April 8, 2018
By Bob O'Donnell
FOSTER CITY, Calif. — With the recent tragic accident involving a self-driving Uber vehicle that struck and killed a pedestrian outside Phoenix, as well as unanswered questions about a recent Tesla Model X crash in Silicon Valley, there’s soul searching in the automotive tech world.
More people are also starting to think about setting realistic expectations for self-driving cars, the essential question being whether they can be expected to completely avoid fatalities or whether it’s good enough that they reduce them.
Obviously, there’s no simple answer. The ethical implications are far-reaching. What makes this question particularly troublesome is that it ties together computing technology with life-and-death consequences. While there have been plenty of hypothetical discussions in the past, this incident has made the possibility of “death by machine” a disturbing potential reality.
Full details on both incidents are still coming out, so there shouldn’t be any rush to judgment as to the ultimate cause of the accidents. However, the technology built into autonomous cars such as the ones involved generate significant amounts of data that is already making the process of determining the cause much faster and more definitive than traditional investigative processes. This is why, for example, Tesla was able to report that it knew that Autopilot, it's partial self-driving feature, was engaged on the Model X that crashed last month in Mountain View, Calif., killing the driver.
From a technical perspective, many of the questions about the potential safety have to do with the sensors which collect all that data. While specific implementations vary by vendor, most self-driving cars have a collection of traditional cameras, radar and LiDAR (a type of sensor that bounces laser light off nearby objects) built into them.
In theory, all of these components work together to enable the creation of a view that provides the car with all the information it needs to make real-time driving decisions. Importantly, radar and LiDAR have the ability to essentially see through objects, allowing them to provide views and perspectives that cannot be seen by humans.
This is relevant for the Uber accident in Tempe, Ariz., because the technology should have been able to “see” that there was a pedestrian on the side of the road, even if she was hidden from human view by cars or other objects, and slam on the brakes. These vehicles are supposed to be able to see things that people can’t and react in ways that are faster and better than a human ever could.
While many in the tech industry have focused on convenience and new business models, the fundamental benefit most carmakers talk about — and most consumers want — is safety. In fact, Tesla CEO Elon Musk has even said that critics of autonomous cars are “killing people” by not enabling their safety benefits.
Thankfully, key automotive tech suppliers recognize this and have been focused on rigorous functional safety standards. After several years of development, they are now able to sell or license parts that meet the latest standards. While those details may never show up on your car’s spec sheets, they provide an important safety net for things such as redundant systems and the ability to operate in challenging weather environments that are essential for building safer, more reliable cars.
As a result of the Uber incident, there also have been changes in autonomous vehicle testing plans by tech companies, as well as regulatory permissions from governmental agencies, such as the state of Arizona. Though tragic, the accident has triggered a level of discussions on both a technological level as well as a societal level that, frankly, should have occurred before it happened.
Realistically, it may be difficult to completely prevent deaths even with autonomous vehicles, particularly because both human-driven and self-driving cars will coexist for decades to come. To make the testing process safer, however, it likely will require different approaches. One particularly interesting approach is to use simulated, virtual driving environments, similar to the new virtual reality-based Nvidia Drive Constellation system the company unveiled this week.
While simulated systems can’t completely replace real-world tests, they can both offer critical benefits as well as reduce potential accidents with development systems. They enable significantly more test miles be driven, and different scenarios to be tested, than can happen with real-world driving. This is important because the safety of autonomous cars is highly dependent on the systems inside them being able to recognize situations they have “seen” before and respond appropriately. The more situations they experience, the safer they will be.
Challenges for autonomous cars still remain, and the realistic time frames for getting them onto the road likely will lengthen as a result of these recent accidents. Nevertheless, they still represent an important step forward in improving the overall safety of everyone on the road.
Here’s a link to the original column: https://www.usatoday.com/story/tech/columnist/2018/04/08/how-safe-should-we-expect-autonomous-cars/451494002/
USA TODAY columnist Bob O'Donnell is the president and chief analyst of TECHnalysis Research, a market research and consulting firm that provides strategic consulting and market research services to the technology industry and professional financial community. His clients are major technology firms including Microsoft, HP, Dell, and Intel. You can follow him on Twitter @bobodtech.