Autonomous Vehicles - by Adam

Introduction

SAE International defines an autonomous vehicle as a vehicle in which the dynamic driving task is performed by the automated driving system, rather than by a human driver. That is to say that the car itself is responsible for the operational (i.e., steering, braking, accelerating) as well as tactical (i.e., lane changes, indicating, responding to changing traffic) aspects of driving, requiring little to no human input (SAE International 2014).

State of the Technology

Vehicle automation can be broken up into six categories, as seen in Figure 1, ranging from no automation at all (a standard, human-driven car), to full automation, that is no human intervention required no matter the conditions or circumstances (SAE International 2014). Levels 0 - 2 are classified as ‘driver support features’, while levels 3 - 5 are considered ‘automated driving features’ (Aptiv 2020).

Trials are being conducted in several locations as to the safety and viability of driverless cars, typically taxis within small, designated testing areas. In Las Vegas it is possible to ride in a robo-taxi via Lyft in partnership with Aptiv, a service that has provided over 100,000 commercial rides as of February 2020 (Iagnemma 2020). While fully autonomous, the rides have always had a safety driver present in order to take over if need be and to operate the vehicles on private property (Akers 2020).

Tesla also claims to have hardware capable of level 5 automation in all new vehicles, along with the ability to upgrade and update relevant software remotely. For them it is a waiting game for the data to show that their vehicles beat human reliability over billions of miles of experience, along with the regulatory approval necessary for these cars to operate legally (Tesla 2021). This data is being gathered by having the autonomous driving system operate in ‘shadow mode’, whereby it records what actions it would have taken without taking them. This then allows Tesla to compare those actions with what the driver did in the case of an accident and to determine if the car’s actions would have succeeded in avoiding it (Golson 2016).

 

 

Figure1: Automated Driving by SAE International, 2014

So, while some measure of autonomous vehicle is available today, with the potential for level 5 automation, it will perhaps be a while before this is readily available for consumers. The timeline, of course, will depend not only on proving the reliability of the software, but also acceptance by regulators, which may vary greatly depending on location (Tesla 2021). It seems certain that driverless cars will grace our roads in the not-too-distant future.

This, of course, is all focused-on passenger vehicles with no driver, which is merely one possibility. Another concept, realised by nuro in early 2021, is a vehicle with nobody inside whatsoever, used to transport goods. The low-speed vehicle has been granted a driverless exemption from the NHTSA as well as an Autonomous Vehicle Deployment permit from California’s DMV (Korosec 2020). In fact, as of April 2021, fully 55 companies have been granted permits to test autonomous vehicles with a driver in place, yet as of May 2021 only eight have been given permits for driverless testing. And only nuro has been granted an actual Deployment permit (California DMV 2021). The complete lack of a person in the low-speed vehicle allows nuro to completely prioritise the safety of other road users over that of the vehicle, going a long way to helping it gain the permits it needs to operate commercially (nuro 2021).

The technology required to make these vehicles safe is manifold. For instance, nuro uses 360° cameras, Lidar (light detection and ranging), short- and long-range radar and ultrasonic sensors (nuro 2021). Lidar, which maps the car’s environment using laser beams emitted in all directions, can be used to assist the deep learning AI typical of autonomous vehicles to recognise objects within range as the AI can ultimately make mistakes. However, it requires that the driving environment be pre-mapped with lidar to function in real time, making it unscalable to all locations and adding great difficulty to maintaining an up-to-date infrastructure. This lack of adaptability, according to Andrej Karpathy (Tesla’s Chief AI Scientist), is why Tesla doesn’t use lidar. They instead have focused their attention on a pure-vision-based approach, relying entirely on eight cameras surrounding the car, and the AI to interpret the information, forgoing even the addition of radar (Dickson 2021).

What is the likely impact?

With nuro delivery vehicles already in operation in Houston, it’s easy to see the potential to revolutionise the freight and delivery industry. In 2017, there were 4,102 deaths in truck wrecks in the US, a 52% increase since 2009, with fatigue listed among the most common causes (Kopestinsky 2021). The opportunity to eventually increase road safety considerably is apparent. Further, as self-driving vehicles become road-legal and then more and more prevalent, research from the University of Cambridge has demonstrated how they can improve the flow of traffic by at least 35% by communicating with each other (Symonds 2019). While current means for drivers to communicate between cars are essentially limited to brake lights and indicators, autonomous vehicles could communicate intent to each other to not only reduce traffic, but also improve safety.

There are also concerns however for example in 2018, it was reported that a self-driving Uber hit and killed a woman at a pedestrian crossing in Arizona. While the local Police Chief noted, after viewing footage of the incident, that it would have been ‘difficult to avoid this collision in any kind of mode (autonomous or human-driven)’ and that ‘it appears that the Uber would not be at fault’, it still raises questions about the current and future safety levels of driverless vehicles (Maki & Sage 2018).

Ultimately, the probable disruption of the taxi and freight industries will lead to those professional drivers being the most affected by the legalisation of autonomous vehicles. While these jobs may one day become entirely replaced through automation, the benefits to the consumer in availability, safety, efficiency, and reliability of vehicles are hard to ignore.

How will this affect you?

As the technology improves and proves itself, gaining the acceptance of lawmakers and regulators, all other road users, including cyclists and pedestrians, will be affected. Whether it be their impacts on safety or their potential to improve traffic, driving will never be the same once autonomous vehicles become common on the roads.

There are other questions to be considered before we can understand the full impact. If communication between the vehicles is utilised, will it be universal or simply Tesla-to-Tesla, nuro-to-nuro? If it is universal, how will that impact and interact with human-drivers? At what level of prevalence on the roads will the benefits of autonomous vehicles have a significant impact?

There are also considerations of responsibility. While the safety of driverless vehicles is often questioned, it is worth noting that human drivers are not without danger. But when a human makes a mistake driving, it is obvious who is responsible. If an autonomous vehicle is in an accident, even if it is one hundredth as often as a human driver, there is a question of who is morally and legally responsible, whether it is the owner of the vehicle, the manufacturer, or humans within the vehicle. Will I be responsible if I am the only person in a driverless taxi that crashes? In this respect, the legislation needs to move quickly to keep pace with the technology.

Finally, there is the question of privacy and security. The potential for dangerous criminal activity, either as a vehicle or as a weapon, if a car can be hacked and controlled remotely cannot be ignored. Even simply protecting location data will be extremely important, as will securing the feeds of the necessary cameras used by the AI.

There is a long way to go before driverless vehicles take over the roads, but the foundation is there already. As the burgeoning technology grows, it seems an inevitable evolution of human transport.

References:

Akers, M 2020, ‘Robotaxi service that proved itself in Las Vegas plans to go national’, Las Vegas Review-Journal , viewed 5th July 2021

Aptiv 2020, What Are the Levels of Automated Driving?, Aptiv, viewed 5th July 2021

California Department of Motor Vehicles 2021, Autonomous Vehicle Testing Permit Holders, viewed 5th July 2021

Dickson, B 2021, ‘Tesla AI chief explains why self-driving cars don’t need lidar’, VentureBeat , viewed 5th July 2021

Golson, J 2016, ‘Tesla’s new Autopilot will run in ‘shadow mode’ to prove that it’s safer than human driving’, The Verge , viewed 5th July 2021

Iagnemma, K 2020, 100,000 Self-Driving Rides Strong , Aptiv, viewed 5th July 2021

Kopestinsky, A 2021, ‘24 Disturbing Truck Accident Statistics (2021 Edition)’, Policy Advice , viewed 5th July 2021

Korosec, K 2020, ‘Nuro can now operate and charge for autonomous delivery services in California’, Tech Crunch , viewed 5th July 2021

Maki, S & Sage, A 2018, ‘Self-driving Uber car kills Arizona woman crossing street’, Reuters , viewed 5th July 2021

nuro 2021, FAQs , nuro inc., viewed 5th July 2021

SAE International 2014, Automated Driving, SAE International, viewed 5th July 2021

SAE International 2014, Automated Driving, Online JPEG Image, viewed 5th July 2021

Symonds, D 2019, ‘Communication between self-driving cars key to improving traffic flow’, Autonomous Vehicles International , viewed 5th July 2021

Tesla 2021, Autopilot , Tesla, viewed 5th July 2021