe. of AVs. most publicly desirable trajectory, In one camp are proponents of The problem with Google’s phi- and these evaluations could develop an older way of thinking about the losophy, however, is that treating AVs and improve as years go by and more relationships between different AVs as independent cells can only move data is gathered and processed. The on the road and between an AV the technology most of the way tests could even be adjusted by auto- and the road itself. They argue that to peak safety. In order to perfect mation level. vehicle-to-vehicle (V2V) commu- self-driving technology, most experts 25 The logic behind giving cars a nication is essential. Not only will believe V2V and/or V2I communi- “driver’s license” is sound, especially we need a single mandated closed cation is still necessary. Even if the as the industry progresses to higher and controlled communication Google approach is 90-95% effective levels of automation. It is anyone’s system for AVs—something akin to at avoiding collisions, there is a very guess whether an independent test air traffic control—but we will need small chance the American public or the reliance on company disclo- to retrofit all vehicles on the road to will accept that outcome, given the sures and data (or both) will ulti- have this capacity, according to this public’s well-established low toler- mately win out. But one scenario position. Furthermore, they argue ance for automobile failures. There will almost inevitably be subject that AV safety requires good “smart” seems to be an opportunity here to a uniform standard for evalua- infrastructure (V2I). In practice, for states to reopen debates over tion, namely the loss-loss situation. it is highly unlikely there will ever infrastructure investment given the Algorithms have already been (and be a single communication system, potential benefits to the future of will continue to be) written to deal in large part because there will be self-driving cars, presupposing they with how an AV will respond to a too many competing algorithms. can get the timing right. Put another scenario where damage, injury, or Besides that, car companies have way, the quality of America’s infra- even death is unavoidable regardless never agreed on standards and they structure is not the red herring it of the car’s actions. These algorithms certainly do not, in most instances, occasionally is made out to be. are plain attempts at solving dilem- view this issue the same way most

7 publicpolicy.wharton.upenn.edu

4. LINKING AV TESTING TO LOCAL both federal rules and any explicit private business transactions, that GOVERNMENT PRIORITIES technological linkage, cities could process could become exceedingly In the face of all of these compet- demand that a certain percentage of expensive. Private automotive insuring philosophies, there is one thing AVs be electric vehicles, at least dur- ance is desirable from the govern- nearly every AV engineer agrees on: ing the testing phase. In this man- ment’s perspective, so regulators may geo-fencing. That is, the practice ner, local governments could push encounter a scenario in which they of limiting the activity of AVs to the industry towards self-driving must mandate access to AV perfor- specific geographic areas that are cars that also happen to be greener, mance data for insurance companies. mapped in great detail (and perhaps should that be a local priority. have dedicated lanes for AVs) is CONCLUSION the best way to advance driverless 5. EXPANDING PUBLIC INSURANCE technology. The need for geo-fencing AND SUPPORTING PRIVATE As this Issue Brief makes clear, is one of the primary differences INSURANCE there are ample technological, safety, between Level 4 and Level 5 auto- Today, private automotive insurance and ethical problems for carmakers, mation. Beyond certain geo-fenced provides a third-party check on the technology companies, legislators, areas, autonomy will not be assured safety of individual drivers. But how and regulators to address before during the years of testing. Cities does insurance work when there are 2021—and beyond. Whether or not naturally are the best-suited envi- no drivers? Assuming that, as cars the U.S. Congress passes legisla- ronments for geo-fencing, and this become fully automated, liability tion mandating some uniform AV presents interesting opportunities for gradually will shift from drivers to performance and safety standards local governments. AV manufacturers, some early-stage in the short-term, there are plenty Some cities such as Austin and ideas include insuring trips or routes of other policy opportunities avail- Pittsburgh have had, in some cases, instead of humans, insuring against able at all levels of government for fleets of AVs on their roads for a cyber attacks, and insuring against boosting innovation and ensuring few years now, despite the fact that product liability like “software bugs, public and consumer safety. Self- their states have passed no explicit memory overflow, and algorithm driving cars may come to dominate 28 AV public testing rules. Cities open defects.” Precisely how this shift the auto market in the next two up their roads to Level 3 and Level in liability will unfold is unknown, decades and completely reshape the 4 road testing for different reasons but it likely will involve some degree way Americans think about and use and with various expectations, and of public insurance, at least during transportation, especially if (as many they likely will continue to be more a period of transition. As a report predict) AV companies introduce active than states in setting param- from the Corporate Partnership their vehicles through their own eters in the short-term. Because they Board states, “Expanding public ride-hailing platforms. There may be offer ideal testing conditions for AV insurance and facilitating greater a dramatic reduction in car-related companies, they have some leverage private insurance could provide crashes, fatalities, and economic to impose demands not necessarily sufficient compensation to those losses as automation improves. And related to autonomous driving. injured by an automated vehicle the economic impacts of AVs on Here is one hypothetical example. while relieving some of the pressure jobs and the economy as a whole are Cities could make AV testing contin- on the tort system to provide such a unclear. All of these questions are the 29 gent upon self-driving cars meeting remedy.” subjects of ongoing research that will certain environmental standards. Automotive insurance companies need to inform how policymakers Ever since “Dieselgate,” there has worry about the potential lack of approach regulating the future of AV been a massive uptick in public inter- access to data. If they are compelled testing and development. est on the topic of fostering electric to purchase AV performance data 27 vehicles. Despite the absence of directly from manufacturers via

8 publicpolicy.wharton.upenn.edu

ABOUT THE AUTHOR

JOHN PAUL MACDUFFIE, PhD Professor of Management, The Wharton School

MacDuffie, in addition to being Professor of Management, serves Prof. MacDuffie’s global research on the determinants of high- as Director of the Program on Vehicle Mobility Innovation, which performance manufacturing is featured centrally in the books The is part of Wharton’s Mack Institute for Innovation Management. Machine That Changed the World and After Lean Production: His research examines the diffusion of lean or flexible produc- Evolving Employment Practices in the World Auto Industry. His tion as an alternative to mass production; the impact of human commentaries on the global automotive industry and trends in resource systems and work organization on economic perfor- employment systems have been featured in the New York Times, mance; collaborative problem-solving within and across firms; Wall Street Journal, Washington Post, Bloomberg Business the impact of distance on managing human and social capital; the Week, Fortune, and on National Public Radio and Knowledge@ relationship between product and organizational architecture; and Wharton. He is a founding board member of the Industry Studies how the interaction of strategic choices with industry structure Association and a member of the Automotive Experts Group at affects competitive dynamics and industry evolution. the Federal Reserve Bank.

ABOUT THE PENN WHARTON ABOUT PENN WHARTON PUBLIC CONTACT THE PENN WHARTON PUBLIC POLICY INITIATIVE POLICY INITIATIVE ISSUE BRIEFS PUBLIC POLICY INITIATIVE

The Penn Wharton Public Policy Initiative Penn Wharton PPI publishes issue briefs at At Penn (PPI) is a hub for research and education, least once a month, tackling issues that are Steinberg Hall-Dietrich Hall, Room 201 engaging faculty and students across the varied but share one common thread: they Philadelphia, PA 19104-6302 University of Pennsylvania and reaching are central to the economic health of the +1.215.898.1197 government decision-makers through inde- nation and the American people. These Issue pendent, practical, timely, and nonpartisan Briefs are nonpartisan, knowledge-driven In Washington, DC policy briefs. With offices both at Penn and documents written by Wharton and Penn 300 New Jersey Avenue, Suite 900 in Washington, DC, the Initiative provides faculty in their specific areas of expertise. Washington, DC 20001 comprehensive research, coverage, and 1+202-870-2655 analysis, anticipating key policy issues on the horizon.

For additional copies, please visit the Penn Wharton PPI website at publicpolicy.wharton.upenn.edu. Follow us on Twitter: @PennWhartonPPI

Founded in 1881 as the first collegiate business school, the Wharton School of the University of Penn- sylvania is recognized globally for intellectual leadership and ongoing innovation across every major discipline of business education. With a broad global community and one of the most published business school faculties, Wharton creates economic and social value around the world.