What are Autonomous Vehicles?
According to the Code of the District Columbia, the legal definition of an autonomous vehicle is “any vehicle capable of navigating District roadways and interpreting traffic-control devices without a driver actively operating any of the vehicle’s control systems”. This definition excludes vehicles with partial automation, such as those with emergency brakes, crash avoidance, blind-spot monitoring, and parking assistance according to BMW.
5 Levels of Driving an Autonomous Vehicle
There are 5 different levels of autonomous driving with only level 5 being considered fully autonomous (2019).
- Level 1: Driver Assistance: Driver assistance systems support the driver, but do not take control. Example: Active Cruise Control.
- Level 2: Partly Automated Driving: systems can also take control, but the driver remains responsible for operating the vehicle. Example: Your vehicle can break automatically, accelerate, and take over steering unlike Level 1.
- Level 3: Highly Automated Driving: in certain situations, the driver can disengage from the driving for extended periods of time. Meaning, the driver will be able to give complete control of the vehicle to the vehicle’s operating systems.
- Level 4: Fully Automated Driving: the vehicle drives independently most of the time. The driver must remain capable of driving but can, for example, take a nap.
- Level 5: Full automation: the vehicle assumes all driving functions, the people in the vehicle are only passengers.
History of Autonomous Vehicles
Autonomous Vehicles (AV) have long been associated with radical thoughts and our favorite Sci-Fi movies. Now, more than ever, they seem to be a very viable option for the future and are expected to play a crucial role in the development and implementation of vehicles moving forward. James M. Anderson, among others, from Rand Corporation classify the history of AV into three different phases.
Three Phases of Autonomous Vehicles
- Phase 1: Foundational Research: From 1980 to 2003, university research centers, in partnership with automotive companies and transportation agencies, undertook studies dealing with autonomous transportation. From this first wave of studies, two essential technological concepts were created. The first was that AV would depend on automated highway systems and highway infrastructure to guide them. The second was to develop a semi AV development and a complete AV development that will depended little, if at all, on highway infrastructure.
- Phase 2: Grand Challenges: From 2003 to 2007, the United States Defense Advanced Research Projects Agency (DARPA) held three “Grand Challenges” which would ultimately accelerate advancements in AV technology, and reignite the general public’s interest and imagination in AV. The Grand Challenges tasked research teams from across the globe with developing vehicles that were fully autonomous and then had them complete a 150-mile race for $1 million- and $2 million-dollar prizes, respectively. In the 2004 Grand Challenge, not a single vehicle completed the challenge with the best competitor completing less than eight total miles of the course. However, by 2007, a 60-mile urban course dubbed the “Urban Challenge” saw six teams complete a course while driving side by side with other autonomous vehicles, real human drivers, and the task of obeying traffic and safety rules. These Grand Challenges proved crucial in spearheading advancements in sensor systems, computing algorithms, and AV in general.
- Phase 3: Commercial Development: The DARPA challenges solidified the future of AV and presented the opportunity for partnerships between the education sector and automobile manufacturers. These partnerships would flourish and ultimately create the GM and Carnegie Mellon partnership and the Volkswagen and Stanford University partnership. Google’s driverless car initiative has brought AV research from university laboratories to commercial research and has developed a tested fleet of cars and initiated campaigns to demonstrate the applications of technology. Since then, Google alone has logged more than 500,000 miles of autonomous driving on public roads. A huge step forward in the implementation of AV.
The Potential Benefits of Autonomous Vehicles
In the United States, alone, roughly 37,000 people are killed and more than two million are injured in crashes every year. U.S Motor Vehicle crashes as a whole can pose social and economic costs of $800 billion in a single year. And, more than 90% of crashes are caused by human errors according to Nidhi Kalra of Rand Corporation. Among these errors are speeding, drunk driving, distracted driving, and fatigue. As General Motors Chairman Bob Lutz said, “The autonomous car doesn’t drink, doesn’t do drugs, doesn’t text while driving, doesn’t get road rage, doesn’t race other autonomous cars, and never gets tired”. And, while technology isn’t perfect and the autonomous car isn’t completely reliable, researchers estimate that autonomous vehicles can reduce accidents by 90%, saving over 30,000 American lives, and avoiding millions of injuries on the road. In a study conducted by the University of Michigan Transportation Research Institute, Researcher Brandon Schoettle observed that autonomous cars are a safer alternative than the traditional conventional car. Albeit, the number of miles driven by autonomous cars was severely less than the number of miles driven by conventional cars. It was found, that the overall severity of injuries suffered by those in autonomous vehicle crashes was much less than those in conventional vehicles. Furthermore, in the study, all the accidents that occurred between autonomous vehicles and conventional vehicles were caused by the conventional vehicle 100% of the time. Subaru has also recently released data for the Japanese market showing that vehicles fitted with adaptive cruise control, automatic emergency braking, and lane departure warning systems (features of AV) were 60% less likely to be involved in an accident than with those without the technology.
Benefits for the Special Needs & Elderly
Driving a car is not an option for many people with visual impairments, ambulatory disabilities, cognitive disabilities, and other disabilities. Special retrofits may allow people with disabilities to drive but with low estimates of $20,000 and higher estimates reaching $80,000, most people with disabilities cannot afford to get their cars retrofitted. Furthermore, public transportation is not always a reliable measure to help drive people with disabilities. Even though most buses are built to be accessible to individuals with disabilities, in many areas bus stops are not suitable for wheelchair access. Henry Claypool from the SAFE and the Ruderman Family Foundation updated the National Academy of Science (NAS) model to estimate that approximately 4.3 million individuals with disabilities face a significant transportation barrier while attempting to travel to their medical appointments. Thus, the implementation of AV would be crucial. Others also argue that AV would provide enough independence of mobility for non-drivers, disabled or not and that this would directly benefit these travelers, reducing chauffeuring burdens on their family and friends, and improving their access to education and employment opportunities.
Decreased Opportunity Cost & Increased Efficiency
When vehicles become fully autonomous, they have the ability to become a Connected Autonomous Vehicle (CAV). CAVs can improve safety and increase efficiency through active, high-frequency transmission of vehicle position, speed, direction, and acceleration rates. Drivers with CAVs might choose their route and travel choices more effectively than drivers without a connected environment. According to Joo Young Lee from the University of Texas at Austin, CAVs would be able to find the shortest path of their trip by knowing the traffic conditions through communications. Thus, allowing them to optimize their paths to satisfy certain objectives, such as finding a path with fewer stops. When route choice algorithms are combined with automated vehicles’ computing power, more effective trips can be made. Route choices of CAVs will affect traffic flow at both the individual level and system level. At the system level, the congestion level of cars can be lowered by dispersing each driver’s path to avoid congestion. CAVs can improve passenger efficiency by allowing them to focus on other tasks such as work while they’re driven by AV and reduces the net time in car by decreasing congestion or choosing faster alternative routes that conventional vehicles would not normally travel.
The Potential Harms of Autonomous Vehicles
Increase Fuel Emissions
Autonomous vehicle technology improves accessibility. This accessibility would greatly benefit people who are elderly, disabled, and unable to drive. However, this new-found accessibility would lead to an increase in trips, in turn, increasing energy use and emissions. Some predict that energy consumption would increase by 10-14% from these newly induced trips. Additionally, since traveling will be made so accessible, the vehicle miles traveled (VMT) would increase exponentially and become greater than those of conventional cars. Accessible travel through cars would decrease airline revenue by 53% and travel to increased distances using AV would increase by 9.6 %. The increased travel distance of AV and CAVs would increase energy consumption in the ground transportation sector and it’s predicted that there will be an increased rate of energy consumption between 6-18%.
Like any technology, autonomous vehicles are prone to hacking. The threats to AV can come through any of the systems that connect to the vehicles sensors, communications applications, processors, and control systems, as well as external inputs from other vehicles, roadways, mapping, and GPS data systems. Andre Weimerskirsh, from the University of Michigan, warns that the control systems of each vehicle for speed, steering, and braking are vulnerable to attacks. These flaws can leave people at risk of hackers who are trying to shut down and take over a vehicle, criminals who could try to ransom the vehicle and the passengers inside, and others trying to burglarize the vehicle. In 2015, under a controlled test, hackers, Dr. Charlie Miller, and Chris Valasek hacked and took over the control system of an internet-connected Jeep Cherokee doing 70mph outside of St. Louis. This hacking helped prove the vulnerabilities of onboard control systems, with the hackers able to control the cars radio, ventilation, braking, and transmission, ultimately stalling the vehicle on the highway. A year later, the same hackers demonstrated the ability to control the same car’s steering and parking brake system, totally bypassing the existing security measures on the car. Additionally, researchers from the University of California, University of San Diego and the University of Washington showed that they could inject messages into the CAN bus of a vehicle, allowing them to make physical changes to the car, such as controlling the display of the speedometer, killing the engine, and affecting the brakes. A follow-up study conducted by the same researchers also showed that these vehicles were vulnerable to being attacked from across the country and not just locally according to Dr. Charlie Miller. Even more frightening, is that deep neural networks (DNNs) are vulnerable to being tricked and hacked by slight modifications to input. DNN’s are considered state-of-the-art and are increasingly being used as part of control pipelines in autonomous cars. However, researchers from the University of Michigan, Washington, and California-Berkeley have shown that these high-tech pipelines can be tricked by placing stickers on stop signs or other street signs. These crafted modifications to the visual input of DNN’s can cause the systems they control to misbehave in unexpected and potentially dangerous ways according to research led by Kevin Eykholt. These malicious modifications to street signs can prove life threatening, as AV can misread stop signs as speed limit signals, speed limit signals as stop signs, and make other errors. Furthermore, researchers from Princeton evaluated the threat of these attacks and were able to demonstrate that “adversarial examples created from either arbitrary point in the image space or traffic signs, can deceive the traffic sign recognition system with high confidence.” In fact, the study showed that the attacker can achieve high attack success rates in excess of 90% in the real-world setting according to data from Chawin Sitawarin of Cornell University.
Legality and Ethics of Self-Driving Cars
The reason that AV’s are so lucrative to society is that they have the potential to eliminate crashes and save thousands of lives. Human error is responsible for 90% of crashes, and AV’s look to remove the human aspect and instead have cars built on math and science. However, one important thing that AV’s lack is a sense of ethics. Ethics refers to the norms of people’s internal values and external behavior. To behave like a human being, it is necessary for a module to interpret ethics and morality into indicators by machine learning says Sixian Li, a professor from Shandong University of Technology and Science. Without a proper sense of ethics, how would autonomous vehicles react in a situation where there is an unavoidable accident? The “trolley problem”, is a typical case of decision making under emergency situations and has become one of the most recognizable scientific examples of ethical and legal situations. The trolley problem has individuals choose whether to change the direction of a trolley and kill one person or stay on the path and kill five. The problem is designed to present a moral dilemma, that many humans have trouble answering.
So how exactly would autonomous vehicles act? Well, the reality is that AV’s cannot make a decision like this on its own and are instead forced to rely on to how it was originally programmed. This then leads to the question as to how accident algorithms should be programmed. Sven Nyholm asks, “Should it always minimize the number of deaths or be programmed to save their own passengers at all cost?” And, if accidents do happen, then who is to blame? There is no clear-cut answer, but liability can either be broken down into tort liability concerning drivers and insurers or manufacturing liability concerning manufacturers and tech companies claims Richard Ni. Ultimately, it would be a slippery slope regarding user or manufacturer error when determining accident responsibility between AV in the future.
On average, there are over 5.7 million vehicle crashes per year. Approximately, 22% of these crashes — nearly 1.3 million — are weather-related. Weather poses a serious risk to AV’s because it makes the vehicle’s crucial functions unable to operate at full efficiency. Adverse weather conditions such as fog, heavy rain, snow, and wind can severely limit the functionality of sensors and cameras. These conditions can result in faulty readings and impact sensor accuracy resulting in car accidents that humans would’ve been able to avoid. The World Economic Forum and Boston Consulting Group concluded that adverse weather like rain or snow “can alter vehicle traction and change how the vehicle’s cameras and sensors perceive the street” writes Kyle Stock of Bloomberg.
Autonomous vehicles are still in their early to mid-age of development and significant progress must be made before mass production and distribution of AVs is realized. Millions of more miles need to be driven and thousands of more scenarios need to be carried out to truly attempt to prepare AV for constant recreational use. Even then, the question “How safe is safe enough?” may still not be answered. Despite the uncertainty of the future in regard to AV, we can see a multitude of benefits in theory. AV would decrease the number of accidents and deaths by eliminating human error, assisting those who can’t drive and greatly increasing accessibility, and improving passenger efficiency while reducing the opportunity cost of driving. However, with every benefit comes a potential harm. AV are projected to increase fuel emissions that damage our environment, face the risk of cyber-attack, have questions regarding their ethics and legal circumstances, and are not fully operational in inclement weather, which is perhaps the biggest concern. Many questions must be answered before even considering the mass production and implementation into society of these autonomous vehicles.
Bmw. “Autonomous Driving – Five Steps to the Self-Driving Car.” Autonomous Driving – 5
Steps to the Self-Driving Car, BMW, 5 July 2019, www.bmw.com/en/automotive-
Anderson, J.M., Kalra, N., Stanley, K.D., Sorensen, P., Samaras, C., Oluwatola, O.A., 2016.
Autonomous Vehicle Technology: A Guide for Policymakers. RAND Corporation, Santa
Monica, Calif. (RR-443-2-RC).
Kalra, Nidhi, and Susan M. Paddock. How Many Miles of Driving Would It Take to Demonstrate
Autonomous Vehicle Reliability? RAND Corporation, 2016.
Torts of the Future: Autonomous Vehicles. US Chamber Institute for Legal Reform, May 2018.
Schoettle, Brandon, and Michael Sivak. A Preliminary Analysis of Real-World Crashes Involving
Self-Driving Vehicles. The University of Michigan Sustainable Worldwide Transportation, 30 Oct. 2015.
“Study: EyeSight-Equipped Subaru Vehicles Reduce Accidents Significantly.” Torque News, 26
Claypool, Henry, et al. SELF-DRIVING CARS: THE IMPACT ON PEOPLE WITH DISABILITIES.
Ruderman Family Foundation, Jan. 2017.
Litman, Todd. Autonomous Vehicle Implementation Predictions Implications for Transport
Planning. Victoria Transport Policy Institute, 18 Mar. 2019.
Lee, Joo Young, and Kara M. Kockelman. ENERGY IMPLICATIONS OF SELF-DRIVING VEHICLES.
University of Texas at Austin, 2019.
LaMondia, J., Fagnant, D., Qu, H., Barrett, J. and Kockelman, K. M. 2016. Long-Distance Travel Mode
Shifts Due to Automated Vehicles: A Statewide Mode-Shift Simulation Experiment and Travel Survey Analysis. Transportation Research Record. No. 2566: pp. 1–10.
Weimerskirch, Andre, and Derrick Dominic. Assessing Risk: Identifying and Analyzing
Cybersecurity Threats to Automated Vehicles. University of Michigan, Jan. 2018.
Miller, Charlie, and Chris Valasek. Remote Exploitation of an Unaltered Passenger Vehicle. 10
Eykholt, Kevin, et al. Robust Physical-World Attacks on Deep Learning Visual Classification. 10 Apr.
Sitawarin, Chawin, et al. DARTS: Deceiving Autonomous Cars with Toxic Signs. 31 May 2018.
Li, Sixian, et al. Ethical and Legal Dilemma of Autonomous Vehicles: Study on Driving
Decision-Making Model under the Emergency Situations of Red-Light Running Behaviors. Department of Transportation Engineering, College of Transportation, Shandong University of Science and Technology, 22 Oct. 2018.
Nyholm, Sven, and Jilles Smids. The Ethics of Accident-Algorithms for Self-Driving Cars: an Applied
Trolley Problem. 28 July 2016.
Stock, Kyle. “Self-Driving Cars Can Handle Neither Rain nor Sleet nor Snow.” Bloomberg.com,
Bloomberg, 17 Sept. 2018, 6:00 am, www.bloomberg.com/news/articles/2018-09-17/self-driving-cars-still-can-t-handle-bad-weather