How Autonomous is Autonomous?
If you work in transportation and know anything about autonomous (or driverless) vehicles (AV), then you have probably heard about the different levels of automation developed by the Society of Automotive Engineers (SAE). These levels describe the scale of control that different autonomous driving systems have in relation to a human operator ranging from 0-5, where 0 means the human controls all aspects of driving, 5 means the human controls no aspects of driving in any condition, and the numbers in between assign different levels of responsibility to the driver in certain conditions. However, even within those levels, the percentage of time an AV is driving itself can vary widely, which is important when assessing a new AV project or company within a massively diverse industry.
With hundreds of billions of dollars invested in autonomous transportation, today’s tip of the industry spear has matured to level 4 along that scale, where the human becomes a passenger that has no responsibility for driving within certain environmental constraints, like the speed limit of the road, weather, or complexity of an intersection. Today’s level 4 AVs come in different forms and shapes, ranging between completely “normal” vehicle platforms you might not notice driving next to you, to completely untraditional vehicles designed from scratch without steering wheels and seats that face inwards. However, what they share is the ability to transition away from a human driver for the majority, if not all of the time, which is probably the most significant cost and supply barrier to better transportation of people and goods.
Level 4 AVs may all share the ability to drive autonomously, but the reality is that there are tremendous discrepancies between the amount of time and the complexity of scenarios an AV can handle on its own. There are hundreds of past, current, and planned Level 4 AV projects in the US. Many of these projects are small pilots demonstrating the ability of currently available automated transport systems in preparation for more robust deployment of “real” commercial applications in public transit, freight or goods delivery.
With that goal in mind, if the AV is only operating by itself 60% of the time along the route, or it transitions control to an onboard operator at critical intersections, those data are important for determining it’s readiness for expansion or, ultimately, the removal of any onboard safety attendant. Most AV companies don’t disclose this data, but their abilities vary widely. The same driving environment (referred to as an operational design domain in AV “lingo”) might be navigable by five different AVs, but they may all navigate them differently, each with different challenges and advantages.
It is important when vetting AV companies to ask them questions about their percentage of time in AV mode for past deployments and expectations for the project you’re discussing. What are the limitations along the route that would likely need human intervention? Some companies may operate at slower speeds or less complex environments but have a much higher percentage of time in autonomous mode, whereas others may drive faster or in more complex environments but depend on a human operator more frequently.
In the case of the Mines Rover AV fleet, the EasyMile shuttles were operating in autonomous mode nearly all of the time, with manual confirmation only at certain intersections. When assessing AVs for a project, one should consider how do technology limitations align with the desired outcomes of the project? If the goal is to progress toward completely driverless operation, then matching an AV company’s ability with your desired route may require the project to sacrifice speed, capacity, or other things in favor of higher autonomy.
What risks do these data unveil? For example, the more frequently a human operator has to intervene, the more risk or liability the operator may take on (as opposed to the AV developer) should something occur. In the case of the Mines Rover in the AvCo program, the operator and the AV company were separate parties.
In the same vein, both transportation planners and the general public must be wary of misleading information and false advertising that promote vehicles with advanced driver assistance systems (ADAS) as being fully driverless. Some of the innovations on the roads today that are labeled as “self-driving” fall on the lower end (1 - 3) of SAE’s levels of automation scale. One harmful effect of this equivocal advertising is that it has the potential to blur this wide range of technology solutions, which all require different levels of human involvement. The reality is that the vast majority of issues that occur with autonomous transport technology and deployments happen when a human is responsible for driving or when the vehicle is operating in manual mode.
Hopefully this is helpful in planning an AV project. Nobody expects a transit agency or other AV “customer” to be an expert on autonomous vehicles, but asking the right questions will equip decision makers with the right information to make the best decisions possible as they push to improve transportation with AVs. Tools like AvCo’s CityForward can help communities prepare for AV deployment by equipping them with the questions that they need to ask and the considerations that they need to address related to autonomy and beyond.
