Transportation Planning Casebook/Autonomous Cars

Summary
Autonomous cars are cars that can detect what is going on around them in order to drive from A to B without humans actively controlling them. They are also known as robot cars and driverless cars. Some features of autonomous cars, such as cruise control and parallel parking assistance, have already been incorporated into traditional vehicles. Fully autonomous cars have not been deployed yet to the general public but technology companies, most notably Google, are actively developing and piloting fully autonomous cars that can be programmed to go to a destination and then carry the rider there. Autonomous cars have the potential to significantly improve safety on the roads but they come with a number of ethical and legal hurdles that still need to be worked out.

List of Actors
Norman Bel Geddes- Creator of the Futurama display at the 1939 World's Fair and author of Magic Motorways, Bel Geddes was a visionary for autonomous transportation as well as the highway system as a whole.

Defense Advanced Research Projects Agency (DARPA)- a United States Department of Defense agency responsible for the development of new technologies for use by the military and funding the development of many new technologies. The DARPA Grand Challenge raised the profile for autonomous vehicles.

Google- Tech giant currently at the forefront of driverless technology.

Mercedes-Benz- Automobile manufacturer that was an early tester of autonomous technology among the large car manufacturer in the 1980's. Mercedes-Benz is one of many manufacturers today with a working prototype of an autonomous vehicle.

Velodyne- Maker of the HDL-64 LIDAR array, a popular choice for LIDAR technology used on autonomous cars of today.

Timeline of Events

 * 1939- Norman Bel Gedde's Futurama exhibit at the World's Fair
 * 1960- The Transport and Road Research laboratory in the United Kingdom tested the Citroen DS driverless car which used magnetic tracks underneath the road and a pair of magnetic sensors in the front and back of the car.
 * 1969- The French government began researching Aramis, a system that ran on magnetic rails, where drivers would sit in their own driverless carriage and join a ‘train’ of similar ones on the way to work.
 * 1995- A Mercedes-Benz van drove 1000 miles from Munich to Copenhagen using cameras to ‘see’ the road and computers to steer.
 * 1997- DEMO 97 demonstration in San Diego shows driverless cars are capable of following each other at high speeds using communication technology.
 * 2004- First DARPA Challenge related to driverless cars helps to raise the profile of autonomous cars through a prize competitions.
 * 2010- Google announces it has been secretly testing driverless cars in traffic

Maps of Locations
Currently, legislation has been passed in four states regarding the use of autonomous cars: California, Nevada, Michigan, and Florida, as well as the District of Columbia. International legislation has also been enacted to allow testing in Germany, Spain, and the Netherlands.

Early Autonomous Cars
Early autonomous car technology was first modeled in the 1920's through the use of radio-control. In these short demonstrations, Radio impulses were sent from nearby controls and caught by a transmitting antenna which operated an electric motor that controlled the car. The dream of a large-scale, driverless transportation system did not garner attention until 1939's World's Fair where Norman Bel Gedde introduced audiences to the Futurama exhibit, his vision of a not-so-far-off 1960's metropolis area. Futurama was sponsored by General Motors and was visionary in its prediction of a nation-wide highway network that featured driverless technology. Though the prediction of 1960 was too soon, the vision for the future was put into place.

In the late 1950's and 1960's, driverless technology was successfully demonstrated on public roads through the use of magnetic tracks placed underneath roadways. Impulses were used to guide a car and determine the velocity of any metal vehicle on its surface. The successful demonstration of the technology led to government backed research into autonomous cars by entities including the Bureau of Public Roads. Many of the research projects were disbanded because of costs, but not before much was learned about the possibilities of autonomous cars

In the 1980's and 1990's, cameras began being used on autonomous cars as a way for the cars to "see" the road. Additionally, communication technology between cars became another way for cars to "talk" to each other and was a demonstrated as a technical success in DEMO 97. Camera and communication technologies were the forefather for many of today’s most promising autonomous innovations including Google car’s prototype.

Modern Autonomous Cars
Today’s fully autonomous cars are covered with different sensors that map nearby features in order to spot road edges and lane markings, read signs and traffic lights, and identify pedestrians. These sensors can include a combination of GPS, cameras, radar, LIDAR (laser radar which use pulses of light rather than radio waves), and ultrasonic detectors which provide accurate mapping of the surroundings at short range for such uses as parking. LIDAR systems can be very expensive and cheaper options such as radar and cameras may be more viable in the immediate future. Additionally, accelerometers and altimeters provide more accurate positioning than is capable with GPS alone. Some of these sensors have already been implemented to varying degrees in production cars of today. Existing autonomous car aids include:
 * lane-keeping systems which follow road markings and sound a warning and correct the steering if a vehicle starts to drift;
 * adaptive cruise control (ACC) which maintains a constant distance from the vehicle in front;
 * auto-parking systems which can reverse a car into a parking space using the ultrasonic detectors discussed above;
 * emergency braking which applies the brakes if an obstacle is detected in front of the car;
 * satellite-navigation systems.

Google’s driverless car uses a LIDAR systems to build a detailed 3D map of its surroundings. Each time a car follows a particular route, it collects data to update the existing 3D map. Google’s software also ciphers data regarding speed limits and recorded accidents. The car’s roof-mounted sensors can "see" in all directions which arguably grants it greater situational awareness than any human driver. Google’s self-driving cars have logged 435,000 miles (700,000 km) under autonomous control without incident. Challenges remain especially with snow-covered roads and temporary signs near roadwork, but the technology is improving all the time.

How Autonomous Cars Work


The environment which both autonomous and non-autonomous vehicles must navigate in urban areas is characterized by two types of phenomena - static and dynamic. The static environment consists largely of the built environment, infrastructure, street geometries, and other physical constraints such as building placement. The dynamic environment includes all other road users - drivers and other vehicles, bicyclists, pedestrians - as well as semi-dynamic elements such as traffic control devices (i.e. the structure is standardized and static, but the operation and timings may change as a function of traffic volume patterns). Each environment offers a wealth of information layers from which drivers sample and use to make operational decisions in the driving of vehicles.

The goal of autonomous cars is to both replace the functionality of the human driver, but also to improve on the efficiency and safety (or lack thereof) of human drivers. As such, a wide array of technology must be used to pay attention to both the static and dynamic information layers in the surrounding environment. In the typical Google-modified Prius, there are five general layers of sensory input: short/medium-range radar located near each of the four wheel wells, a rooftop LIDAR array, videographic monitoring of traffic lights (placed near rearview mirror), precision GPS, and speed & positioning information from the vehicle’s computer & wheel sensors. The four radar sensors placed near the wheels are designed to monitor other moving objects and vehicles that are not immediately next to the autonomous car, allowing the system to create a longer-range map of vehicle positions and movements relevant to higher-speed environments such as highways and high-speed merging. The rooftop LIDAR array (perhaps the most iconic element of the system), often a Velodyne HDL-64, creates a near-field 3D mapping of the surrounding environment by rotating a unit with 64 independently vertically angled laser emitters and measuring the attributes of reflected light. The front-mounted video camera monitors the presence and progression of traffic lights - this is very important information, as traffic lights can be dynamically timed depending on time of day and traffic flow volume, and information regarding precise signal timing is not normally available to vehicles and their operators. High-precision GPS is used to monitor vehicle progression and to assist in route mapping and changes in routing, and on-board sensing of the vehicle’s speed and computer system information is vital in the case of GPS connectivity failure.

The autonomous system integrates the various layers of information scraped from the environment, and together with pre-programmed limitations on vehicular operation from traffic laws, guides the vehicle forward in its environment. The general operational regime is to create a first-draft route map with GPS information, and then use environmental sensory information to inform the vehicle controllers’ execution parameters. A number of specific guidance challenges exist, and have been addressed, in autonomous vehicle motion, such as lane changing behavior, passing and overtaking, parallel parking, and navigation of parking ramps (Laugier, 2001). Several fully- or partially-autonomous movements currently exist in legally purchasable vehicles, and a few motion regimes have existed for many years in cars - namely, ABS (automatic braking system), but also (parallel) parking assist.

A technical difficulty is associated with the severance of GPS connectivity. Autonomous vehicles use GPS not just for gathering information about vehicle placement, but for crucial route planning - if a vehicle cannot know its current position on the street network, or its current progression along a route, the route progression is possibly hindered. In the case of GPS failure, local route planning can be determined via the locally-sensed speed information of the vehicle, and off-line route information such as segment distance and number and relative positioning of turning points (Kümmerle, 2009). The case of planned environmental GPS severance (e.g. a parking structure) can be mitigated by mapping and storing additional information regarding generic parking structure geometries, floor plans, and visual control points. This allows off-line route planning, for continued operation of the autonomous vehicle.

Safety
One of the most significant advantages to autonomous cars is that they will not be subject to human error. Worldwide, about 1.2 million people are killed each year due to motor vehicle crashes. In the U.S, in 2012, 37,000 people were killed in motor vehicle crashes, and over 90% of those crashes were attributed to driver error. Human error includes problems such as drunk driving, distracted driving (i.e. driving while texting), falling asleep at the wheel, running red lights, speeding, tailgating, and more. According to an All State survey in 2011, Americans drastically overestimate their driving ability. The study found that although 64% of Americans rate themselves as “excellent” or “very good” drivers, 45% admit to having driven while excessively tired, 44% have received 3 or more speeding tickets, 34% admit to texting while driving, and 15% (23% of men) admit to having driven while intoxicated. Autonomous cars, when programmed correctly, would eliminate these serious road dangers. Not only will autonomous cars be free from human error, but they will have the ability to recognizes obstacles and dangers much farther away than humans and will be able to respond much more quickly. Autonomous cars will be designed without blind spots; they will be able to see 360 degrees as well as far into the distance. Additionally, autonomous cars will be able to communicate with each other, so if there is a sudden slow-down on the highway, or dangerous weather conditions, or a road hazard such as ice or an oil spill, cars warn each other to brake early to avoid collisions. The car in front of you could warn you that a child is running towards the street, instructing your car to brake before you ever see the child. Even if you do see the child, the car’s reaction time will be much faster than humans; currently, driverless cars can make decisions 20 times per second, as compared to human driver reaction time which typically ranges from 0.7 to 1.5 seconds.

Congestion
According to Texas A&M’s 2011 mobility study, on average, Americans spent 38 hours per year in traffic congestion. In large cities like Washington D.C., Los Angeles, and San Francisco, the congestion delay per commuter is over 60 hours per year. Congestion delays cost each American commuter an average of $800 additional in fuel costs for a total of $121 billion total nationwide.

Autonomous cars could reduce congestion by reducing the number of car crashes. Additionally, having cars constantly communicating with each other will allow following distances to be drastically reduced, which could pack more cars into the same amount of road space and reduce congestion. Autonomous cars could coordinate routes to use the road system most efficiently. If cars can self-park, the car could drop you off at your destination and find a parking spot far away without you, reducing the significant amount of congestion in cities caused by people circling for parking. Autonomous cars could take on the role of taxis, reducing the need for people to own a car and reducing the amount of space taken up by parked cars.

Fuel efficiency
Autonomous cars improve fuel efficiency by being programmed to accelerate in a more fuel-efficient manner than most humans. Smaller following distance could also reduce air resistance on cars driving close behind other cars.

Time savings
Mean driving time to work in Minneapolis/St. Paul is about 25 minutes (each direction, each day). In more populous states like New York and California, this figure is closer to 30 minutes. Quite simply, if cars drive themselves, people can spend that time doing work, reading, sleeping, or doing any number of things with that time.

Non-drivers
Autonomous cars have the potential to increase travel options for people who cannot drive. This includes large swaths of the population including children, people with physical or mental disabilities, and elderly or incapacitated people. Autonomous cars could drastically improve transportation for these populations, which have traditionally relied on public transportation or on others to drive them around.

Potential for computer malfunction
There is always the potential that a computer malfunction could cause a crash. In order for autonomous cars to pass the technological, legal, and regulatory hurdles to be present on the road in large numbers, it is likely that the technology will be highly refined and safe, but perfect safety is perhaps impossible to guarantee. Consider air travel, which relies heavily on technology and on auto-pilot functions. Air travel is extremely safe, far safer than driving, but not completely without incident. As autonomous vehicles are deployed and people forget (or never learn) to drive, people may not be able to take over in the event of a computer malfunction.

As of 2014, google’s cars have driven over 300,000 miles on California roads without incident, but at this time they haven’t been tested rough conditions including snow, unanticipated construction, and a multitude of other unpredictable and unusual situations.

Cost
At least during the early phases of deployment, autonomous cars are likely to be significantly more expensive than traditional cars. The elements of autonomous cars that have are already in use, such as lane correction (which alerts and even corrects drivers who stray from their lanes), adaptive cruise control (which detect how far away the car in front of you is and change your speed to maintain a certain distance), and automatic braking (which stops the car if the driver is not braking and a collision is imminent), are found only in luxury cars. Autonomous cars, like many transportation technologies, are most efficient and effective when widely deployed. Autonomous cars are designed for a system where there is extensive car-to-car communication. It if it too costly for the mainstream to afford autonomous cars, car-to-car communication may be limited for many years to come.

Job loss
Autonomous cars, when fully deployed, would eliminate the need for drivers, likely including both taxi drivers and truck drivers. Job loss for this entire class of low-skilled workers could be economically significant.

Potential for crime
Autonomous cars would present a drastic change for law enforcement on the roads. Some crimes, such as speeding and driving drunk, could be eliminated. But other, less common but more catastrophic, crimes could be facilitated by autonomous cars. For example, criminals fleeing law enforcement would have their hands free to shoot. A self driving car could be turned into a self driving bomb. Criminals could take over the car-to-car communication network and purposefully cause crashes. Clearly, the security of the computers and the communications network that govern autonomous cars would be paramount.

Enjoyment of driving
Finally, some people like driving. They may like driving for any variety of reasons, including feelings of power, speed, and control. Autonomous cars, fully deployed, would take away the fun of piloting a powerful and fast moving vehicle.

Ethics of Autonomous Cars
When an accident occurs involving vehicles driven by humans, the prevalence of human error or intent lends easily to the determination of fault, and the subsequent legal consequences thereof. Because non-autonomous vehicles historically have required the input and maintenance of control signals from the operators for the duration of vehicle operation, the responsibility and due diligence for careful operation within the law always rests with the operator, except in rare cases of abject mechanical failure. However, when an autonomous car appears to have caused an accident, determination of fault and legal consequences are much more difficult. There are two sides to this - the human occupants of the autonomous car may or may not be legally responsible for the safe operation of the autonomous car, particularly if they are the owner, and legal issues are bound to arise regarding fault of occupant vs. fault of software manufacturer. However, in the case of an accident between an autonomous vehicle and a non-autonomous vehicle, there are data collected in the autonomous vehicle regarding vehicle speed, trajectory, brake pressure & deceleration rate, among other details. These data can serve to strongly suggest fault of a non-autonomous vehicle operator if the autonomous car were operating safely within tolerances.

A much more nuanced and difficult ethical area to examine is how autonomous cars make moral decisions, in cases of accident avoidance and accident inevitability. In the New Yorker, Gary Marcus begins his discussion of autonomous car ethics by citing, as many have done when discussing robotic ethics, Isaac Asimov’s laws of robotic behavior:
 * A robot may not injure a human being or, through inaction, allow a human being to come to harm.
 * A robot must obey the orders given to it by human beings, except where such orders would conflict with the first law.
 * A robot must protect its own existence as long as such protection does not conflict with the first or second laws.

More heady discussions may address the issue that Asimov’s laws serve to relegate robots and autonomous machines to the status of human servant, and that this regime may be inconsistent with the possibility of ethically self-aware Artificial Intelligence. Autonomous cars in their current, primitive forms are nowhere near self-aware - rather, a mechanized, automated tool used to serve human transport needs. But even though they are not self-aware in the sense of “conscious” robots, the software driving the vehicles must still make ethical decisions. Should an autonomous vehicle attempt to minimize harm to its owner and/or occupants first and foremost, and minimize harm to other road users (e.g. a human-driven vehicle making an error and threatening to collide with the autonomous vehicle) second? Should an autonomous car be allowed to “illegally” swerve into the oncoming traffic lane to avoid obstacles and maintain traffic flow, or should it interrupt traffic flow to avoid breaking the law? Asimov’s laws are rather difficult to translate into sensible code that can be executed in real-time, and we do not have good answers to the above questions regarding harm minimization and the conduction of technically illegal maneuvers.

There are several considerations which mitigate the above moral dilemmas. If pedestrians suddenly dart out in front of a non-autonomous car, the attention level and reaction time of the driver, as well as the speed of the auto (elected by the driver), all determine whether the pedestrians are likely to be struck. Johansson (1971) reported a median of 0.9 seconds in the distribution of sudden-braking reaction times for human drivers. Reaction times for an autonomous car are faster than the median for humans, differential pressure ABS is more efficient at rapid stopping than manual systems, and an autonomous car operating appropriately is less likely to be traveling too quickly to stop (e.g. not speeding as humans do). Thus, an autonomous car is less likely to hit these pedestrians than a non-autonomous vehicle - the overall collision rate involving autonomous vehicles should be lower. Additionally, wider-spread adoption of autonomous car technology will serve to drastically reduce accident risk - in a system in which all but a few operators are communicating to their nearest-neighbors (near-field communication), the “off-grid” operators pose the greatest collision risk, all else held equal. Autonomous cars become safer the more of them there are, by way of removing unsafe human drivers from the road.

Finally, there is the issue of changing behavior of the driver, be it a human or computer. There is a tradeoff between having a human driver to blame for more frequent errors and collisions, and having a faceless, unfamiliar entity to blame for infrequent errors. The behavior of one of these actors is significantly easier to change than that of the other - essentially, pushing code patches to fleets of on-grid vehicles, versus humans unlearning bad driving behavior. A non-autonomous collision may be easier to prosecute, but each autonomous collision provides an engineering opportunity to reduce the likelihood of similar collisions in the future.

Legal Issues (Insurance and Liability)
A group of researchers from the University of California PATH Program documented that for automobiles there are three basic theories of tort liability; traditional negligence, no-fault liability, and strict liability. The majority of the states in the U.S follow negligence liability rule. The idea of liability for negligence is that a party should be held liable for harms caused by it unreasonably failing to prevent the risk. For example, suppose that Tom drives a car with defective low tire pressure despite. If he injures a pedestrian because of his low tire pressure, Tom will probably be found negligent and responsible for the accident. In addition, twelve states currently use the no-fault systems for automobile insurance. In these states, automobile crash victims are not permitted to sue other drivers in the tort system unless their injuries reach a certain degree of severity. Instead, victims recover their losses through their own insurance, which directly compensates them for their losses. Furthermore, the strict liability is the rule makes a person legally responsible for the damage and loss caused by her or his acts and omissions regardless of culpability.

How autonomous fits into the liability system above is very much undecided at this time. For example, National Highway Traffic Safety Administration (NHTSA) defines vehicle automation as having five levels, from level 0 which is completely no automation to level 5 which is completely automation. The debate is at which level the manufacturer should be responsible, and at which level the driver should be responsible.

Legal issues also involve insurance companies. For example, a driver might program the autonomous car to drive him/her to work in the morning, then tell the cars to go park with instructions to come back after work to pick the driver up. While the car is given instructions by a human, it is not being driven in the traditional sense. If a car were truly in passenger mode any violation would be a malfunction on the part of the vehicle. However, under the same circumstances, the only difference is that the driver is in the car. When the car senses danger and send safety alerts to the driver, the issue is whether the operator has a duty to override the vehicle. In this case, liability might be different depending on whether or not if the driver fails to override when obligated to, or acts to override in an inappropriate way. Either way, another challenge is that how to verify whether a person was operating a vehicle in a meaningful way, or whether the car was operating autonomously.

Policy Issues
As of the end of 2014, in the United States, legislation has been passed in four states regarding the legal use of autonomous cars: California, Nevada, Michigan, Florida and the District of Columbia. Meanwhile, International legislation has also been enacted to allow testing in Germany, Spain, and the Netherlands.


 * In June 2011, the Nevada Legislature passed a law to authorize the use of autonomous cars for testing purposes. The bill states: "Currently, the DMV is accepting applications for testing only. Autonomous vehicles are not available to the general public. "
 * In July 2012, the State of Florida legalized the test use of autonomous vehicles . The bill states: "autonomous technology may be operated on roads in this state by employees, contractors, or other persons designated by manufacturers of autonomous technology for the purpose of testing the technology. "
 * In September 2012, California passed the bill allowing the legalization of autonomous vehicles for testing purposes . The bill indicates that the State of California would authorize the operation of an autonomous vehicle for testing purpose only.
 * In December 2013, Michigan passed legislation allowing the testing of autonomous vehicles . The bill indicates that "Only a person who possesses a valid operator's or chauffeur's license may operate an automated vehicle in automatic mode on a highway or street of this State for testing purposes.

Even though the above places have legalized limited use of autonomous cars, there is still hesitance towards legalizing commercial use. Most of the laws enacted only allow for testing of the vehicle. In addition, there are certain rules for operators to use the autonomous vehicle. The Nevada Statues Chapter 482A states a human operator must sit in the driver's seat and ready to take immediate manual control of the vehicle in the event of a failure. In other States, the similar concern is also presented. For example, in Michigan, the bill states that a human operator shall be present in the automated vehicle to monitor the vehicle's performance. However, except in states where law mandates that operators of a vehicle must have a hand on the wheel at all times (and even then, you can just rest your hand on the wheel without applying force), autonomous cars are basically legal, because it hasn't been outlawed.

Future Directions
The future of autonomous cars appears to be a bright one, technologically - cars have been successfully testing in myriads of on-the-road situations for several years, and the sensory hardware and particularly controller software continues to be refined further. There are both clearly well-defined niche markets for autonomous cars, consisting primarily of users otherwise unable to travel in private vehicles such as the elderly or disabled, and broader markets consisting of everyone else for whom autonomous vehicular travel would be an optional choice. It is difficult to see autonomous travel not taking some sort of hold within the transportation market in North America, though it remains to be seen whether autonomous travel will exist primarily within, and bolster, the market of individual automobile ownership, or take stronger hold in car-sharing markets such as ZipCar and Car2Go. If autonomous cars become popular as a personal convenience item, saving their owners time during travel and, perhaps eventually, operating without their owners aboard on their behalf, then trips per capita are likely to increase, even further clogging roads. Sufficient legislation including very strict congestion pricing and prohibitively expensive parking costs in central business districts will be needed to discourage private ownership of autonomous vehicles. This is no different than the current legislative environment surrounding non-autonomous vehicles - congestion pricing and decreased parking subsidies are used to discourage excessive trips per capita. Autonomy is often equated with efficiency, but personal vehicle ownership is very inefficient - most cars spend 95% of their time parked. Autonomous vehicles have the opportunity to drastically shift the culture of car use and ownership.

Another consequence of the rise of autonomous car use is the potential death of public transit service in all but the most densely populated cores. Particularly in suburban communities where transit trips per capita and boardings per transit vehicle are lowest, service may be scaled back or cut altogether, if autonomous car-sharing becomes prominent. Until that point, transit service is not likely to disappear, as vehicular autonomy within the current framework of auto ownership is an add-on feature in the making - one still probably has to buy a car in the first place, which is economically prohibitive particularly for new vehicles. Autonomous cars will never replace heavy rail or high-frequency bus service in sufficiently dense corridors, because the urban form simply doesn't allow enough room for one autonomous vehicle per every 2 people to fit, let alone the likely situation of many single-occupancy vehicles. The potential for autonomous vehicles to impact mass transit services is limited to suburban disruption & replacement of low-demand, low-frequency, inefficient fixed-route service, and selective augmentation of more frequently used services by providing connections to system entry points.

Discussion Questions

 * 1) How do you feel about potentially relinquishing control of your car to a computer sometime in the future? How do you think your parents or grandparents would feel about it?
 * 2) How safe do you think autonomous cars need to be in order to be freely allowed on the roads? As safe as human drivers? Completely accident free?
 * 3) How do you think autonomous cars will impact the viability of public transportation?
 * 4) Why do you think most government transportation and transit agencies are not actively discussing and planning for a future with autonomous cars?  What are the possible repercussions of this?