The Social Integration Gap Facing Driverless Cars
Autonomous vehicles are significantly more complex than anything else on the road today, their behavior is still developing and mammoth amounts of data collection and analysis will be needed to get them to a state of safe integration without a trained backup driver.
Automakers and tech developers have demonstrated they know how to create cars that drive themselves.
At this point, AV basic driving skills are nearly impeccable. AVs can quickly plot a path and dynamically alter that path as traffic, construction and other obstacles become known. They can merge into traffic, identify most objects with an array of sensors and avoid collisions.
However, converting cars from being controlled by humans to being independent decision-makers is not just about going from one place to another, it also is about how they interact with others along the way.
The underlying issue is that AVs are significantly more complex than anything else on the road today. Their behavior is still developing and mammoth amounts of data collection and analysis will be needed to get them to a state of safe integration without a trained backup driver.
As a result, Artificial Intelligence/Machine Learning (AI/ML) systems for AVs have been designed to be very cautious drivers. We all have experienced very cautious human drivers. They tend to slow or stop when you least expect it. They drive slowly and behave in ways that can be unpredictable and frustrating. They also seem to have little ability to anticipate human behaviors and estimate what human drivers, pedestrians or bike riders are likely to do.
Similarly, it seems an AV may do a great job of identifying objects in its direct line of sight, but it doesn’t appear to be able to look ahead of the vehicle, out in front as humans have learned to do, to drive defensively.
In other words, AVs have a social integration gap (SIG). The SIG looms large! Closing the SIG certainly will be a process, not an event; it will take a lot of time for both humans and AVs to adjust to each other. How can we get from our current state of autonomy to a more advanced and human-like autonomy?
AV Database Still in Formative Stages
For starters, with AVs still in the testing phase, automakers are collecting, storing and analyzing AV incident data in‑house, using a myriad of hardware and software tools; many are proprietary.
In October 2017, General Motors’ Cruise autonomous test vehicle was rear-ended while passing through a green-light intersection. GM announced in its 2018 Self-Driving Safety Report it is collecting data on both an event data recorder (EDR) and an unspecified data logger. While that may be feasible at this early stage, as AVs are more widely deployed, third parties likely will be ultimately responsible for collecting and analyzing the information.
Part and parcel, the U.S. Census in 2009 stated 8% of all drivers on the road were involved in collisions that year. If we put 200,000 AVs on the roads over the next couple of years and only 4% of them (it is assumed AVs will drive more safely) are in accidents, there will be 16,000 incidents annually that will need detailed investigations. That number would grow each year as more AVs were released to the public.
As the incident volume grows with the release of more AVs, many stakeholders will be eager to investigate them for a variety of reasons (improve performance, avoid liability, assign liability, evaluate safety, et al).
Thus, a standard and common way to see what happened and how the AI/ML handled the vehicle and situation will be essential. This will save time and will deliver consistent analysis input so that supporting tools can be developed that will speed evaluations and minimize tampering with the data to manipulate outcomes.
Standardization also will provide a common signal interface that first responders can use to get an idea of what happened while at the scene.
A standardized advanced data recorder, or black box, also will be an important tool to move data and video collection from the hands of developers to others who need the information. Video capture alone delivers a great benefit, as many situations will arise where occupants will not be dependable witnesses.
Combined with system data, a complete picture of events leading to an incident and how the AV behaved immediately afterward will provide important value-added in the pursuit of closing the AV SIG.