False positives plague the automobile crash detection industry, threatening to kill consumer trust and diminish the viability and profitability of business models based around crash detection programs, such as driver and fleet safety, post-crash and white glove services.
This problem was highlighted last year by the tragic accident as a result of a false interpretation of a sensor reading that led to the death of a Tesla driver. This incident was highly publicized and created significant public sentiment against automobile automation, making acceptance and adoption more difficult.
Those of us in the transportation safety industry should take note. False positives in the crash detection and response industry destroy consumer trust, introduce unnecessary costs, create negative brand impressions, and can even be dangerous to everyone on the road surrounding the incident, which invites legal risk.
For these reasons, one of the most important factors in developing a viable crash detection algorithm is to reduce—with the intent to eliminate—false positives. Subsequently, it’s also one of the most important factors to consider when evaluating a solution.
Let’s start with a simple definition. False positives are when a crash detection technology falsely identifies a crash event that in reality hasn’t occurred.
False positives occur in all three types of crash detection technologies, embedded, aftermarket and smartphone-based. So, to eliminate false positives, the first place you have to look is at the data collection source itself, the sensors. Embedded and aftermarket sensors have operational limitations relative to those in smartphones. The former rely on airbag sensors that are designed to trigger only at speeds of about 30 mph and above. Even newer vehicles enabled with ultrasonic bumper sensors that can identify low-speed accidents don’t fully address the problem, due to physical limitations outside of the speeds that they are designed for (typical parking speeds). Therefore, most solutions don’t even capture a significant portion of crashes on the road.
To capture the full range of crashes at various speeds, including those in the lower speed bands, smartphone-based solutions have more and sometimes even better sensors than those in embedded and aftermarket solutions on the road today. If that sounds unreasonable to you, simply consider the cost. Embedded and aftermarket solutions are under incredible pressure to keep the cost down. Traditionally, those device might cost about $75.00 and contain only a minimum number of sensors. Whereas smartphones can now cost upwards of $800.00 and usually contain about 14 sensors.
Another disadvantage of embedded and aftermarket solutions is their lack of portability. Smartphone-based protection covers users no matter which vehicle they are in, or whether they are the driver or the passenger.
Inadequate development and testing
For individuals evaluating solutions, embedded and supported aftermarket solutions do have one benefit. They are generally thoroughly tested by the automotive OEMs who support them. Most smartphone-based solutions don’t go through that thorough of a vetting process. Their accuracy most often must be taken for granted based on the word of the service provider.
Which means there are many solutions on the market inadequate to the task of saving lives or sustaining business models. In good part, because of the high numbers of false positives they generate. A grossly inferior crash detection algorithm might indicate that a dropped or tossed phone is a crash. But even reasonably effective algorithms often misinterpret hard braking and cornering, and even simply moving the phone, as crashes.
The problem is exacerbated in the lower speed bands, where the tradeoffs become more difficult when tweaking the sensitivity of the algorithm. This is one reason why there are only a few solutions coming to market that even claim detection in the lower speed bands.
Perfecting the algorithm involves identifying real world considerations and making decisions at every level of development to ensure the developers are intelligently responding to those considerations, from system architecture (such as on-device or cloud-based processing) to crash testing methodologies. It can take years of testing, generating data and developing an AI classifier (if they’re using one, and they should). Yet, not having all these critical items in place doesn’t seem to stop many companies from bringing their products to market.
Which leads us to the four ways that those inadequate solutions and their false positives can destroy the business of crash detection programs.
#1 Loss of consumer trust
If consumers consistently received false crash notifications from their supposed safety app, they will lose trust and no longer feel secure. They might seek out another alternative or give up on the category of safety services entirely. That erosion of trust hurts everyone in the industry. #2 and #4 below are additional situations that fall under loss of trust.
#2 Negative brand impressions
Negative experiences lead to negative brand impressions. False positives are definitely a negative experience (see #4). This is most problematic when the crash service is white labeled, now carrying a brand that wasn’t involved in the development of the algorithm, but is the brand offering the service to its customers, employees and their families. When things go awry, people blame the brand with the logo in the app.
#3 Increased costs
False positives can also skyrocket the budgets at call centers, where every call costs money, $8.00 on average just to say, “Hello, how may I assist you.” In addition, in fleet situations, the calls and questions about false positives don’t stop there. They’re escalated to managers, who spend time assessing the situation, talking to the provider and generally trying to quell the rising dissatisfaction and loss of trust within the fleet. After all, they must defend their decision to implement the service.
#4 Dangerous workarounds that lead to legal risk
False positives are such a huge problem in the industry that many solutions are turning to dangerous workarounds that open themselves up to legal risk. Those solutions send a notice after an accident detection that reads, “We detected a crash. Do you need help?”
They do this because the road leading to reducing false positives is long, expensive and complicated. The easy out is to put the burden on the smartphone holder. But that’s not an option if he or she is unconscious. Or startled and in shock. Or out of reach of the phone. Or simply tired of answering those pop-ups.
Plus, they might be driving. Receiving a pop-up is a distraction in itself that can cause an accident and even a lawsuit.
Imagine a situation where a mother is driving her kids to school. It’s raining. Visibility is poor. A truck is bearing down on them from the rear. Suddenly, she sees a large pothole, swerves to miss it, but catches the corner, which causes a thump and sets off her crash detection service. She steers the car toward the right lane. But the truck has misjudged her move and is now heading on a collision course. In milliseconds, her heart is racing, her blood pressure is up. The kids are yelling. And now, at that moment when she needs all her wits to react quickly and effectively, the phone “pings” with a message that says, “We detected a crash. Do you need help?” It doesn’t take a great imagination to envision a bad outcome with legal ramifications.
Reducing false positives isn’t easy
It’s extremely difficult to get a crash detection algorithm right and reduce false positives. To do so is costly and time consuming. Most players in the space who are not focused on enabling safety technology don’t have the resources to properly develop their technology to reduce false positives and to show the data to prove it.
The more sophisticated solutions spend considerable resources conducting crash tests, collecting data and then tweaking their algorithms and artificial intelligence (if they use it).
It takes consistent effort and budget to achieve the optimal balance necessary for an accurate algorithm. Which is perhaps why the author of a Wired article on false positives chose to include these words in his title, “The Agony of Knowing What Matters.”
Protect the Industry by Making Smart Choices
The only way to ensure that the crash detection industry remains strong and has a robust and secure future is to educate the companies purchasing these solutions. They should select only top-tier crash detection services from companies dedicated to safety technology (where it’s their core business, not ancillary to it). If substandard services continue in the marketplace, false positives will continue to inflict damage to the industry.