A few weeks ago I attended MIT’s Second Machine Age Conference, where I heard a number of very interesting presentations on the evolution of AI, robotics, and other advanced technologies. The prospects for truly autonomous vehicles was one of the main topics of discussion. With most other topics, there was considerable audience consensus, but not so with self-driving cars. While many thought that fully autonomous vehicles will be all around us within a decade, others, myself included, were not quite so sure due to the many technical and societal issues involved.
What do we really mean by self-driving cars? There seems to be no precise definition. Are we talking about a human driver assisted by all kinds of advanced technologies, or is there no driver whatsoever? Will such vehicles operate amidst regular human-driven ones, or will they be confined to special lanes equipped with sophisticated navigational technologies? And, is self-driving per se the actual objective, or is it a metaphor for the development of near-crashless cars regardless of whether human drivers are still in the picture.
These questions are not surprising given the very early stages of such a complex area. When exciting new initiatives are first launched, we sometimes describe them using an attention-grabbing phrase that, while potentially unattainable in practice, should be taken more as a marketing pointer to a general direction rather than as a realistic near-term objective.
Similarly, we are now talking about the cashless economy as a metaphor for the evolution toward mobile digital payments. But while expecting that over time a much larger percentage of payments will become digital, few believe that cash will disappear any time soon, - if ever.
While highly sophisticated robots are now being developed for all kinds of exciting applications, they have not quite caught our imagination because, so far, they’ve been generally aimed at industrial applications that few of us have any direct experience with. Cars, on the other hand, are a major part of our daily lives. A self-driven car is thus a concept we can immediately grasp, concrete proof that our machines are becoming really intelligent. The notion that in the not too-distant future we’ll be able to read, work or sleep while an autonomous car drives us around feels like the stuff of science fiction. But, how real is this future, and how long will it take us to get there?
In August of 2012, KPMG and the Center for Automotive Research published a comprehensive report, - Self-driving cars: The next Revolution, - that examined the evolution of self-driving vehicles and their impact on society. The report notes that “in the early decades of the 21st century, the [automobile] industry appears to be on the cusp of revolutionary change - with potential to dramatically reshape not just the competitive landscape but also the way we interact with vehicles and, indeed, the future design of our roads and cities. The revolution, when it comes, will be engendered by the advent of autonomous or self-driving vehicles. And the timing may be sooner than you think.”
These changes are being driven (no pun intended) by a number of forces, most prominently vehicle safety. “In 2010, there were approximately six million vehicle crashes leading to 32,788 traffic deaths, or approximately 15 deaths per 100,000 people. Vehicle crashes are the leading cause of death for Americans aged 4–34. And of the 6 million crashes, 93 percent are attributable to human error. The economic impact of crashes is also significant. More than 2.3 million adult drivers and passengers were treated in U.S. emergency rooms in 2009. According to research from the American Automobile Association (AAA), traffic crashes cost Americans $299.5 billion annually.”
Also, cars are expensive and inefficient. A $21,000 car driven an average of 15,000 miles per year ends up costing its owner more than $40,000 over 5 years, and sits unused almost 22 hours of every day. In addition, “The average American commuter now spends 250 hours a year behind the wheel of a vehicle; whether the value of that time is measured in lost productivity, lost time pursuing other interests, or lost serenity, the cost is high.”
At some point in the future, cars might be able to truly drive themselves and give us back many of those hours we now spend commuting. But, improving driving safety is the overriding objective of most vehicle automation efforts, - a more realistic and achievable objective than coming up with a truly autonomous vehicle.
“Automakers are sneaking features into their newest models that could earn them billions and save 30,000 lives a year - but only if they can convince you to give up control of your car,” notes Fast Company in a recent article, - Self-Driving Cars: Inside the Road Revolution. “Audi, BMW, GM, Nissan, Toyota, and Volvo all have announced plans to unveil an autonomous car by 2020. Google says to watch for its public debut of its prototype in 2016. Still, a truly autonomous vehicle, one capable of dealing with any real-world situation, won't hit showrooms coast-to-coast for years after that.”
The National Highway Traffic Safety Administration (NHTSA) has defined 5 levels of vehicle automation.
- No-Automation (Level 0): The driver is in complete and sole control at all times.
- Function-specific Automation (Level 1): Driver has complete authority but is helped by automation in certain normal or crash potential situations. Examples include electronic stability control, anti-lock braking system and adaptive cruise control.
- Combined Function Automation (Level 2): Driver can cede active control to at least two primary functions but is still responsible for monitoring and safe operations and expected to be available at all times. Examples includes the combination of adaptive cruise control and lane departure warning system.
- Limited Self-Driving Automation (Level 3): Driver can cede full control under certain traffic and environmental conditions, but must be available at all times to take back controls when needed. Google’s self-driving car is such an example.
- Full Self-Driving Automation (Level 4): Human provides destination and/or navigation input but is not expected to be available to control the vehicle at any time during the trip. Vehicle is solely responsible for safe operations and could be traveling occupied or empty.
The current state-of-the-art is someplace between levels 2 and 3. The next steps toward levels 3 and 4 are achievable but will require considerable R&D, according to the Self-driving cars: The next Revolution report.
Current automation functions use sensor-based systems with limited capabilities. “So far, the fusion of available sensors and artificial intelligence is not capable of seeing and understanding the vehicle’s surroundings as accurately as a human being can. Humans use a combination of stored memories and sensory input to interpret events as they occur and anticipate likely scenarios. For example, if a ball were to roll onto a road, a human might expect that a child could follow. Artificial intelligence cannot yet provide that level of inferential thinking, nor can it communicate in real time with the environment.”
Moreover, these automation functions are quite expensive. “Creating a 360-degree view of the vehicle’s environment requires a combination of sensors and may cost more than consumers are willing to pay. Light Detection and Ranging (LIDAR)-based systems provide 360-degree imaging but are complex, expensive, and not yet ready for the market.”
But, adds the report, sensor-based solution are not enough. “What’s missing, so far, is the convergence of sensor-based technologies and connected-vehicle communications that is needed to enable truly autonomous vehicles… Connected-vehicle systems use wireless technologies to communicate in real time from vehicle to vehicle (V2V) and from vehicle to infrastructure (V2I), and vice versa.” To make progress with connected-vehicle systems, a number of major hurdles must be overcome, including:
- Improved positioning technologies: our present GPS isn’t accurate enough for safety-critical applications;
- High resolution mapping: today’s digital maps lack the necessary resolution for such applications;
- Reliable and intuitive human-machine-interfaces: drivers must be able to seamlessly, instantaneously and safely hand off control to the vehicle and take it back; and
- Standardization: standards must be defined to ensure full interoperability amongst all vehicles and infrastructures.
Despite all these challenges, a follow-up 2013 KPMG report, Self-Driving Cars: Are We Ready, was quite bullish: “A year later the momentum around self-driving vehicles is astonishing. In some ways, the industry is moving even faster than we predicted. Rarely does a day go by without another announcement about a new technological breakthrough or a new joint venture. Traditional automotive manufacturers are teaming up with high tech companies; innovative start-ups are seeking and finding investors. The landscape is shifting before our eyes.”
The Fast Company article is somewhat more conservative in assessing the prospects for fully autonomous vehicles (AV): “[T]he first decade of AVs would probably be a mixed bag of semi-autonomous vehicles. Fully self-driving cars are more likely in 20 years. The fleet of cars on the road turns over roughly every 10 to 15 years, so even if driverless cars were in production today it would be many years before they dominated our highways and started delivering the promised benefits.”
“In the meantime, you’re going to see carmaker rolling out the technology in small doses, getting people used to using driver aids and having less responsibility behind the wheel.”
This brave new world is indeed coming, but it will likely get here in stages as features are added to make our cars safer, until one day we can truly take our hands off the wheel.
Excellent essay. I agree entirely. My take: we can get to 95% of the safety benefits without going all the way to the Level 5 car, which will probably cost twice as much (or more) than the Level 3. But the allure of the "Jetsons" car (Level 5) is strong.
Frankly, I think the industry is wrong to pitch safety as the main benefit of these cars: 33,000 traffic fatalities is 33,000 too many, of course. But given we drive about 3 trillion miles a year in the USA, that means we drive about 90,000,000 miles between fatalities. So while with one optic we can say humans cause 90% of these fatalities (true) with another we can say humans are actually pretty good at driving. Will IT be able to beat this 1 in 90,000,000 record soon? I am not sure. And if there is a failure, you can be sure of the headline: "FrankenCar Kills Family of Four." And that will set this all back a long, long way... unfortunately.
I would stress instead the mobility aspect: the opportunity to get millions of elderly or disabled Americans who cannot drive out of their homes and down to the local CVS for shopping or to the clubhouse to meet friends. The value here would be immense, and if we limited top speed to 25 mph the risk of fatality quite low. I see this thinking as behind Google's 'pivot' from high-speed on-highway uses to low-speed local-road uses: the safety claim was going to be just too hard to deliver on, while the mobility claim has immense value and lower chances of failure.
Posted by: Glenn Mercer | November 01, 2014 at 01:07 PM