Lentis/Autonomous Vehicles

Stanley is an autonomous vehicle created by Stanford University's Stanford Racing Team in cooperation with the Volkswagen Electronics Research Laboratory (ERL). It competed in, and won, the 2005 DARPA Grand Challenge, earning the Stanford Racing Team the 2 million dollar prize, the largest prize money in robotic history.

An autonomous vehicle is defined as a motor vehicle that uses artificial intelligence, sensors and global positioning system coordinates to drive itself without the active intervention of a human operator. These "sensors" can include, without limitation, cameras, lasers and radar. In the past several years, autonomous vehicle designers and advocates have made significant steps towards integration into the transit system and, eventually, universal autonomous transportation. This technology can revolutionize the way we approach transportation and, according to Jim McBride of Ford Research and Innovation, "there is no technology barrier from going where we are now to the autonomous car. There are affordability issues, but the big barrier to overcome is customer acceptance." As autonomous vehicle advocates move closer to large-scale implementation, social factors will prove to be their largest obstacles, and they must look to the past for lessons on how to proceed.

Case Study: Lessons From the First Cars

Automobiles have evolved from horseless carriages to vehicles resembling our modern cars.

The introduction of automobiles in the late 19th and early 20th century radically changed personal transportation. These "horseless carriages" were initially modeled after the carriages that people were already accustomed to. As people gradually overcame these mental models, automobiles evolved from the horse-and-buggy style into a design similar to modern vehicles. As automobiles evolved, so did their environment. The early days of automobiles included debates on whether cars should even be allowed on roads, since roads were for carriages and pedestrians. As the ubiquity of cars increased, entire infrastructures, both physical and legislative, had to adapt to the changes in transportation; Licenses and traffic control signals became necessary for safely operating vehicles. Much like the introduction of automobiles, the acceptance of autonomous cars embodies a drastic change in the mental models of transportation. People have been driving cars for over a century, and learning to surrender all control to electronics will likely be a transition rivaling the introduction the automobile itself.

Levels of Autonomy

Levels of Driving Autonomy

A misconception with consumers about automated vehicles is that they all will be able to drive with no human intervention. In actuality, there is a scale from Level 0 to 5 describing how much and how often a vehicle operates independently. Level 4 autonomy is the point where humans won't need to take over, and many of the benefits promised from self-driving will take shape at this level. Currently, most automated vehicles in use are Level 2, with progress toward 3 and 4 continuing at a rapid pace.

Current Automated Technology: 2020

A Google Self-Driving Car in a high speed control test.

Many advanced systems are in use today, with GM’s Super Cruise, Tesla Autopilot, BMW ADAS, and other’s all offering active driver assistance. While the performance and capabilities of each system vary, those in use today have automated systems for helping drivers. For example, automakers like Jeep and Toyota have released vehicles equipped with automated systems such as adaptive cruise control and lane departure warning systems.

State of The Art and Developments

Of the many automotive companies in the world, three notable ones looking to develop fully autonomous vehicles include General Motors, Tesla, Inc., and Waymo.

Waymo, known formerly as Google’s Self-Driving Car Project, has evaluated its technology's performance with millions of miles of testing and has started an autonomous ride hailing service called Waymo One in Phoenix, Arizona. Waymo extensively uses LiDar and high definition maps for depth perception. The company is actively working on decreasing the cost of its lidar sensors. Waymo also trains their cars in a simulator dubbed “Carcraft”, to speed up the learning process.

General Motors has a similar sensor suite to Waymo, using LiDar and high definition maps for scene composition. But because of this mapping based approach, GM and Waymo’s cars cannot operate unmapped roads. Super Cruise has an infrared camera in the cabin to tell when the driver is looking at the road, and will turn off automated functionality if the driver's gaze is averted.

Tesla has developed a feature called Tesla Autopilot that, according the NFTSA’s published levels of autonomy, operates with “partial automation;” the company is hopeful to develop software to support full autonomy in the future. Tesla’s system has no LiDar and only uses cameras and radar, with claims that they can accurately estimate depth from camera vision alone. Tesla’s intuition is that performance can only be achieved with variety and magnitude in training, and therefore the best learning experience is real world experience. With this goal in mind, Tesla will have estimated 5 billion miles driven on Autopilot as of January 2021.

Benefits

Safety

According to the Insurance Institute for Highway Safety, there were over 37,000 deaths from car crashes in the United States in 2017.  A study by the Rand Corporation published in 2017 found that adopting Highly Autonomous Vehicles (HAVs) sooner rather than later could save lives. While perfected autonomous vehicles are still in development, there is a demonstrable increase in safety with the systems today. As of Q3 2020, Tesla registered one accident for every 4.59 million miles driven in which drivers had Autopilot engaged. By comparison, NHTSA’s most recent data shows that in the United States there is an automobile crash every 479,000 miles.

A central problem autonomous cars have when performing in the real world is in the variety of situations encountered in everyday life. Any real-time system must process all the complex information in an environment and be able to act rationally in new situations. While the underlying neural nets in self-driving cars excel at recognizing objects previously trained on, they struggle to transfer this learning toward new objects and situations. Safety is directly proportional with accurate object perception, and Tesla hopes to iterate fast enough to overcome this deficiency. As they continuously run into novel situations and objects, and by seeing where their system fails, they can retrain the underlying HydraNets and prevent entire categories of accidents stemming from mislabeled objects in a scene.

Fuel Efficiency and Traffic

Widespread adoption of autonomous vehicles could improve fuel efficiency, decreasing fuel consumption. Adaptive cruise control (ACC), considered level 1 autonomy, is already making a difference. A study by the Institute of Electrical and Electronics Engineers (IEEE) found that vehicles equipped with Adaptive Cruise Control consumed between 5 and 7 percent less fuel than vehicles not equipped with ACC. As higher levels of autonomy are adopted, fuel efficiency may improve as well.

Optimal fuel efficiency can be attained in the more distant future. If autonomous vehicles are proven to be as safe as some experts predict, cars could be completely redesigned. Nady Boules, the director of GM’s Electrical and Controls Integration Lab predicts, “You could remove the weight dedicated to crash protection, using very light materials for the skin instead of metals.” By removing unnecessary weight, vehicles will become much more fuel efficient. This could also eliminate the cost of safety technologies such as airbags.

Vehicle-to-vehicle communication may improve fuel efficiency and traffic congestion, and safety. Results from a simulation of Autonomous Car Following Control systems and Cooperative Car Following Control (C-CFC) systems published by the IEEE found that both control algorithms improve traffic flow in traffic jams. Cars in the simulation using the C-CFC algorithm, a type of vehicle-to-vehicle communication, were more fuel efficient than cars not using the C-CFC algorithm.

Other Benefits

A 2018 study at the University of Toronto found that autonomous vehicles could reduce the space needed in parking lots by 62 percent. Fully autonomous vehicles drop passengers off at their destination. With no occupants to exit vehicles in parking garages, autonomous vehicles could park much closer together. Vehicle-to-vehicle communication would allow for a grid-like layout of cars. When a car exits the garage, the surrounding cars rearrange. These two factors contribute to a decrease in space needed for public parking.

Barriers to Adoption

Legal

Legal issues are one of the most significant obstacles to the large-scale production and use of autonomous vehicles. The introduction of such vehicles has impacted and will further impact most aspects of society, so policymakers must move towards a comprehensive set of applicable legislation.

Governance

Nevada (June 6, 2011) was the first state to effectively allow the operation of autonomous vehicles by authorizing the Nevada Department of Transportation to set forth rules and regulations governing these vehicles. Twenty-nine additional states have enacted autonomous vehicle-related legislation since. Governors of ten states have issued executive orders to help advance legal progress.

On the federal level, legislators seek to facilitate the creation of a framework for the testing and operation of autonomous vehicles. Since Nevada first enacted legislation in 2011, little federal progress has ensued, as legal concerns have become controversial and multifaceted. 2017 was a recent turning point — two bipartisan bills were introduced in Congress. The SELF DRIVE Act passed unanimously through committee and passed a verbal vote on the United States House of Representatives floor, and the AV START Act passed unanimously through committee but was never brought to the United States Senate floor. Differences between the two bills were not reconciled because of concern over safety details, and neither bill was passed. Advocates for Highway and Auto Safety, Consumer Reports, the Consumer Federation of America, and the Center for Auto Safety shared safety concerns, asserting that “rushing a bill to usher in mass deployment of risky, unproven systems . . . using dubious claims about safety is misleading and disingenuous at best, and reckless and deadly at worst.” Both pieces of legislation were fairly similar. They would have established “the federal role in ensuring the safety of highly automated vehicles by encouraging the testing and deployment of such vehicles,” while preventing states from enacting related laws unless in strict accordance with federal laws. They also would have directed the National Highway Traffic Safety Administration (NHTSA) to publish safety standards for autonomous vehicles. Nevertheless, in 2018, the Department of Transportation released new autonomous vehicle guidelines, called Automated Vehicles 3.0. The voluntary guidance advanced safety principles and implementation strategies, reduced policy uncertainty, and advised on working with the Department of Transportation in the future.

In August 2019, the United_States House Committee on Energy and Commerce and the United States Senate Committee on Commerce, Science, and Transportation solicited input from industry, nonprofits, and disability advocates to help craft a new bipartisan bill. In October 2019, the Senate committee began to circulate the first few sections of the first draft of a prospective bill, the language of which is comparable to that of the SELF DRIVE Act. The timeline of the new bill is still unclear.

Liability

A major challenge to the widespread adoption of autonomous vehicles is liability uncertainty. This is a particularly important challenge because it will determine the landscape of the commercialization of autonomous vehicles. It remains uncertain who will be liable when there is a car accident or other violation for which there is a party at fault, as “no clear legal framework exists that outlines how liability is apportioned between third parties responsible for designing AV systems – the manufacturer, supplier, software provider or the software operator.” Some question if the owner or operator of the vehicle, if there is one, should be compensated for damages to the vehicle. It is also unclear if there will be standardized crash algorithms to allow for a more generally applicable approach to liability assessment.

Data Privacy

Data privacy is intertwined with liability. Insurance companies weigh how much and what type of data will be accessible or sold for assigning liability more accurately. Autonomous vehicles rely on a myriad of sensors sampling continuously to ensure safe operation. Location data could be exposed to external networks, allowing others to track the vehicle. Autonomous vehicle users could face highly targeted advertisements from their autonomous vehicle data, which could give companies private information.

Current policy addresses some privacy concerns related to vehicle data. Security and Privacy in Your Car Study Act of 2017 required the National Highway Traffic Safety Administration to protect against unauthorized access to driving data, including for marketing and advertising purposes. The bill applies to autonomous vehicles, but does not specify specific additional language for those vehicles.

Ethics

Depiction of the Trolley Problem

In March of 2018, an Uber operating under a self driving system killed a pedestrian. The pedestrian, Elaine Hertzberg, was walking across the street with her bike. Though the car detected an object in its path, the algorithm did not apply the breaks. It is expected that the back-up human driver will apply the breaks in this situation. This outcome is concerning to many and begs the question: Can machines act as moral agents? As the prevalence of autonomous vehicles increase, engineers will have to face questions similar to the Trolley_problem, which contemplates a trolley moving down a track. The trolley can continue in its course, hitting 5 people tied to the track, or switch paths hitting 1 person. Ethics scholars are discussing these hard questions.

Cyber Security

Cyber Security poses a threat to autonomous vehicles. Autonomous vehicles are vulnerable to hackers, making the many networks connected to these vehicles vulnerable as well. The more autonomous vehicles become, the more threats are posed.

Public Perceptions

A 2019 study found that public perceptions impact the ease of adoption for autonomous vehicles. Familiarity plays a role in these perceptions. Public opinion of autonomous vehicles improves with increased interaction with the vehicles. The study recommends implementing policy that allows increased interaction with autonomous vehicles.

Economics

Impacts on Employment

The transportation industry accounts for a large percentage of the workforce in the United States. With 207k taxi drivers, 2.4 million truck drivers, 160k uber drivers, and 570k bus drivers, there are over 3.4 million driving jobs. This number doesn't account for the truck stops, support workers, etc. that support truckers/drivers, and when those are factored in there are a whopping 7 million endangered jobs as a result of autonomous vehicles. With a current workforce size of 164.6 million, this accounts for 4.25% of the US workforce. Despite this large makeup, it is only estimated that unemployment rates will decrease 0.06-0.13% in 2045-2055, the predicted time period for when these autonomous vehicles will have the greatest effect.

Potential Effects

The relatively small effect on employment is a result of the benefits of autonomous vehicles and a predicted job shift in the United States. With the introduction of autonomous vehicles, access to transportation and the ease associated with it will drastically increase. The number of individuals who were unemployed due to a lack of transportation will drastically decrease, positively effecting unemployment. Additionally, the US spends $840 billion annually responding to and dealing with automobile accidents. As a result of the drastically smaller accident/crash rate with new self-driving technologies, it is predicted that the US will benefit $800 billion annually by 2050. These savings can take the form of job-conducive government investments or economic stimulation in one form or another. Similar to the job shift from agriculture to industry a century ago, a job shift from driving to other industries is expected to occur. As said by Amitai Bin-Nun, vice president of autonomous vehicles and mobility innovation at Securing America’s Future Energy, “It’s not because Americans stopped eating.” There was the same demand for food before the introduction of job-stealing agricultural technologies as after, yet there was not a huge hit to unemployment rates. The introduction of driving technologies echoes this same pattern, and thus predicts a similar result.

Responses

Policies to address the advent of autonomous vehicles and their effect on American jobs became a topic of debate with presidential candidate Andrew Yang. While previous legislation had been passed about the development of self-driving technologies, Yang was the first politician to make his responses to correlating unemployment issues a central policy on the debate stage and throughout his run for the presidency in 2020. Famously quoting that "truck driving is the most common job in 29 states," Yang attempted to appeal to the working class and propose solutions that would ease the transition to self-driving technologies. Notably, Yang's policy had two major points:

• A Trucking Czar would be appointed to oversee the successful transition of truck drivers to other professions
• It is estimated that automated freight would result in an annual savings of $168 billion for corporations. These profits would be taxed and reallocated as severance packages for truckers who've lost their jobs as a result of these new technologies.

Future Implications

Traffic Flow

Bumper-to-bumper traffic and rush hours are all too common situations that city commuters face on a daily basis. A single driver could have a monumental effect on traffic flow if he or she breaks hard, causing every car in her rear view mirror to follow in a domino effect and leading to congestion. Scientists and mathematicians have been researching traffic flow since the 1930's, using partial differential equations to model driver behavior in congested environments. With the computing power of Amazon Web Services (AWS), today's researchers are able to use stronger machine learning models to simulate traffic flow. Alexandre Bayen, a professor at the University of California Berkeley, has been using Deep Reinforcement Learning - a subset of machine learning - to simulate car behavior on a closed circuit. Bayen, along with other scientists in the field analyze the effects on moderating the speeds of every other car on the road and the impact it has on congestion.

Rideshare and Taxi Services

The intersection of Ridesharing and Autonomous Vehicles has become more prominent as companies like Uber and Lyft enter the market for self-driving technologies. In June of 2020, Lyft announced its development of Level 5 autonomous vehicles that would allow for safer and more accessible form of transportation. Using data from over one billion rides, Lyft's software is able to use visual localization technology in conjunction with real-life driving scenarios to simulate more sophisticated and realistic driving behavoir. Riders could summon autonomous vehicles at any time of day; the availability of a driverless ride hailing service would not have to depend on drivers in the area being awake and alert. Unsafe interactions between a rider and a driver would be impossible without a driver in the vehicle.

With the ridesharing industry expanding all across the world, the Taxi industry has been taking a hit. In contrast to ridesharing services, the taxi service business model charges based on moving and stop-and-go traffic, making it hard for passengers to predict prices and drop off times. Also, the convenience of hailing a ride from your smartphone and the option of choosing the right car type for your needs has allowed rideshare companies to overtake traditional taxi services. In July of 2017, Uber surpassed yellow cabs in average daily ridership in New York City with an average of 289,000 rides per day, whereas yellow cabs only managed 277,000. . Government restrictions have been put in place to protect the jobs and livelihoods of taxi cab drivers, but with the rise of autonomous vehicles, the future of traditional ride hailing services may face trouble in the near future.

Trust of Technology

The development of autonomous vehicles have escalated at a rate faster than regular people are understanding how the technology work. This gap has led many people to lack the trust of relying on computerized systems on their transportation needs. With full autonomy, the driving decisions that were once made mostly by humans would be made by a combination of their vehicles’ hardware and software. Surveys from AAA, Kelly Blue Book, and Partners for Automated Vehicle Education (PAVE) have yielded the following statistics:

• Nearly 3 in 4 Americans say autonomous vehicle technology “is not ready for primetime.”
• About 48 percent said they would never get in a taxi or ride-sharing vehicle that was self-driving.
• Another 20 percent think autonomous vehicles will never be safe.
• Only 34 percent think the advantages of AVs will outweigh the disadvantages.

Through computer simulated trials and models, many researchers argue that human error cause the vast majority of car crashes and these detrimental situations can be mitigated with autonomous vehicles. However, gaining the trust of the people to use this technology is an issue that must be addressed further. The effectiveness in high levels of autonomous driving relies on the support of the people using the software as these technologies work better synchronously rather than in isolated instances.