How AI meets the auto industry’s safety imperative
Gil Dotan, CEO of Guardian Optical Technologies, writes that today’s artificial intelligence technology is taking automotive safety far beyond the airbag.
From the day that the first automobiles began frightening the horses, safety has been a prime concern for drivers and everyone sharing the streets with the amazing invention. The indisputable need for safer cars has driven automotive innovation ever since. Hydraulic brakes and laminated windshields were among the earliest developments, while seat belts and airbags are the most familiar safety features of this era. However, the need to produce new and improved features never ends. Given the safety imperative, it’s only natural that industry leaders are now tapping the most advanced technology of our time—artificial intelligence (AI)—to reach the next level of automotive safety.
Watching the road
AI technology proliferates in the external safety features offered by advanced driver-assistance systems. ADASs employ outward-looking cameras, LiDAR, and sensors to watch the road, in effect becoming an extra set of eyes for the driver. The systems then use image recognition and computer vision to assess potential dangers, and alert drivers in real time.
Machine learning algorithms/deep neural networks classify the situations the cameras observe on the road; some highly advanced systems go even further by “understanding” what’s happening outside the car. Also classified are other vehicles, such as cars, trucks, motorcycles, pedestrians—even bicycles.
This capability is already a reality. It all started with the Israeli company Mobileye, which was the trailblazer in terms of ADAS and, more specifically, applying AI to solve these challenges. Now Tesla is developing its own solution applying advanced AI methodologies, and it is on its way to mass-market vehicles in coming years.
Watching drivers and passengers
Inside the car, inward-looking cameras keep an eye on drivers, monitoring their behavior and instantly alerting them to dangers like drowsiness and distraction. Internal cameras use deep neural networks to track and understand the driver’s behavior and recognize signs of danger. Is the driver’s attention focused on the road or not? Are their eyes closing, or is their gaze leaving the road? Are two hands on the wheel, only one—or no hands? That is a dangerous scenario, and something needs to be done.
Mobility services are emerging as another AI-assisted boon to drivers. These systems will steer drivers toward the best route to help them avoid accident scenes and traffic tie-ups. Most importantly, they’ll take other drivers into account and suggest routes that will be the best possible solutions for all drivers—routes that will minimize drive time for everyone and keep everyone safe on the road.
Understanding passenger status is also important—where they’re seated, their size, and body position or posture—to make sure that airbags operate more intelligently in case of an accident. Additionally, it helps seatbelt reminders and pre-tensioners operate more safely.
A true leap forward
Despite its unmistakable promise, artificial intelligence isn’t completely here yet. Many of the companies working with AI technology are still in the development stage, building systems that aren’t ready for mass-market products.
The time is right, however, to reach the early adopters. Consider the marketing of the latest Samsung Galaxy Fold smartphone, notable for both its folding screen and its $2000 price tag. It’s not likely to be widely adopted at that price, but its launch does provide an effective in-market testing bed—a safe place to identify any issues and iron them out before going to high-volume production. That strategy would serve AI systems well.
Whatever the timetable, we can trust that artificial intelligence will grow more common because it can be used to hedge risks in smart ways. The introduction of machine learning in 2012 instead of classical computer vision reduced errors significantly for the well-known ImageNet classification problem. Since then, classification errors have continued to decrease dramatically and reach better-than-human performance.
Disillusionment and optimism
A widely seen Gartner Group graph charts the 2018 “hype cycle” for connected vehicles, plotting the elevation of expectations over time. Technologies start with low expectations at the trigger point of innovation, rise sharply upward, reach the “peak of inflated expectations,” then sink down to the “trough of disillusionment.” Novel technologies begin to seem less magical when they hit bumps in the development path, but start rising up the “slope of enlightenment” once developers have grinded through every challenge and solved every issue. Expectations rise again as innovations actually become available to consumers.
It appears that disillusionment with AI has set in. Yet the technology is becoming more mature, and systems now in development are on their way to becoming commercial. And there’s at least one more reason for optimism: A 2017 Tractica market report predicts that AI automotive revenues — from hardware, software, and services—will hit the $14 billion mark by 2025, compared to 2016 revenue of $404 million.
The challenges currently posed by AI:
It’s data driven: Applying AI requires using massive data sets to “train” algorithms and give them “life experience.”
It’s only as good as you train it: Developers and researchers need to be extremely mindful of training algorithms with complete, well-rounded data. Algorithms trained only on U.S. data, for instance, are worthless when facing real-life scenarios in other countries around the globe.
It’s a black box: Since algorithms don’t reveal the patterns they’ve found, developers really don’t know how they work. Trying to improve what’s hidden in a virtual black box is deviously difficult.
Back in 1896, Alfred Sennett wrote about the dangers of horseless carriages replacing horse-drawn carriages: “We should not overlook the fact that the driving of a horseless carriage calls for a larger amount of attention, for he has not the advantage of the intelligence of the horse in shaping his path.” We might say that AI technology can stand in for the horses now. In any case, AI is taking automobile safety far beyond anything those pioneering drivers dreamed of more than a century ago.