Follow
Subscribe

Autonomous Car Startup Scale API Believes Sharing Car Data Will Make Self-Driving Cars Better

Home > News > Content

【Summary】Scale API, which trains artificial intelligence systems through the examination and categorization of visual data, claims that companies will be able to develop better autonomous vehicles if they share data.

Original Vineeth Joel Patel    Aug 10, 2018 10:15 AM PT
Autonomous Car Startup Scale API Believes Sharing Car Data Will Make Self-Driving Cars Better

Coming out with an autonomous vehicle, as automakers have learned, isn't easy. There are numerous things that go into giving a self-driving vehicle the ability to drive on its own. At the moment, companies are looking into developing self-driving vehicles on their own. Some have opted to create small partnerships with startups for necessary software, hardware, or equipment, but the majority of large companies and automakers have opted to forgo getting help from other major companies. 


For the most part, autonomous vehicles have to deal with erratic drivers and the occasional pedestrian that chooses to forgo using a crosswalk on a regular basis. The situation is worse in cities, as there are more things going on, which include more cars, more pedestrians, and differing road conditions. In urban locations, driverless vehicles won't have to deal with congested roads, but animals, like deer. 


While these are things that companies need to account for when they develop autonomous vehicles, there's a completely different side that driverless cars will need to be able to tackle. Days like July 4 are incredibly difficult for autonomous cars. Not only are there more cars on the road during the holiday, but pedestrians behave in a much more erratic fashion, fireworks illuminate the road in odd ways, and numerous items, including flags, are being waved around. That, as Wired points out, is the making of a perfect storm for driverless cars. 


Scale API Labeling.jpeg


How Scale API Is Training Autonomous Vehicles


Technology companies and automakers that are having trouble being able to have their autonomous vehicles deal with the chaos of holidays like the Fourth of July, could benefit from a helping hand from Scale API. The startup, as Wired points out in a lengthy piece, has developed automated systems that takes data from what self-driving cars see and examines it to label necessary items. The process of creating labels helps the software inside an autonomous vehicle learn to recognize a specific situation to be better prepped for the future. 


Obviously, labeling is an important part of what Scale API does. As Wired points out, wrongly labeling a human as a vehicle could confuse an autonomous vehicle's software to the point where it's constantly making mistakes and could even lead to an incident. The flip side of that, is when Scale API does get it right, a driverless system safely and effectively learns how to deal with both tricky situation and routine ones. As the outlet claims, Scale has made labeling a necessary and integral part of developing an autonomous vehicle. 


The unfortunate thing is that labels aren't shared with one another. Wired claims that Scale's customer base, which includes Cruise, Nuro, Lyft, Zoox, Nutonomy, Starsky Robotics, and Embark all send data to the start up to label. That data, though, doesn't get shared with one another, which is a shame. If it's one thing we've seen, it's that autonomous cars can always use more real-word testing and more data. Information from precarious situations, like the ones that take place on July 4, would be even more helpful for autonomous cars. 


As Wired points out, it's understandable that companies would want to keep their data private, even if it is data that other companies will probably gather on their own. Companies spend a lot of money gathering data, testing autonomous vehicles, and developing high-tech systems. 


"I don't know how you get competitors to share their most valuable information," Oscar Beijbom, the head of machine learning at Nutonomy, told Wired. "In a way, these corner cases are very previous." 


While it's easy to understand why companies keep things to themselves, Alexandr Wang, Scale's 21-year-old founder and CEO, thinks it's a bad decision. "Right now each company is so in its own lane and secretive," Wang stated. "In reality, these edge cases, these are things that should be probably be shared or standardized across the industry at some point." 


Scale API Sensors.jpeg


What Does Scale API Actually Do?


Scale doesn't just label everyday things, but everything a company provides it. Autonomous vehicles, as we've outlined multiple times before, have numerous systems that include cameras, LiDAR, radar, and sensors that soak up their environment. So data that Scale receives includes everything from regular traffic and cyclists to freak situations – Wired claims that the company had to label an incident that involved a cargo truck that lost its logs on the road while driving. 


"Scale is basically providing the ground truth for our perception systems,"Anantha Kancherla, the individual in charge of developing autonomous software at Lyft told the outlet. "It's a very, very critical piece for us to develop." 


Scale's employees in Asia and Europe examine the data, primarily the images that the vehicles' camera and 3-D LiDAR systems generate and draw boxes around important items in the picture. Important items in a picture include pedestrians, cyclists, and cars. The team also identifies items that aren't important, pointing things out that could be confused and labeling those, too. Scale has a system that can label items on its own, but employees check those to make sure that they're accurate. 


The odd part of Scale's operation is that it goes through the process of labeling data for different clients at different times, since there's no sharing between its clients. That means Scale probably goes through the same data numerous times – even if it's the same case. Sharing would not only reduce the amount of data Scale goes through, but it would also help everyone developing autonomous technology. 


"It's a little bit ridiculous that the same companies do almost the exact same annotation work," Beijbom told Wired. "It does feel very wasteful and suboptimal." 


There's also the issue of not accounting for the unknown. Scale and one of its companies may have labeled and documented a wacky situation that helps one company's vehicles prepare for the situation. Since another brand's autonomous vehicle didn't interact with the same incident, their self-driving cars won't be prepared for the specific situation.


"If you're worried about your system missing edge cases, the ‘unknown unknowns,' then the more examples you have, and the more conditions the car encounters, the more opportunities you have to train the system to do a better job," Michael Wagner, CEO of Edge Case Research, told the outlet.


If autonomous companies were to share the load of coming out with autonomous vehicles as a group instead of attempting to solve the dilemma individually, it could result in self-driving cars coming out on the road quicker and better technology. 

Prev                  Next
Writer's other posts
Comments:
    Related Content