As technology advances and city design evolves, new mobility services are emerging to bring transportation solutions into the 21stcentury. One of the big trends that is emerging is Autonomous Vehicles. Major automakers and startup companies, such as General Motors and Waymo, are racing to bring AVs to the street. While there are many benefits to AVs, such as giving drivers back time and reducing the number of crashes and deaths from auto collisions, much skepticism surrounds the concept of driverless cars, especially from planners.
The vision for AVs was grand until the public took a step back after Uber’s self driving Volvo XC90 struck down and killed a pedestrian in Arizona back in March. After a lengthy investigation, it was clear that while the vehicle detected the woman crossing the street, it determined the detection of the woman to be a “false positive,” meaning it saw her but believed she was nothing of real concern. It then continued forward, striking and killing the woman. People began to question, if this happened once, what means it will not happen again? AVs should be saving lives according to the manufacturers, not taking them. With this in mind, a major question has arisen: do pedestrians need to change their behaviors in order for AVs to function safely as they hit the road?
When examining how Autonomous Vehicles function, a major flaw has been uncovered that, if not perfected, could cause the loss of more lives and a call to action in the future. This flaw is artificial intelligence systems that cannot predict pedestrian behaviors. While AVs are outfitted with expensive lidar (3D mapping systems), radar, and camera systems to see and analyze the world far better than any human, it is not enough. The systems are not yet smart enough to be able to predict how people will behave, and most systems cannot communicate effectively with pedestrians, as a driver can, to piece together a pedestrian’s intent when they are in the street.
In Amir Rasouli and John Tsotsos’s Autonomous Vehicles that Interact with Pedestrians: A Survey of Theory and Practice, the typical behaviors of pedestrians are analyzed and brought into perspective when relating to AVs. In order for AVs to predict pedestrian behaviors, they first need to understand the intent of pedestrians. Unfortunately, there are many factors that pedestrians take into account when crossing streets, and certain people respond to situations differently as various factors come into play. People fall into different demographics based on gender, race, and age, all of which affect their decision making process. Other physical demographics, such as location, time of day, and weather, in addition to social factors can also affect how a pedestrian may decide to cross a specific street. For example, if a group of people are crossing the road, people tend to walk slower and pay less attention, especially when they have the right of way. They are also less likely to follow another’s actions if they are breaking the law, such as crossing a street when the light is red and conditions may not look adequate for crossing.
Due to this long list of factors that affect decision making, and the general unpredictability of people that may just run out into the street, it is difficult for AVs to be sure what a pedestrian may decide to do. It is believed that the perceived risk of crossing the street should be non-existent once AVs are present due to the fact that they should be able to predict what the pedestrians are doing. The solid fact though is that they cannot predict everything yet, which poses safety issues.
Among the many factors that come into play when pedestrians make decisions, none play such a vital role as vehicle size and speed. By watching a vehicle as it approaches and taking into account both size and speed, pedestrians establish a gap acceptance to decide for themselves if they can make it across the street safely. At signalized intersections, people only look at vehicles 69.5% of the time, while at un-signalized intersections, people look at vehicles 86% of the time because there will be no signal giving them the right of way to cross. In the situations where people are crossing roads without a dedicated signal, regardless of the presence of a crosswalk, communication with the vehicle becomes crucial for a pedestrian to make their final decision to cross.
While 63% of pedestrians claim the right of way by stepping into the road, it does not mean they will necessarily continue to cross the street. They need some piece of communication with the vehicle in order to make their final decision. Usually, communication with the driver is established to make this final decision, most typically eye contact at the very least. A pedestrian will know the driver’s intent from a flash of the lights, a nod, or a hand signal, most times accompanied with this eye contact. In situations where vehicles have fully tinted windows and pedestrians cannot see the driver at all, people are more hesitant because they cannot establish communication, which in essence is what a driverless vehicle will be like. There will be no driver to establish communication with. So then, how are pedestrians going to communicate with the vehicles navigating on the roadway?
As companies continue to develop Autonomous Vehicles, they are still trying to figure out ways for vehicles to communicate with pedestrians. Some ideas currently being explored include LED displays that relay messages, vehicle speed, and intent so pedestrians have more information on what the vehicle is thinking. Another practical approach being considered is tracking pedestrians from the location statistics their phones hold, although this could pose as a personal security issue. Unfortunately, pedestrians cannot make eye contact with these vehicles regardless of what is put in place, and that is a crucial part of how people make decisions when interacting with cars today.
While it is known that AVs are suppose to see pedestrians and stop for them, there currently is no way for a pedestrian to communicate and interact with the autonomous vehicles on the road. The cars cannot fully predict a person’s behavior due to the many factors at play, and without effective communication with pedestrians, there poses a serious risk. Not only can the vehicle not be able to fully predict what a pedestrian will decide to do, but the pedestrian doesn’t know how the vehicle is perceiving their actions and what its reaction will be. While vehicles are being refined in physical design and software, pedestrian behavior studies are being overlooked and underutilized, so cars are not yet learning how to predict pedestrian behavior. If vehicles cannot get this component correct, more pedestrians could be killed in the future by AVs.
After looking at the stated reasoning and current situation, there is an answer to the question: do pedestrians need to change their behavior for Autonomous Vehicles to function properly? As of today, the answer is yes. Andrew Ng, a machine learner who works with AVs, even said “we tell people, please be lawful and please be considerate”(Kahn, 2018). In other words, he is saying, “follow the law and do not jaywalk or act in unpredictable ways, given the cars only know how to react to so much.” This is why current AVs only operate in certain geo-fenced areas, areas they have traveled through many times and are familiar enough with to navigate safely most of the time. Automakers are pushing this type of conversation, as they want to put autonomous vehicles on the road sooner, despite this shortcoming. This has sparked outrage, as people are proclaiming “Jaywalking should not be punishable by death”, as it so happened for the woman killed by the autonomous Uber in March (Schmitt, 2018). There is still no legislation or policies that explain who is to blame when there is a collision between an AV and pedestrian as well, so even the finer points are not sorted out should there be any major collisions.
Regardless of what has been done thus far, the technology is still too new and not smart enough to predict pedestrian behavior and effectively communicate with them. The one shortcut to being able to bring AVs to the road sooner is having pedestrians be more aware and change their behaviors so AVs can understand their intent better. This shortcoming, which manufacturers are pushing on the public, is something many are not happy with adhering to. While the situation could change in the future as machine learning and communication between AVs and pedestrians advances, for now pedestrians will have to be weary if they want to be safe as AVs begin to navigate some of America’s roadways.
Kahn, Jeremy. "To Get Ready for Robot Driving, Some Want to Reprogram Pedestrians." 16 August 2018. Bloomberg. 18 December 2019 <https://www.bloomberg.com/news/articles/2018-08-16/to-get-ready-for-robot-driving-some-want-to-reprogram-pedestrians>.
Schmitt, Angie. "Self-Driving Car Makers Prepare to Blame "Jaywalkers"." 17 August 2018. Streetsblog USA. 18 December 2018 <https://usa.streetsblog.org/2018/08/17/if-self-driving-cars-cant-detect-jaywalkers-they-shouldnt-be-on-the-roads/>.
Tsotsos, Amir Rasouli & John K. "Autonomous Vehicles that Interact with Pedestrians: A Survey of Theory and Practice." Report. 2018.