Abstract
Autonomous vehicles are automobiles that require no human intervention, but they still require user awareness. They are not yet fully developed, and their goal is to be completely autonomous. Despite having stricter safety standards than traditional vehicles, they are not yet safe enough.
So, what is “safe enough?” Before going into depth about changes that should be made to enhance safety culture, a detailed explanation of the current information on autonomous vehicles will set up the improvements to follow: (1) the definition of autonomous vehicles (background), (2) the problem, (3) current state of autonomous vehicles and (4) future actions to take. Through extensive research, autonomous vehicles have been found to lack clear expectations from the user and efficacy from the existing safety regulations, which leads the research of this paper towards the next steps for an appropriate safety culture.
Introduction
“Even when designed with the best intentions, automated driving technologies affect human performance in ways that could negate the potential benefits.” (Human Factors and Ergonomics Society, 2022)
With a projected $64.88 Billion value in 2026, the autonomous vehicles (AVs) market is widely recognized as “the future of automobiles” for its potential to optimize traffic flow and road safety (Rajasekhar). Automated systems, specifically in cars, have become a hot commodity. Engineers and investors promise autonomous vehicles (AVs) can improve air quality, reduce traffic congestion and accidents, and enhance human mental and physical health. But at what cost? According to the National Transportation Safety Board (NTSB), car crashes kill 30,000 people annually. As Dr. Missy Cummings points out, “the technology is being abused by humans” (Metz). In February of 2022, the Human Factors and Ergonomics Society identified the loss of driver engagement and low situational awareness, poor understanding of and overreliance on automated systems, and loss of human manual skills as problems that arise with autonomous vehicles that must be addressed. As automation continues to become more advanced with increasing levels, we must not consider these AV systems as independent replacements for humans, but rather as part of a human-AI team with complex interdependencies. Philip Koopman further emphasizes that “it is [even] unknown how long it will take to be safer than human drivers–if ever” (Koopman).
Levels of Automation
Regarding autonomous vehicles there are six different levels of automation that place autonomous vehicles into a specific category based on the features and its capabilities. The word “autonomous” is often misleading in the subject of vehicles and misperceived as meaning vehicles that can perform without any human intervention. However, according to the Cambridge Dictionary, the term autonomous defines a “machine or system able to operate without being controlled directly by humans” (Cambridge Dictionary). Individuals read this definition with emphasis on the word “without” rather than on the word “direct”; to most consumers, “without” means that the autonomous vehicle is fully capable of driving on its own as a car would with a human behind the wheel, but the word “direct” implies that user attention is still necessary for its operation.
Current State of Autonomous Vehicles
Technological Maturity of Autonomous Vehicles
In the 21st century, more technology is seen as better. Technological determinism governs Western society. While innovations can revolutionize society for the better, sometimes solutions are chosen not because they are better but rather because they are high-tech. It’s easy to become overenthusiastic about new engineering and ignore its limitations. As Professor Najmedin Meshkati puts it, “just because something can be done, sometimes it shouldn’t be” (The Tesla Experiment). Technology is neither inherently bad nor good. If autonomous vehicles cause just as many accidents as traditional ones, there is no benefit to implementing them. In 2021, Elon Musk, CEO of Tesla, claimed he was “highly confident the car will be able to drive itself,” which of course was an unmet goal (Metz). As Dr. Missy Cummings points out, “The reality that such systems may not actually be as capable as envisioned is slowly creeping into the national consciousness” (Cummings).
While several automotive companies claim to be working on “fully autonomous vehicles,” we must consider the actual technological maturity of these systems and how they can be implemented. These companies are promoting “inflated and unrealistic expectations” (Cummings). Machine-learning effectively performs bottom-up reasoning after being trained with millions of data points, but it lacks top-down reasoning which is a “critical element in uncertain safety-critical systems” (Cummings). Deep learning algorithms that AVs depend on suffer from brittleness, meaning that they cannot generalize or adapt conditions outside of a narrow set of assumptions. For example, if a stop sign is covered by a tree branch or lane markings are covered in snow, automation cannot adapt. With deadly potential consequences, this limitation is concerning.
The Human Readiness Level (HRL) scale is “mechanism to evaluate, track, and communicate the readiness of a technology or system for safe and effective human use that can be applied in the context of existing systems engineering and human systems integration (HSI) processes” (Human Factors and Ergonomics Society). Technological maturity is not solely about how many functions the automation can perform – rather, the HRL scale measures “the level of maturity of a technology with respect to its readiness for human use” (Human Factors and Ergonomics Society). Autonomous technologies are not 1-to-1 replacements for humans. They operate in a larger network including the human operator, other human drivers, and other automation. Therefore, we must consider the importance of these interactions.
AI-Human Interactions
While automation is often seen as a “single element,” it is useful to consider AVs in the context of a larger network. As Josh Lee writes, “As automation becomes more capable, the need to craft designs compatible with people becomes more important” (Lee). Though the evidence related to autonomous vehicle’s impact on human behavior is limited, aviation systems can provide interesting parallels. When unexpected events occur, autopilot can actually create new risks. In 2009, Air France Flight 447 killed 228 people after autopilot disconnected due to bad weather, and the pilots did not know how to respond. Four years later, Asiana Flight 214 crashed because the pilot didn’t understand the automation system well enough (Gottlieb). According to Larry Sanberg, these examples show how pilots can become overly reliant on cockpit automation, and consequently lose their skills and situational awareness. Some pilots with excellent training and no infractions become complacent, and they ignore precautionary measures like checklists. Most of the time, this doesn’t cause any harm. If anything does go awry, though, the result could be deadly. Automobiles introduce even higher risks: airplanes have thousands of feet of clearance and minutes to make decisions. By contrast, the National Highway Traffic Safety Administration (NHTSA) claims drivers have only 17 seconds, or about ¼ mile, to take control of a car (Gottlieb). Whereas airline pilots are extensively trained, drivers do not receive any official automation training. It’s unrealistic to assume all owners will read detailed manuals. Judging aviation safety and considering the compounding factors on the ground, safety hazards are substantial.
Despite the novelty of this technology, there is already some research demonstrating the impact of autonomous systems on driver behavior. Google employees were filmed early on in their Waymo AV project and became distracted. MIT research published in the Accident Analysis & Prevention journal presents a model for people’s behavior while using Tesla’s Autopilot. Using on-road and off-road glance data, they found that visual behavior patterns changed before and after Autopilot was disengaged (Morando). When using Autopilot, people were less attentive, which is well-known to have adverse safety effects. A study conducted in boredom research determined that boredom is tied to vigilance, attention, and performance. The results of the study suggest that increased automation in the workplace will influence these factors, and we must be able to measure boredom based on psychological signals. Inattention in crashes is estimated to cause ¼ of all driving accidents (Basu). Engineers must actively prevent driver complacency and distraction to avoid accidents.
Existing Regulations
NHTSA’s Automated Vehicles Policy
The NHTSA is known as a top safety regulator and this agency of the United States federal government has the mission to “Save lives, prevent injuries, reduce vehicle-related crashes.” The Federal Automated Vehicles Policy includes vehicle performance guidance for automated vehicles, the model state policy, current regulatory tools, and modern regulatory tools. Vehicle performance guidance for automated vehicles serves as a set of 15 guidelines to establish “safe” design and implementation for manufacturers.
Model Sate Policy
When considering the Model State Policy there is a clear distinction between the responsibilities expected from the government and the responsibilities expected from the state. On the U.S. Department of Transportation website, there is a fact sheet labeled AV Policy Section II and a set of responsibilities is described for each. The Federal responsibilities consist of establishing and ensuring the adherence to safety standards, examining, and improving any safety recalls, and “educating the public about motor vehicle safety issues,” etc. State responsibilities include “licensing (human) drivers”, establishing traffic laws, performing safety examinations, etc. (U.S. Department of Transportation).
Analysis of Policies
Through analysis, one federal and one state responsibility listed above are worthy of focus and will be highlighted throughout the rest of this paper: educating the public and licensing human drivers. If autonomous vehicles are to increase safety on the road compared to regularly motorized vehicles that are most common today, what is the need for licensing human drivers if there is self-driving capability? And to what extent do we need to educate the public? Currently autonomous vehicles are not following through with higher safety nor are users provided with adequate knowledge to control the operations of these vehicles. In fact, “the self-driving car accident rate is higher than the one of human-driven vehicles” (Milenkovic).
When it comes to autonomous vehicles, NHTSA has resisted to further the extent of their safety regulations for the reason that they do not want to slow down innovations (Plungis). But at what cost is NHTSA willing to stop the advancement of their regulatory guidelines in exchange for the successful implementation of autonomous vehicles? There needs to be a limit to the further innovations of autonomous vehicles and a clear vision that puts the safety of the user first, because as of now it seems as if NHTSA is willing to push back user safety for the progress of autonomous vehicles.
Issues with Regulation
Having different standards for different states is an issue. There should not be different expectations for the same law regarding the same entity (autonomous vehicles). Yet, this is not the primary problem with current autonomous vehicle laws, it is the fact that there is not enough regulation imposed. As described for the state of Colorado, there was no consideration taken into account for the presence of an operator. But at the current condition that autonomous vehicles are in now, this should not be an optional decision to address. Having an operator inside the vehicle must be a requirement at all times because currently there are no autonomous cars with the higher levels of automation (where no operator is needed).
The importance is that there is a gap between the development of autonomous vehicle laws and the understanding of the levels of automation. The guidelines in the current laws don’t correlate with what is expected at each autonomous level. Since autonomous vehicles have not reached the level of automation in which no human intervention is required, the presence of operators is necessary. Similarly, the operator must be licensed and the individual needs to have the sufficient knowledge to at least have experience in operating a vehicle in general. But even still autonomous vehicles operate differently than regular motorized vehicles. An alternative to this issue will be detailed for future improvement.
Road safety campaigner, Ralph Nader, speaks on the topic of inadequate regulations and has his own opinion on the future of self-driving cars. Nader claims that we need to “send an urgent message to the casually-minded regulators that Americans must not be test dummies for a powerful, high-profile corporation and its celebrity CEO. No-one is above the laws of man-slaughter” (Gover). This strong statement Nader made regarding Tesla and autonomous vehicles highlights the way in which individuals and even regulators, such as NHTSA, praise self-driving cars to the extent that safety culture is thrown into its shadows. He also makes clear that NHTSA has the authority to execute safety regulations and should use this authority to remove Full Self-Driving technology in Teslas for good. Nader sheds light on NHTSA’s failure to improve safety through further regulation, however, this should not be a reason to completely end the implementation of autonomous vehicles. Instead NHTSA should use their authority to tighten the restrictions enforced by public policy and safety regulations. In these restrictions, the requirement of an operator, licensing, and insurance needs to be addressed for each different level of autonomy that a vehicle is assigned.
Future Actions
Autonomous vehicles have the potential to revolutionize the transportation industry and greatly enhance the safety of our roads. However, in order for this technology to be fully embraced and integrated into society, there must be strict regulations and guidelines in place to ensure that safety is the top priority. The government must enforce strict regulations that are applicable to all states. These regulations must include a set of guidelines for the machine-user interaction, as well as standards for testing of autonomous vehicles. Currently, some states allow for driverless testing, such as in California where they have established the Autonomous Vehicle Tester Driverless Program. However, there should be more testing stages and procedures that involve humans to continue to assess the vehicle’s safety.
Taking these recommendations and changes into consideration will brighten the future of autonomous vehicles and make the idea of self-driving cars a reality. Therefore, to continue the technological advancements of automation, the costs imposed on humanity must be acknowledged and the current errors must be fixed in order to move a step forward in the right direction. A step closer to a distinguished safety culture.
Conclusion
Autonomous vehicles have the capability to either save lives or claim several more depending on how they are implemented. The widespread enthusiasm for this technology is understandable, but we must proceed with caution. Patience is essential for design, testing, and regulation. While “fully self-driving” vehicles have a futuristic appeal, it’s more realistic to take automation step-by-step. Business pressure should not crack our safety systems. By designing for improved human-machine interaction and considering larger system and organizational factors, Tesla’s series of accidents may not have occurred. AVs will be safer with more cognitive engineering research and higher levels of protection.