The sine qua non for the CAV (connected autonomous vehicle) is communications. It is at the same time its strength and, borrowing from Greek mythology, its Achille’s heal. To function, autonomous vehicles must rely on a tremendous amount of inter and intra connectivity. All of the on-board sensors (lidar, radar and cameras, engine parameters, lane departure, etc.) have to flawlessly communicate with one another, as well as vehicle to vehicle, and communicate with traffic management (lights, flow, emergency vehicles, etc.).
Sounds great in theory, but in actuality this is astoundingly difficult to pull off. Keep in mind, this connectivity has to function flawlessly all of the time. There was a bit of irony at CES 2017 in that every presentation I attended experienced a problem at least once with the remote presentation control unit communicating correctly with the media controller equipment. And this was connectivity at its very basic level! On a more complex level, there was Faraday’s problem during the press review where their car failed to accept the command to self-park.
Obviously, you can’t have a break in connectivity or the autonomous vehicle will come to a complete (unintended) halt (hopefully), and in doing so will become a potential accident instigator for both other autonomous and non-autonomous vehicles. What level of redundancy will be sufficient to prevent a loss of connectivity? While it seems feasible that intra- vehicle (between its numerous components necessary to have an autonomous system) redundancy is reasonably surmountable, what will be necessary to ensure the inter-vehicle, and traffic management, along with live web connectivity, is flawless?
Simultaneously with ensuring the continuous flow of connectivity, there are still two large problems to solve: All communication has to be hack proof (we have seen the videos of someone remotely gaining access to a vehicle’s electronics via one of the communication channels, and taking over one or more of the vehicles systems- acceleration, braking, steering. Hackers have demonstrated this remotely on cars ranging from Jeeps to Teslas.). Further a great deal, if not all, of the information has to maintain the privacy of the vehicle (and its occupants).
Additionally, complicating the connectivity issue is what was tagged “Babel” at the CES 2017 A United Language for the Connected Car session. The general definition of babel is a confused noise, typically made by a number of voices. Unfortunately, it applies to the current status of proprietary software designed for many of the components needed for a connected vehicle. The herculean challenge is to get a universal open language used across all components/systems for autonomous vehicles. Beyond the current Babel-of-software-language is the growing quagmire of state and federal regulations aimed at controlling autonomous vehicle access to our roads. Currently, an autonomous vehicle approved by nascent laws in one state, may not be able to continue driving when it crosses into an adjacent state. For example, while an autonomous car can be driven in Nevada, it can’t legally continue into nearby Oregon or Idaho, and if you are in an autonomous car in Florida, you could not continue on into any of its adjoining states.
The RAND Corporation pointed out in their 2016 publication Autonomous Vehicle Technology: A Guide for Policymakers, that rather than autonomous vehicles reducing congestion on our roads, they may, in fact, increase congestion. This conclusion is based on the reduced transportation costs borne by individuals. For example, the cost of automotive insurance shifts from the owner to that of the manufacturer of autonomous vehicles. This, combined with increased access (potentially no need for individual driver licenses), could see a substantial surge in the number of individuals travelling at the same time. Of course, it could be moderated by increased reliance on mass vs low occupancy vehicles. The elimination of the hassle often associated with finding a parking space (your autonomous vehicle could drop you off and then continue on to a remote parking area, awaiting your request for it to comeback and pick you up) may also contribute to a significant increase in willingness to ‘hop’ into your vehicle and head to a dense, high-use, urban area.
What are the implications for the potential loss of transportation sector jobs, their respective incomes and loss of tax revenues from reduced or eliminated parking garages, meters, etc.?
And while most believe that autonomous vehicles (or even semi-autonomous) will significantly reduce the number of deaths caused by crashes, the is one part of our society that has depended on these deaths- that of organ donations. “It’s morbid, but the truth is that due to limitations on who can contribute transplants, among the most reliable sources for healthy organs and tissues are the more than 35,000 people killed each year on American roads (a number that, after years of falling mortality rates, has recently been trending upward). Currently, 1 in 5 organ donations comes from the victim of a vehicular accident.” [From Future Tense: The Citizen’s Guide To The Future. Dec. 30 2016] The potential impact is catastrophic on an already stretched organ donation system. “All of this has led to a widening gap between the number of patients on the organ wait list and the number of people who actually receive transplants. More than 123,000 people in the U.S. are currently in need of an organ, and 18 people die each day waiting, according to the Department of Health & Human Services. Though the wait list has grown each year for the past two decades, the number of transplants per year has held steady in the last decade, at around 28,000.”[ Fortune: If driverless cars save lives, where will we get organs? By Erin Griffith Aug 15, 2014].
You may be familiar with the paradox of Buridan’s ass. As the story goes, a hungry donkey was placed equidistant between two identical bales of hay. Unable to choose which one to go to, the donkey died of starvation. The movement towards autonomous vehicles has at least two analogous conundrums: how many deaths by autonomous vehicles is an acceptable number of deaths, and, who is going to have the final approval of the algorithms designed to make a decision for an autonomous vehicle as to who should be sacrificed when a choice has to be made between certain death in a pending accident. The analogy is that if we can’t reach agreement on both of these issues, the movement towards autonomous vehicles may come to a halt.
Even though these two conundrums are inextricably related, let me briefly explore each separately. We know factually that autonomous vehicles can lower deaths currently associated with driver error, and that the number won’t rapidly be reduced to zero. Using the approximately 32,000 automotive related deaths per year (cited in my Part 1), what percent reduction would be ‘acceptable’? Would a 50% reduction resulting in 16,000 fewer deaths per year, but also 16,000 remaining deaths per year by autonomous vehicles be OK? Would it take a 75% reduction resulting in 8,000 deaths per year by autonomous vehicles to be considered OK? The consensus appears to be that while the astounding number of 32,000 deaths per year caused by human error behind the wheel, isn’t good, we seem to have ‘accepted’ it without demanding immediate action on a national or global level. However, few believe we would be as complacent if the news was filled with 16,000 or even 8,000 deaths per year as a result of autonomous vehicles.
Recently a number of articles have appeared highlighting the other conundrum: algorithms being designed to decide who lives and who dies when the outcome of a pending accident is unavoidable. For example: “A self-driving car carrying a family of four on a rural two-lane highway spots a bouncing ball ahead. As the vehicle approaches a child runs out to retrieve the ball. Should the car risk its passengers’ lives by swerving to the side—where the edge of the road meets a steep cliff? Or should the car continue on its path, ensuring its passengers’ safety at the child’s expense?” [Driverless Cars Will Face Moral Dilemmas by Larry Greenemeier, June 23, 2016, Scientific American] Or:” Imagine you’re behind the wheel when your brakes fail. As you speed toward a crowded crosswalk, you’re confronted with an impossible choice: veer right and mow down a large group of elderly people or veer left into a woman pushing a stroller.” [Driverless cars create moral dilemma. By Matt O’Brien, The Associated Press January18, 2017]. Who should be entrusted with developing and ultimately approving the necessary algorithms? Shall there be one algorithm for all autonomous vehicles globally or will there have to be country/culturally specific versions?
Real World Impediments To Fully Autonomous Vehicles:
At this point, autonomous vehicle developers have not been able to handle several frequent occurrences typical to our driving environments. If a fully autonomous car comes upon road construction, it doesn’t know how to ignore the programming that tells it not to cross a double yellow line, or purposely drive into a temporary lane without lane markers. It is basically programmed to shut down- or, in Nissan’s case, phone ‘home.’ At CES 2017, Carlos Ghosn, Chairman and CEO of Nissan, during his keynote speech said they are planning on having a centralized station staffed 24/7, to handle “edge” circumstances for their autonomous cars. In logic, the human contacted by the autonomous car would review the information available from the on-board sensors, and map an alternative route or action. It is unclear how would this approach be able to scale up instantaneously, for example, when a large section of a country has an extreme disrupter such as flooding, earthquake, etc.?
Similarly, autonomous vehicles cannot negotiate a dirt road, or a road that lacks up-to-date gps mapping. Neal Boudette in his article “5 Things That Give Self-Driving Cars Headaches” points out, autonomous cars will have a very hard time with unpredictable reckless drivers on the same road in a non-connected vehicle [New York Times, June 4, 2016].
Current thinking of many developers, is to require a (human) driver to serve as ‘back-up’ in those circumstances where the autonomous or semi-autonomous vehicle encounters a situation it isn’t programmed to handle. Unfortunately, there are severe limitations to how well most drivers would be able cope with such an unexpected/instantaneous hand-off (one doesn’t have to look any further than the tremendous increase in accidents attributable to drivers distracted by texting). The biggest problem is with a lack of sufficient reaction time even at moderate speeds, let alone highway speeds. This is further complicated by the well documented fact of vigilance decrement. The longer the autonomous vehicle is properly handling the driving, the less attentiveness and readiness the ‘back-up’ human will have to properly respond to the hand-off.
In order to succeed, there is going to have to be a significant educational effort of the current, and potential, driving public during the transition period when autonomous and semi-autonomous vehicles share the road with traditional non-connected vehicles. Part of this education will need to focus on the trust issue confounded by demographic and age differences in acceptance.
In some ways, many of the concerns today are parallel to those around one of the earliest autonomous vehicles designed to transport people- the elevator. Original elevators were relatively dangerous vertical transport platforms, operated by a trained elevator operator. As safety concerns were addressed, elevators vastly improved including having doors, fixed stopping points, redundant mechanisms to prevent free fall, etc. Shortly after the turn of the twentieth century push buttons were introduced that would permit selecting a specific floor and the elevator to proceed automatically to that floor. However, it wasn’t until after World War II -forty years after automation- that elevator operators were no longer placed in most elevators. One of the main reasons for the slow transition from manually operated to fully automated elevators was people were fearful of getting into an elevator that did not have a human operator. How likely are you to entrust your life to the newest mode of autonomous vehicles?
Autonomous Vehicles Part 3 will explore: What is next? Is the light at the end of the tunnel daylight or an oncoming train?