Published on THECONNECTEDCAR.COM
Link to publication
What is the worst confluence of driving conditions you can possibly imagine?
It’s probably nighttime in a densely populated city. There would be a vision-obscuring blizzard making the roads slick and unreliable. Traffic would be thick and moving irregularly. Worse, the drivers around you would have seemingly little regard for your safety, or even their own. Even the road signs would be difficult to see.
Welcome to life for drivers in Moscow.
Poorly laid out roads, bad weather and not-so-cautious driving practices make for dangerous automobile treks in the Russian capital. Moscow streets are home to 20 serious car accidents per day, and the road fatality rate in Russia is double that of the United States.
Not surprisingly, that makes it a difficult landscape for autonomous vehicles to navigate. But the conditions in Russia are not all that different from other countries in the world, which is why autonomous developers believe that, with apologies to New York, “if they can make it there, they can make it anywhere.”
At a three-day hackathon in Moscow, engineering students from around the globe and corporate sponsors like Nvidia and Uber gathered to take a crack at developing autonomous systems for Moscow’s roads.
“The event had another purpose: to advance a credo that when it comes to autonomous cars, tougher conditions produce smarter technology,” writes Gaus. “Lidar — the expensive, light-pulsing sensors relied upon by current autonomous car models — is worthless in snow … Instead, cars should be trained to operate using high-definition cameras, low-cost radars and powerful AI that mimics the human brain.”
The idea that Lidar is “worthless in snow,” advanced by Olga Uskova of Russian AV software developer Cognitive Technologies may be an extreme position — nearly all driverless car manufacturers incorporate Lidar sensors in some form.
Nevertheless, Lidar does not work as effectively in the snow, and developers have relied more heavily on other hardware to navigate in adverse conditions. An autonomous vehicle in Finland primarily used radar sensors to complete a journey through a wintry mix.
As for the hackathon engineers in Moscow, cobbling together an autonomous driving system over the course of three sleepless days proved difficult. The top team only managed to achieve 40% accuracy in identifying road signs. The expected culprits were to blame for the difficulties: Snow-covered road signs were difficult for systems to detect, and non-Russian speakers had an even more challenging time differentiating between similar looking road signs.
While success proved elusive at the hackathon, in some ways, that was beside the point.
The fact that driverless car developers are moving beyond building cars that can work in the idealistic sunny climates of Arizona and California and shifting to the more challenging task of creating vehicles that can work in more realistic scenarios is an important sign of progress.
Many expect that self-driving cars will eventually operate 24/7, but they will only be able to do that if they can handle the weather and unique road conditions that present themselves at every moment of the day all over the world.