An Analysis of Economically Efficient Insurance Schemes for Automated Vehicles

After there is more market saturation and thus, more data is collected, the insurance industry will be better able to understand the impact of automation systems and consequently, will b

Brandon Xavier Rhodes

June 2014 Advised by Professor Alain Kornhauser

Submitted in partial fulfillment

of the requirements for the degree of Bachelor of Science in Engineering Department of Operations Research and Financial Engineering

Princeton University

I hereby declare that I am the sole author of this thesis

I authorize Princeton University to lend this thesis to other institutions or individuals for the purpose of scholarly research

Brandon Xavier Rhodes

I further authorize Princeton University to reproduce this thesis by photocopying or by other means, in total or in part, at the request of other institutions or individuals for the purpose of scholarly research

Brandon Xavier Rhodes

Acknowledgements

I would like to first thank my family for being there to support me throughout my Princeton journey and this process in particular Your guidance has been invaluable I would also like to thank Allie, whose gestures of support, from cupcakes to words of encouragement, provided me with persistent motivation Finally, I would like to thank

my advisor Professor Alain Kornhauser who inspired my thesis topic In addition,

Professor Kornhauser provided much appreciated guidance throughout my writing and was always available to discuss ideas or concerns I had Professor Kornhauser, your excitement and passion for automated vehicles is evident both in the discussions we had about my thesis and in your lectures It’s been a pleasure to work with you on this

research

Abstract

Automated vehicles have the potential to have a profound impact on the

automobile insurance industry Preliminary research suggests that various automation systems are correlated with a drastic reduction in the frequency of insurance claims However, American companies have yet to offer unique plans for customers who own vehicles equipped with an automation system, thereby resulting in economic inefficiency

This thesis suggests various insurance schemes that are designed to insure

automated vehicles in an economically efficient manner First, this thesis provides an overview of the automobile insurance industry and an overview of automated vehicles Afterwards, the dilemma automated vehicles pose to current liability laws is also

examined This is because the legal treatment of vehicles equipped with automation systems will have a significant effect on how they are insured Next, before offering suggested insurance schemes, the results of a preliminary analysis on the effect of

automation systems on insurance claims are examined The thesis concludes with a

theoretical analysis of the impact automation systems will have on various sectors of the economy once they achieve greater market penetration

TABLE OF CONTENTS

Introduction 1

Chapter 1: Current Automobile Insurance Schemes 8

Chapter 2: Probability Models 13

Chapter 3: The Legal Dilemma Posed By Automated Vehicles 17

Chapter 4: HLDI Research Study Of Crash Avoidance Systems 28

Chapter 5: HLDI Research Study Results categorized by Crash Avoidance System 65

Chapter 6: Examination of the Availability of Forward Collision Warning System

80

Chapter 7: Suggested Insurance Schemes for Vehicles Equipped with Automation

Technology 97

Chapter 8: Financial Implications of Passenger Vehicles Equipped with Automation Technology 115

Chapter 9: Conclusion 120

Chapter 10: Literary Review 131

Bibliography……… ………… … 135

it is important to first understand the current progress of the technology and the legal implications of introducing cars that drive without human input It is also beneficial to consider how insurance companies currently price insurance plans, because aspects of the current pricing method can be used when pricing plans for autonomous cars

Before continuing, it is important to realize that there are varying levels of

automation Thus, the National Highway Traffic Safety Administration (NHTSA) has released a report detailing what defines these levels The first level is 0 At this level, according to NHTSA (2013), drivers are in complete control of the vehicle and there is

no automation (p.5) At level 1, NHTSA (2013) states that the vehicle has specific automation” (p.5) In the report, NHTSA mentions some examples of functions

“function-at this level such as stability control, cruise control, autom“function-atic braking, and lane keeping (p.5) At level 1, the automation technology is classified as driver assistance technology

Therefore, the driver is to be in control of the vehicle and maintain full attention while driving NHTSA asserts that at this level, the driver is not to have both hands off of the wheel and feet off of the pedals at the same time (NHTSA, 2013, p 5) Level 2

automation “involves automation of at least two primary control functions designed to work in unison to relieve the driver of control of those functions.” (NHTSA, 2013, p 5) Thus, at level 2, the driver is able to have both hands off the wheel, feet off of the pedals, and cede control of primary driving functions to the vehicle (NHTSA, 2013, p.5)

However, the driver is to maintain alertness and must be ready to take immediate control

of the vehicle if necessary (NHTSA, 2013, p.5) Level 3 automation is characterized by vehicles with the capability to “enable the driver to cede full control of all safety-critical functions under certain traffic or environmental conditions and in those conditions, to rely heavily on the vehicle to monitor for changes in those conditions requiring transition back to the driver.” (NHTSA, 2013, p.5) As in the case of level 2, the driver is to be available for taking control of the vehicle, but a key difference is that the driver has to have a “sufficiently comfortable transition time” (NHTSA, 2013, p 5) To achieve this transition time, the vehicle alerts the driver when it is no longer capable of using

automation to drive As NHTSA states, the driver is not expected to monitor the roadway with the same diligence when operating a vehicle with level 3 automation as he or she is expected to use when operating a vehicle with level 2 automation (NHTSA, 2013, p.5) Level 4 is characterized by full automation (NHTSA, 2013, p.5) The vehicle is expected

to carry out all aspects of driving for an entire trip whether a driver is present in the vehicle or not (NHTSA, 2013, p 5) A vehicle system with level 4 automation

technology is treated as a system that replaces a human driver A vehicle with level 4

automation is not expected to drive in circumstances that are not suitable for human drivers, such as severe weather conditions

The goal of autonomous cars is to have these cars drive as if there was human input from the best possible human driver Therefore, the car would minimize the

possibility of having an accident in all situations and would maximize other aspects of driving, such as fuel efficiency In order to do so, the cars must be able to perceive the world around them as humans do Thus, the cars will need to be aware of their

surroundings, the road conditions, traffic lights, lane markings, and so on To achieve this objective, autonomous cars use a combination of sensors that include cameras, radar, GPS, and LIDAR to gather the necessary information from the world around them and to use that information to “see” what is needed to drive in a safe manner The information collected from the sensors is then sent to computers on board the car to be analyzed by algorithms written to understand the information Once the information is analyzed, commands are sent to the systems that control the steering, braking, and acceleration of the car so it can operate in the proper manner For example, a camera mounted onboard the car captures an image of the lane markings on either side of the car, while radar detects an object moving in front of the car The algorithms recognize the image captured

by the camera as lane markings, as another algorithm determines that the object moving

in front of the car can be classified as another vehicle As a result, commands are sent to the system controlling the steering so that the steering wheel moves in a manner that allows the car to remain in the lane Also, a command is sent to the systems controlling the acceleration of the car, directing it to accelerate enough to keep a safe following distance from the car in front of it This example is clearly a simplification of the process

involved However, it is designed to capture the main idea of one of the key capabilities needed for the autonomous car to function properly The process of interpreting the information collected by the sensors is very technical in nature and as such, is not the focus of this paper

The success of autonomous cars is only possible through seamless connections among the hardware sensors, the data they collect, and the sophisticated algorithms needed to process that data, determine the appropriate action, and respond properly The unique hardware necessary for autonomous cars includes sensors to collect information about the surrounding world, along with systems to control basic driving actions, such as acceleration, braking, and steering Currently, the sensors being used by researchers in this field are a combination of radar, GPS, and LIDAR Different companies use different combinations of these sensors, and some companies have attempted to forgo using

LIDAR The algorithms that are used to process the information can be grouped into four main categories These categories are: lane detection, object detection, road analysis, and control of systems involved in basic driving actions The details of the algorithms and the mathematics involved are beyond the scope of this paper Consequently, the paper makes the assumption that the algorithms used by the different car companies to program for automation are uniform in their functionality and reliability This assumption aids in reducing the complexity of pricing the insurance plans, for without this assumption one would need to take into consideration the different limitations of each algorithm used, which may affect the performance of the automated function

At the time this thesis was written, car companies and other players in the field have made significant progress in developing reliable level 4 automated cars The rapid

innovation in this technical domain could arguably be traced to the 2005 DARPA Grand Challenge Since then, automobile manufacturers and other companies have been

working at a feverish pace to develop this technology, so that it can be introduced to the public To aid in the maturation of this technology, select states have legalized testing automated cars on public roadways, after the technology has proven to work reliably on closed test tracks Google has been the most successful company to date Google’s fleet

of cars has driven over 300,000 miles while the technology was operating the vehicles and has yet to experience an accident (Murray, 2012) Google is thus seen as a leader in the effort to achieve level 4 automation Nevertheless, alongside Google, car

manufacturers are working towards generating level 4 automation Car manufacturers differ from Google in that they offer lower level automation systems as driver assistance options on existing vehicles Mercedes Benz exemplifies these options in the additional safety features available on new 2014 S and E class sedans Mercedes Benz has been the leader amongst automotive manufacturers in developing and offering automation

technology, but other manufacturers have begun to follow suit When this research was collected, there were at least 24 car manufacturers that offered at least level 1 automation technology as an option for consumers The addition of these features has a two-fold impact on automobile insurance First, it immediately increases the expected safety of each vehicle and thus reduces the expected frequency and severity of claims for any vehicle owner with this technology in place After there is more market saturation and thus, more data is collected, the insurance industry will be better able to understand the impact of automation systems and consequently, will be able to improve models for vehicles equipped with automation technology Secondly, this addition allows car

manufacturers to amass data on the performance of their automation technology, using lower level systems as a baseline point of comparison in preparation for vehicles

equipped with higher levels of automation systems

Looking to the future, successful implementation of levels 3 and 4 automation technology will most likely involve the abandonment of LIDAR LIDAR has been

extremely useful and performs very well However, it is doubtful its success will be implemented in any role greater than as equipment used for testing While LIDAR is extremely successful and reliable, it is also prohibitively expensive Currently, the

LIDAR Google uses on its fleet of automated cars is approximately $70,000 per unit (Priddle &Woodyard, 2012) In most cases this one piece of equipment is more than the MSRP of current mid range vehicles, and in some cases it is more than the base MSRP of luxury vehicles Unless there is a severe drop in the cost of LIDAR, it would not be cost effective for car manufacturers to equip onto vehicles The cost is too high to warrant the technology as an option for consumers, because the cost prevents high market

penetration Because cameras, GPS, and radar units are inexpensive, it is predicted that a combination of these technologies will be used for vehicle systems offered to the public with levels 3 and 4 automation

This thesis will examine the impact automation technology will have on the automobile insurance industry Automobile insurance has been driver-centric Insurance companies use historical data to assess the risk of insuring a particular driver With the advent of automation technology, automobile insurance will have to shift focus away from a predominant emphasis on individual driver performance to an increased emphasis

on vehicular performance Because the importance of the driver compared to the vehicle

varies amongst automation levels, vehicles equipped with different levels will need to be treated differently by insurance companies As such, the models used to insure the owner

of the vehicle will have to vary accordingly Level 0 is driven purely by a human driver and thus should have a purely driver-centric model; Level 4 is purely driven by the vehicle and therefore should have vehicle-centric model; and since the driving operations are performed by a combination of the human driver and the vehicle in levels 1 through

3, these models should have a corresponding mixture of driver and vehicle elements This thesis will examine different methods of insuring vehicles of each level type In addition, the legal issues automation technology has caused will be discussed How vehicles with automation technology are treated in the legal system will have a direct impact on how the automobile industry insures them Finally, to conclude, this thesis will offer

suggestions for future research

Chapter 1

Current Automobile Insurance Schemes

The objective of insurance companies is to maximize expected profits, while

minimizing expected losses To do so, insurance companies pool policyholders with similar characteristics into groups, or risk classes Each broad risk class is then further subdivided so as to determine a base rate that accurately reflects the level of risk the company is assuming by insuring a member of that group With regard to automobile insurance, these divisions are made on the basis of, but are not limited to, age, gender, marital status, type of vehicle owned, and zip code The company then uses information unique to the consumer, such as previous driving record, to determine the specific rate for that policyholder Insurance companies charge premiums whose total expected sum for the risk pool will be greater than its expected sum of claim payments (Anderson &

Brown, 2005, p.2) This differential is achieved by using historical statistics to predict both the estimated number of claims and the expected severity of each claim for the pool

If an insurance company deems a driver to be too high risk, they reserve the right to refuse to voluntarily insure him or her Insurance companies have adopted various pricing schemes to maximize their expected return After examining the current schemes used, this paper will address whether a variation of the available schemes should be applied to autonomous cars

1.1 Basic Monthly Premium

The most basic pricing plan is a monthly premium Insurance companies take the information they have gathered and determine a monthly price to be paid by customers that maximizes their expected profit By looking at historical data of drivers with similar characteristics, insurance companies predict the expected cost associated with each new customer Examples of the descriptive information used by insurance companies when determining the premium to be charged include the applicant’s gender, age, marital status, zip code, type of car to be insured, and driving history The mathematical models most commonly used by insurance companies will be discussed in detail later in the paper After the applicant supplies this information he or she then can choose from varying levels of coverage Coverage levels range from the state minimum to the

maximum amount of coverage permitted by the company

Another pricing plan is a monthly premium with a deductible This plan requires the insurance company to only reimburse a policyholder if the cost of the accident

exceeded a predetermined threshold (Anderson & Brown, 2005, p.7) If an accident occurs and its expense exceeds the threshold value, the insurance company pays the difference between the cost of the accident and the threshold (Anderson & Brown, 2005, p.7) Typically, the monthly premiums of a plan with a deductible are less than those of a comparable plan without one, thereby incentivizing some policyholders to choose a plan with a deductible (Anderson & Brown, 2005, p.7)

One option for PAYD pricing is to charge a per mile premium (Todd Litman,

2005, p.1) The premium would be calculated in conjunction with the current risk class assessment used by all automobile insurance companies This pricing method therefore improves actuarial accuracy because it uses previous data alongside data of how often the vehicle is driven per month (Todd Litman, 2005, p.1)

However, a more actuarially accurate option is to use incentive based pricing Automobile insurance companies would install a device on a policyholder’s vehicle that records when the vehicle is driven, the time of day, the miles driven, and driving

characteristics such as hard braking and quick accelerations (Iqbal & Lim, 2006, p.5) In fact, Progressive Insurance has experimented with this mode of pricing in the United States by offering their product TripSense to consumers (Iqbal & Lim, 2006, p.3) This method is more actuarially accurate than a per mile method because there is additional information about both the driver and the driving conditions The recorded speed and the

frequency of hard braking and quick accelerations gives insight into whether the driver tends to be more aggressive or defensive The record on the number of miles driven and the time of day reveal how frequent and how long the driver is in conditions that are known to be of high risk for an accident

The most actuarial accurate option is to use GPS based pricing (Todd Litman, p.1) To achieve this option, automobile insurance companies would either use the GPS system already installed on the vehicle or a GPS unit if the vehicle does not have one already The use of GPS technology allows insurance corporations to use data on how many miles were driven, where they were driven, and a what time of day they were driven in conjunction with driver data to determine the monthly insurance premium The new data provided by GPS technology is hugely important, because the risk of an

automobile accident is dynamic and changes depending on the multitudinous factors for which the GPS technology provides information Those factors include whether the vehicle is being driven in an urban or suburban area, at what time of day the vehicle is driven, and how often a vehicle is driven

If a customer elects this option then the insurance company installs a device that monitors the number of miles driven per month The user then pays a premium on a per mile basis Some insurance companies have taken this concept even further and offer discounts based on how the customer drives By collecting additional data such as where one drives, the average driving speed, and the amount of braking force applied, insurance companies can further determine the risk associated with the specific customer This additional information not only benefits insurance companies, but also drivers that elect this option Assuming that only a safe driver would elect to have this information about

their driving habits collected, he or she would now be able to receive previously

unavailable discounts Thus, these safe driving discounts eliminate excess spending on behalf of the customer, while simultaneously providing additional information on which drivers are riskier than others

While Pay As You Drive insurance seems attractive, especially for low mileage drivers, there are nonetheless privacy concerns associated with it An insurance company knowing a driver’s exact location at any given time is disconcerting for many people There are the concerns of who will have access to this information, for how long they will have access, and how securely the information will be held

Poisson distribution will be used in this thesis because it is the most widespread approach

to model claim frequency In addition, the analyses done by the HLDI referenced in chapters 5 and 6 use the Poisson distribution The equations below define the probability mass function, the expection, and the variance for a Poisson random variable X

𝑝 𝑖 =  𝑃 𝑋 = 𝑖 =   𝑒!! !!

!!  ,      𝑖 = 0,1, …      𝜆   > 0

𝔼 𝑋 = 𝜆 𝑉𝑎𝑟 𝑋 = 𝜆

From the equations, it is evident that the accuracy of λ is essential for the

insurance companies because it represents both the expected value and variance of the claim frequency An inaccurate λ could potentially have severe consequences for the profitability of the company As a result, insurance companies implement sophisticated

methods to calculate λ As stated previously, driver characteristics such as the driver’s age, gender, marital status, ZIP code, type of vehicle owned, and previous accident record are used to group drivers of similar characteristics into risk pools These driver

characteristics are then used as explanatory variables in a regression analysis conducted with historical accident data to estimate the appropriate λ for each risk pool of drivers The complexity and type of regression model implemented varies, and the details of the various types commonly implemented are beyond the scope of this thesis As a result, the regression model used by the insurance companies is assumed to be the best option available, as the focus of this thesis is not the effectiveness of different regression

models, but rather the effectiveness of insurance schemes as applied to automated

vehicles Thus, the research explores the effectiveness of the explanatory variables used

in the regression

2.2 Claim Severity

The other component of determining the insurance plan is predicting the accident severity There is a vast amount of data on automobile accident severity; fortunately, this data is applicable to all levels of automation The best approach in assessing accident severity has yet to be determined, but modeling it as a Gamma random variable appears

to be best available option In fact, HLDI used a Gamma distribution when they modeled accident severity in their studies on accident avoidance technology (HLDI, 2011) The Gamma random variable X has parameters (α, λ) The density, expectation, and variance

of X are given in the equations below

𝑓 𝑥 =  𝜆𝑒!!" 𝜆𝑥 !!!

𝛤 𝛼      0 ≤ 𝑥 < ∞

𝔼 𝑋 =  𝛼

𝜆𝑉𝑎𝑟 𝑋 =  𝛼

𝜆!

To increase the accuracy of the model, there should be distinctions based on location For example, it is suggested that there be a variable for vehicles registered in suburban areas and one for those registered in urban area because in urban locations one expects to see multiple accidents that are small in financial severity However, for

suburban locations, one anticipates a small number of accidents that are more financially severe This pattern occurs because the density of vehicles is higher in urban locales while speed limits and congestion prevent the car from reaching high velocities In

contrast, suburban locations experience less vehicle density, and the speed limits are higher This combination, on average, permits cars to drive faster than vehicles in urban locations; the increase in speed increases the accident severity While this assertion seems plausible more research should be done to validate it

Insurance companies may also wish to consider separating the predicted accident severity for vehicles with automation technology from those without it It is suggested to take this measure if corporations see a noticeable trend in the difference of the average severity for vehicles with automation technology compared to the average for those without it It may be that on average, vehicles with automation technology do not have minor accidents, but tend to have accidents that are very severe The trend to have severe accidents could be the result of accidents that only result in a catastrophic failure in either

the hardware or software This theoretical phenomenon can only be substantiated after the technology is introduced to the public and sufficient data is gathered

Companies may also consider classifying accident severity by vehicle

manufacturer and vehicle type For example, foreign vehicles are typically more costly to repair compared to their domestic counterparts In addition, luxury vehicles are typically constructed with more expensive parts and are therefore more expensive to repair after an accident Insurance companies may also wish to take into account the cost and location of the equipped automation technology when estimating the expected average accident severity It is highly unlikely that the LIDAR Google is currently using will be

implemented in vehicles offered to the public But, if it were to be used, the LIDAR system costs approximately $70,000 (Priddle & Woodyard, 2012) Clearly, this cost will need to be taken into consideration when predicting accident severity The location of the technology is also important because if, for instance, it is known that a rear end collision

is the most prevalent accident in an urban area and that the vehicle is equipped with expensive sensors on the rear bumper, it is likely that on average when the vehicle is in a collision the sensor will be damaged and need replacement

is at fault and orders that driver to pay for the damages he or she caused In some cases it

is determined that both drivers are partially liable, in which case both drivers pay a portion of the ensuing damages Concerning automation, as previously discussed, levels 2 through 4 are not available to the public, but levels 0 and 1 are available and in use The courts have determined to treat levels 0 and 1 as additional safety features that do not reduce the liability of the driver using the technology For example, suppose a driver relies on his installed blind spot assist to make a lane change, but the technology does not work and the driver causes an accident The driver will still be liable for the accident, because the technology is designed to be used in a supportive role rather than as a

replacement for the driver’s own awareness of his or her blind spot

According to Dylan LeValley (2013), this is not the case if the blind spot assist on the owner’s vehicle was delivered with a manufacturing defect (p.13) If the accident was due to such a defect then according to product liability law the manufacturer is strictly liable (LeValley, 2013, p.13) Under product liability, if the accident is due to the blind spot assist failing, for the manufacturer to be liable absent a manufacturing defect the

owner must prove a design defect (LeValley, 2013, p.13) However, as LeValley (2013) stated, “design defects are inherent in the design of the product, and are intended”(p.13)

So, to prove the manufacturer is strictly liable, the driver must prove that the design of the product is unnecessarily hazardous (LeValley, 2013, p.13) The decision determining whether a product has a design defect or not is imperfect and subjective As LeValley (2013) stated, “courts … are often weighing the utility of the product, the severity of the harm caused, and the burden on the manufacturer of an alternative design” (p.14) For this example, the blind spot assist would have a design defect if the manufacturer

continued to use an outdated algorithm even after a more reliable one had been

discovered and thoroughly tested

3.1 Common Carrier Laws Applied to Automated Vehicles

Product liability laws are sufficient in dealing with cases involving vehicles with levels 0 through 2 automation technology, but are not able to encompass vehicles with levels 3 or 4 automation technology LeValley (2013) argued that autonomous vehicle manufacturers should be treated as common carriers (p.6) While interpretations vary, it is assumed that LeValley was referring to vehicles with either level 3 or level 4 technology installed Common carriers “are held strictly liable of any injury to goods, but are held to owe the passengers the highest standard of care, bound to “extraordinary diligence…to protect the lives and persons of his passengers.”(LeValley, 2013, p 18) In addition, he asserts that, “under this heightened duty of care, a common carrier of passengers is liable for injuries to passengers resulting from even the slightest negligence.”(LeValley, 2013, p.19)

If autonomous vehicle manufacturers are treated as common carriers in cases of accidents caused by a vehicle equipped with level 3 or 4 automation technology, the manufacturer would be held liable while the vehicle is in automated driving mode It is conceivable that if prior to the accident the owner did not properly maintain the vehicle,

or if the owner had the vehicle operate in severe weather conditions, the owner would be held liable However, it is also possible that the vehicle manufacturers could be held responsible for programming the software so as to prevent such incidents from occurring

In other words, the vehicle should be programmed to cease operation in certain severe weather conditions In addition, the vehicle should not be capable of operating in

autonomous mode if it is not maintained properly through routine visits to the service station and the maintenance of up to date software

It is proposed that the manufacturers of the autonomous vehicle technology be treated as common carriers while the vehicle is operating in autonomous mode

Therefore, for this law to be effective, the technology must maintain a record of when the vehicle is in automated driving mode that can be accessed after an accident

3.1.1 Implications for the Insurance Industry

Automobile insurance has a large financial incentive to encourage car buyers to purchase vehicles with level 3 or 4 autonomous technology if autonomous vehicle

manufacturers are treated as common carriers This is a legitimate contention because there is no risk for companies when their vehicles are driving in autonomous mode Therefore, automobile insurance companies not only want car buyers to purchase

vehicles with this technology, but they also want owners to enable automated driving

whenever possible One may argue that vehicles with level 3 or 4 autonomous technology

do not need to be insured This argument is more valid with regard to vehicles with level

4 autonomous technology because such vehicles are able to operate without any human input However, vehicles with level 3 automated technology will still need to be insured because, there is still the potential for human driving

In this scenario, a new type of automobile insurance may emerge If insurance companies offered to insure autonomous vehicle manufacturers, this would call for a radical change in how automobile insurance companies assess risk and determine rates Instead of focusing on a class of drivers, companies would have to analyze

manufacturers

3.1.2 Implications for the Market

Autonomous vehicle manufacturers will be hesitant to offer these products to the market if they are treated as common carriers in the legal system Autonomous vehicle manufacturers may decide that it is not financially beneficial to put the products on the market if the expected cost of accident claims is greater than the expected gains from product sales If it is a profitable venture, one can predict that the rate at which the

products are offered to the market and the subsequent rate of market saturation will be slower than if the manufacturers were not held to such a high standard of care For

instance, companies that are sufficiently large and a have a high risk tolerance will

predictably offer the technology first In addition, the technology will be offered as an option for only the highest end models available If history serves as an indicator, at first, the option to equip level 3 or 4 automated technology would be offered only on a

Mercedes-Benz S Class Afterwards, the option would be offered on other models within Mercedes-Benz before finally being offered by other manufacturers on their luxury models

Treating autonomous vehicle manufacturers as common carriers would therefore not necessarily prevent companies from both developing and offering this technology to the market LeValley (2013) compared manufacturers automated vehicles to

manufacturers of automated elevators, who are also treated as common carriers (p.11) As one can see today, elevators have been extremely successful in that the market is

sufficiently saturated with them LeValley (2013) argued one cause of the success of the automated elevator was the creation of “elevator liability” insurance (p.11) Thus, to insure autonomous vehicles, manufacturers could implement the model provided by

“elevator liability” insurance Since there is precedent for how to transition from manual technology to automated technology when the technology is treated as a common carrier,

it is conceivable that automated vehicle technology will still be successful even if

manufacturers are treated as common carriers Elevators are not completely analogous to automated vehicles, as LeValley admits (2013, p.9), yet there exists enough similarities such that using the precedent set by the automated elevator industry is a reasonable option

3.2 Product Liability Laws Applied to Automated Vehicles

The American legal system may instead decide to extend the current product liability laws used for vehicles equipped with level 0 through 2 automation technology to vehicles equipped with level 3 or 4 automation technology Under product liability,

manufacturers are only held strictly liable if there is either a design defect or a

manufacturing defect (LeValley, 2013, p.13) This extension of current autonomous vehicular liability law therefore becomes analogous to a “buyer beware” scenario in which the buyer of a vehicle equipped with automated technology assumes liability if the vehicle causes an accident while in automated driving mode

3.2.1 Implications for the Insurance Industry

If manufacturers are held to existing standards of product liability, then there will

be no structural change within the automobile insurance industry Vehicle owners will be held responsible for purchasing automobile insurance, and they will remain the focus of the risk assessments conducted by insurance companies It is assumed that automated driving will be as safe as, if not safer than, the average human driver If this is the case, insurance companies will still have a financial incentive to foster the adoption of this technology amongst car buyers Clearly, however, the financial incentive is not as great

as if autonomous vehicle manufacturers are treated as common carriers

3.2.2 Implications for the Market

The rate at which the technology is offered to the market will be greater compared

to the rate at which the technology is offered if the manufacturers were to be treated as common carriers This assertion is intuitive because with the significantly reduced

liability, the expected cost of introducing the technology will be reduced as well

Therefore, the expected profit accrued from introducing the technology will be much greater Conversely, the expected impact on the rate of market saturation if manufacturers were not treated as common carriers remains unclear Similarly, the impact on car buyers treating automated driving as a “buyer beware” scenario is equally unknown On one

hand, some car buyers may be unaffected by liability concerns and continue to purchase the vehicles at the same rate as if automated vehicles with level 3 or 4 technology are treated as common carriers On the other hand, car buyers may not feel comfortable relinquishing control of driving functions to the vehicle knowing that he or she will be held liable in an accident Jeffery Gurney (2013) supports this claim, stating, “people may hesitate purchasing these vehicles if they bear the liability for accidents caused in

autonomous mode” (p.34) This is not an assertion that car buyers will be hesitant, but rather acknowledges the non-zero probability that this hesitance may occur Thus, more work needs to be done to assess the potential for this prediction to materialize Without having a complete understanding on how the rate of market saturation will be affected, it

is difficult to accurately assess the impact of product liability laws on the financial

success of the automation technology

3.3 Risk-Utility Test Applied to Automated Vehicles

Alexander Herd (2013) suggested that courts use the risk-utility test when

determining liability for an accident caused by an automated vehicle (p.38) As described

by Herd (2013), “the risk-utility test requires the plaintiff to demonstrate that his injuries were a proximate cause of the product design and that the injury could have been prevent

by a reasonable design alternative.” (p.38) To successfully defend themselves,

manufacturers would have to demonstrate that the occasional accidents caused by

automated vehicles are worth the safety improvements and the value the technology creates, using the seven factors of the risk-utility test (Alexander Herd, 2013, p.39) Pertinent examples of these seven factors are: the usefulness and desirability of the

product; the manufacturer’s ability to cost effectively eliminate the unsafe character of

the product without diminishing its usefulness; and the user’s expected awareness of the inherent dangers associated with using the product (Alexander Herd, 2013, p 39-40)

Thus, under this scenario, the courts are charged with conducting a cost benefit analysis of the automation technology in the case of an accident This proposed method can be considered a middle ground between common carrier and product liability

Vehicle owners are held liable unless it is determined that the autonomous technology manufacturers were negligent by producing a product whose benefits do not exceed its costs However, the amount of power and the role the courts play in this regime is

concerning because the courts are charged with doing the cost benefit analysis of the technology Therefore, the courts will, in certain cases, have direct control of what products, or what version of the product, is offered to consumers Since the courts are not experts in the field, it is probable that mistakes will be made when determining whether a technology is meeting the expectations of the risk-utility test Economically, the market tends to determine the most economically efficient outcome, not the courts

3.3.1 Implications for the Insurance Industry

If the risk-utility test is used to determine liability then the automobile insurance industry will not structurally change Instead, it will resemble its current structure

However, insurance companies will have to bear the additional cost of staying

continually informed on the state of automation technology This additional information will become necessary if a suit is brought against vehicle manufacturers for being

negligent in continuing to offer a technology whose benefits no longer exceed its risks

3.3.2 Implications for the Market

The market implications of using the risk-utility test resemble the market

implications of a product liability regime One difference may be that under the utility framework, there is a higher probability that more lawsuits will be brought against autonomous vehicle manufacturers This higher probability is due to the increase in the probability that a plaintiff will win against a manufacturer by being able to disprove the risk assessment conducted by said manufacturer The lawsuits would most likely be centered on the algorithms used by the manufacturers Every algorithm has limitations and the potential for bugs A plaintiff would merely have to prove that the algorithm used unnecessary estimations when doing calculations or was poorly written and was therefore unnecessarily vulnerable to bugs

risk-The increase in the expected number of lawsuits increases the expected cost to manufacturers compared to the expected cost under a product liability regime However, the expected legal cost is not as high as the expected legal cost under a common carrier regime The rate of market saturation may in fact be slightly higher than the rate under product liability This could be the result of vehicle buyers feeling more comfortable because the vehicle manufacturers have an increased incentive to offer the safest

technology so as to reduce the number of lawsuits they incur However, it is uncertain if this increase in incentive would be significant, if there were in fact an increase

3.4 Cost of Litigation

Consistent throughout all potential regimes is the high cost of litigation against an autonomous vehicle manufacturer As Jeffrey Gurney stated, “current law is too cost prohibitive to all for many of the suits; whereas, manufacturers should be liable similar to

how traditional car drivers are liable.” (p.34) Therefore, a single vehicle owner could not bring a suit against a manufacturer Under a regime in which the autonomous vehicle manufacturer has some form of liability, in order to operate effectively the automobile insurance companies will need to handle litigation against the manufacturers This will increase the expected legal cost of insurance companies; however, it is unlikely that the expected cost will exceed the expected savings the technology will bring them The legal system should be altered to allow for both a time and cost efficient manner in which legal suits can be brought against autonomous vehicle manufacturers to ensure they are held to the proper liability standards

3.5 Conclusion

Before cars equipped with level 3 or 4 automation technology can be offered to the public, the legal conundrum such technology poses must first be resolved The legal system is currently founded upon the principle that in an accident one or both parties involved are liable If an automated vehicle causes an accident whether due to a hardware

or software failure, the question becomes whether the owner of the vehicle or the

manufacturer of the technology is liable Two possible solutions are either to treat level 3

or 4 autonomous vehicle technology manufacturers as common carriers, or to extend existing product liability laws to vehicles equipped with this technology Using product liability laws on all vehicles equipped with automation technology would be the simplest solution, but the potential consumer response to this decision is unclear In a similar fashion, the manufacturers response to the technology being treated as a common carrier

is also uncertain, but there are precedents one can examine to predict their responses, such as the elevator example discussed earlier A third solution to the legal conundrum

caused by automated vehicles that represents a partial compromise between the two proposed solutions is to determine liability by the risk-utility test By using the risk-utility test vehicle owners assume most of the liability, but manufacturers are held more

accountable for offering the safest product possible compared to their level of

accountability under product liability laws This solution is problematic because it

requires courts to do the cost benefit analysis It is probable that court’s analysis would be flawed in some regard because the courts are not experts in the technology If the flaws in the cost benefit analysis are severe, they could lead to the courts coming to the wrong conclusion, thus negatively affecting the market As mentioned earlier, it is generally understood that the market determines the most economically efficient outcomes; this regime asks the court to decide these outcomes Which legal laws that will ultimately be used to govern vehicles with level 3 or 4 automation technology is uncertain, and the impact it will have on the market for this technology is equally unclear However, it is evident that how automated technology is treated in the American legal system is not trivial and will have an impact on both the rate at which manufacturers offer the

technology and the subsequent rate of market saturation

Chapter 4

HLDI Research Study of Crash

Avoidance Technologies

In late 2011 and early 2012 HDLI conducted a preliminary analysis of the impact

of automated collision avoidance technology on both the frequency and severity of

insurance claims The scope of the research is limited to select vehicles from Acura, Buick, Mazda, Mercedes-Benz, and Volvo HLDI notes the research conducted has various limitations One of the acknowledged limitations is the unknown status of

whether the automation technology was engaged at the time of the accident or not (HLDI, 2011) The true effectiveness of the technology cannot be discerned without this

knowledge In addition, the accident information presented in the study does not

differentiate by crash type (HLDI, 2011) This impacts the analysis because each crash avoidance technology is designed to avoid or mitigate a particular type of accident For instance, blind spot assist’s purpose is to prevent or mitigate sideswipe collisions Yet, the data on blind spot assist includes both front and rear end collisions in addition to sideswipe collisions Thus, it is indiscernible if an increase or decrease in the frequency

of accidents seen in vehicles equipped with blind spot assist is due the technology itself

or the rates of accidents not meant to be ameliorated by the technology These limitations

do not discredit the study in its entirety, as it serves as a sufficient basis for research into the effectiveness of the technology The limitations do however demonstrate the inherent

uncertainty in the conclusions made As such, the conclusions made by the statistical analysis should be viewed as correlation effects rather than as causation effects

The research done by HLDI was conducted at various times between 2011 and 2012; consequently, results pertaining to each manufacturer were released in separate bulletins Searching through the various bulletins and parsing through the pertinent

information proved to be cumbersome and inefficient To serve as a reference, the results from HLDI’s research studies are reproduced in the sections below The data was also aggregated to facilitate the discussion and dissemination of the information obtained from HLDI’s results

4.1 Acura

HLDI examined the Collision Mitigation Braking System, the Active Front Lighting System, and the Blind Spot Information offered on select Acura models The feature exposure for each of the various models examined in the analysis is shown in the table below

Acura: Feature exposure by vehicle series

Make Series

Model year range

Active Front Lighting System

Collision Mitigation System

Blind Spot Information

Total Exposure Acura MDX 4dr 4WD 2010-11 0 0.12 0.12 42,123

Acura RL 4dr 4WD 2005-11 0.97 0.04 0 174,044 Acura ZDX 4dr 4WD 1010-11 0 0.28 0.28 2,034

The Blind Spot Information and Active Front Lighting System are classified as level 0 automation technologies by NHTSA They are purely driver assistance systems and lack the control of function critical systems The Collision Mitigation System is however considered a level 1 automation technology because in an emergency scenario the system can control the vehicle’s brakes The Collision Mitigation Braking System alerts the driver through auditory and visual warnings if the technology determines there

is a significant risk of frontal collision (HLDI, 2011) If the driver does not heed these warnings the system will initiate automated emergency braking that will mitigate the ensuing damage (HLDI, 2011) It is important to note that while the driver can deactivate the Collision Mitigation Braking System, the system reactivates on every engine startup (HLDI, 2011) Thus, there is a significantly higher probability that the technology will be enabled since the driver must opt out of it at every startup This should warrant a higher discount than a forward collision system that remembers the previous setting, or requires the driver to opt in The results from HLDI’s analysis of the Collision Mitigation Braking System are shown in the table below

Acura : Change in insurance losses for Collision Mitigation Braking System (includes Adaptive Cruise Control)

Vehicle

coverage

type

Lower bound Frequency

Upper bound

Lower Bound Severity

Upper bound

Lower bound

Overall Losses

Upper bound Collision -11.2% -3.1% 5.7% -$452 $31 $567 -$52 -$9 $41 Property damage liability -25.9% -14.2% -0.6% -$323 $69 $523 -$24 -$10 $7 Injury coverage type Lower bound Frequency Upper bound Lower Bound

Low Severity Frequency Upper bound Lower bound

High Severity Frequency Upper bound Bodily injury liability -46.5% -15.0% 35.0% -45.5% 9.8% 121.1% -78.8% -41.3% 62.5% Medical payments -40.8% -3% 58.8% -12.9% 119.5% 453.4% -67.7% -25% 74% Personal injury protection -40.1% -16.5% 16.4% -74.3% -36% 59.4% -42.7% -13.1% 31.8%

Acura’s Active Front Lighting System uses driver steering input to calculate the appropriate amount to turn the headlights so as to facilitate the driver’s area of vision when driving around a curve (HLDI, 2011) Clearly, this is particularly useful when driving at night on a road with low visibility, thereby allowing the driver to clearly see both oncoming traffic and obstacles, such as a deer, in the road Similar to the Collision Mitigation Braking System, the Active Front Lighting System allows the driver to

deactivate it, but the system reactivates every time the headlights are turned on (HLDI, 2011) The results of the analysis for the system are shown in the table below

Acura’s Blind Spot Information visually alerts the driver if a vehicle is detected in

a blind spot (HLDI, 2011) Acura’s technology does not employ any form of emergency braking if the driver does not heed the warning and instead chooses to enter the lane where the other vehicle is located As with the previous crash avoidance systems, the driver is able to deactivate the system However, upon startup, Blind Spot Information does not require the driver to opt out again; rather, it remembers the previous setting before the vehicle was shutdown (HLDI, 2011) This negatively impacts the probability that the technology will be engaged, and as such, should be taken into consideration by

Acura : Change in insurance losses for Active Front Lighting System

Vehicle

coverage

type

Lower bound Frequency

Upper bound

Lower Bound Severity

Upper bound

Lower bound

Overall Losses

Upper bound Collision -11.9% -2% 9% -$466 $12 $556 -$40 -$4 $38 Property damage liability -20.3% -6.3% 10.3% -$418 -$9 $473 -$20 -$5 $14 Injury coverage type

Lower bound Frequency

Upper bound

Lower Bound

Low Severity Frequency

Upper bound

Lower bound

High Severity Frequency

Upper bound Bodily injury liability -38.2% 8.7% 91% -51.9% 39.4% 304.1% -68% -23.6% 82.7% Medical payments -59.7% -28.2% 27.8% -92.1% -25.9% 597.1% -65.5% -24.9% 63.3% Personal injury protection -38.6% -7.9% 38.1% -43.9% 88.7% 535.2% -50.1% -16.7% 39.3%

insurance companies when determining the discount offered for having the technology installed The results of the analysis are displayed in the table below

Of the three systems, the analysis done by HLDI concluded that only the Collision Mitigation Braking System had a statistically significant effect on insurance losses The analysis estimated that the effect of implementing the technology was a 14.2% decrease

in the frequency of property damage claims than expected No other estimated effect was statistically significant Also, no pattern is discernable in the other results because the confidence intervals span a wide range (HLDI, 2011)

4.2 Buick

HLDI examined the Lane Departure Warning, Side Blind Zone Alert, and the Ultrasonic Rear Parking Assist crash avoidance technologies offered on select Buick models; this research study examined the 2008 to 2009 Lucerne These crash avoidance systems would be classified as level 0 automated technology because they assist the

Acura : Change in insurance losses for Blind Spot Information

Vehicle

coverage

type

Lower bound Frequency

Upper bound

Lower Bound Severity

Upper bound

Lower bound

Overall Losses

Upper bound Collision -18.5% -5.4% 9.7% -$523 $315 $1,315 -$70 $3 $94 Property damage liability -34% -16.2% 6.3% -$739 -$187 $512 -$38 -$19 $8 Injury coverage type

Lower bound Frequency Upper bound Lower Bound

Low Severity Frequency Upper bound Lower bound

High Severity Frequency Upper bound Bodily injury liability -47% 24.1% 190.6% -37.9% 116% 651.6% -43.5% 197.3% 1463.9% Medical payments -60% -5% 125.7% -89.6% -37.8% 272.4% -60.7% 41.8% 411.3% Personal injury protection -21.5% 43.1% 161% -81.8% -0.2% 446.5% -26.8% 58.5% 243.3%

driver in operating the vehicle, but they lack the capability to control vehicle systems critical to driving The feature exposure for each technology is shown in the table below

The Lane Departure Warning alerts the driver if the vehicle strays from the

current lane through audio and visual alerts (HLDI, 2011) The system can be deactivated

by the driver, and upon vehicle start up the system remains in the previous setting (HLDI, 2011)

Buick’s Side Blind Zone Alert notifies the driver through visual warnings if a vehicle is detected in one of the blind spots The system can be deactivated by the driver, and upon vehicle start up the system remains in the previous setting (HLDI, 2011)

The Ultrasonic Rear Parking Assist uses both auditory and visual systems to aid the driver in parking The system utilizes a combination of display and a rear-facing camera to show the driver a clear view of the area behind the vehicle (HLDI, 2011) In addition to the visual display, a tone sounds if an object is detected and continues to sound if the object is within one foot of the rear of the vehicle (HLDI, 2011) In

conjunction with the tone, the display visually communicates the distance an object is from the rear of the vehicle through the use of indicator lights (HLDI, 2011) The driver,

as in the other crash avoidance technologies, has the capability to deactivate the

Ultrasonic Rear Parking Assist (HLDI, 2011) However, unlike the Lane Departure

Buick: Feature exposure by vehicle series

Lane Departure Warning and SZBA

Ultrasonic Rear

Warning and Side Blind Zone Alert, the system activates upon vehicle start up regardless

of the previous setting Thus, the driver is required to opt out at every vehicle start up, or

at every parking attempt

HLDI examined the combined effects of the Lane Departure Warning and Side Blind Alert systems on insurance losses, while Ultrasonic Rear Parking Assist was

examined separately The analysis concluded that the combination of Lane Departure Warning and Side Blind Zone Alert had no statistically significant effect on insurance losses (HLDI, 2011)

Conversely, Ultrasonic Rear Parking Assist had statistically significant effects on multiple aspects of insurance loss claims The results of the analysis show that the

frequency of collision and property damage liability claims decreased by 5% and 16.6% respectively Furthermore, the overall financial losses accompanying property damage liability claims decreased by $11, and the technology was associated with a 30% decrease

in the frequency of high severity medical payment claims However, as HLDI notes, there

is no reasonable explanation for the association between Ultrasonic Rear Parking Assist and the decrease in high severity medical payment claims (HLDI, 2011) Though the

Buick : Change in insurance losses for Lane Departure Warning and Side Blind Zone Alert

Vehicle damage coverage type

Lower bound Frequency

Upper bound

Lower bound Severity

Upper Bound

Lower bound

Overall Losses

Upper bound Collision -1.1% 4.2% 9.7% -$212 -$34 $154 -$10 $6 $24 Property damage liability -1.3% 7.2% 16.4% -$138 $46 $247 -$2 $6 $15

injury coverage type

Lower bound Frequency

Upper bound

Lower bound

Low Severity Frequency

Upper Bound

Lower bound

High Severity Frequency

Upper bound Bodily injury liability -24.2% -1.5% 27.9% -33.7% 1.3% 54.9% -38.3% -3.4% 51.1% Medical payments -15% 12.5% 48.9% -25.1% 39.4% 159.4% -32.9% 0.1% 49.2% Personal injury protection -11.6% 11.6% 40.8% -20% 25.8% 97.7% -34.8% -9% 26.9%