Autonomous vehicles have been getting lots of attention lately. Loup Ventures predicts that by 2040, we’ll see a full-scale transformation of the automotive industry, focusing on autonomous and/or electric cars with a higher focus on ridesharing services, rather than personal ownership.
However, autonomous vehicle technology isn’t quite there yet. Despite advancements from companies like Waymo in the field of fully autonomous vehicles, there’s yet to be a commercially available, totally self-driving car. Top-named luxury cars are starting to roll out with autonomous features, like lane-keeping and parking assistance, but there’s no fully autonomous solution yet.
The Testing Problem
Regulations are part of the problem, but a bigger issue is with testing. Autonomous vehicle engineers are responsible for ensuring their vehicles drive safer than their human counterparts, which means driving millions—or even billions—of miles to test for every contingency. The real hazards aren’t everyday occurrences, but instead are random occurrences and rare circumstances that will truly put these algorithms to the test.
The trouble is, companies are reluctant to put autonomous vehicles on the road while in test mode; after all, an untested design can be a public safety hazard, and beyond that, a failed test could turn into a PR nightmare. Accordingly, many engineers have turned to simulated road experiences to test their self-driving algorithms, but simulated environments are hard to come by.
Enter Cognata
This is where Cognata comes in. Cognata is an Israel-based startup launching VR software that tests and trains autonomous vehicle programs in a simulated driving environment. Collectively, Emerge, Maniv Mobility, and Airbus Ventures have invested $5 million into the startup. Relying on AI, deep learning, and computer vision, Cognata specializes in producing several different road conditions that mimic the inputs a self-driving car would receive if it were actually on the road.
Moreover, it uses 3D simulated landscapes of real cities, such as San Francisco, modeling everything from buildings, trees, and even other vehicles as realistically as possible. The company relies on historical traffic data to mimic real-life conditions, and can therefore test cars in each of several different urban environments.
Cognata’s approach solves several problems related to testing:
- You might feel confident about your driving abilities after a few years on the road, but automated technology requires much more thorough testing and practice. In fact, Rand Corporation estimates that autonomous cars should be tested for more than 11 billion miles before they can demonstrably prove themselves to be more reliable for humans. Assuming you have 100 cars running 24 hours a day, that would take 500 years of testing. Simulated road software could cut that time dramatically, serving hundreds of thousands of simulated vehicles at once.
- Testing an autonomous vehicle in real-life conditions is inarguably the most “realistic” way to test them—but it’s also the riskiest. One wrong decision could hurt, or even kill another driver. In a simulated environment, there’s no such safety risk. Cars can be tested in complete safety for millions of miles before they even hit the pavement.
- Condition diversity. Cognata also provides more conditional diversity than even real streets can offer. They provide infrastructural layouts for multiple different cities, and can simulate different conditions related to traffic and weather as well. This gives autonomous vehicles a more thorough exploration, and can help pinpoint weaknesses in the technology early.
- Sensor inputs. There are several components of self-driving cars, each of which send and receive information at a constant rate while driving. Cognata’s software emulates data that responds to each of these sensors, such as image sensors, radar, LIDAR, and GPS data.
Despite the advancements they’ve seen over the past several years, self-driving cars are still a technology in its infancy. Thanks to companies like Cognata, engineers may find it easier to reach those 11 billion miles necessary to truly test the safety of their self-driving vehicles—and in much safer, more thorough conditions. Though most of the public’s attention remains on the engineers and manufacturers of the cars themselves, the background players—producing the hardware, infrastructure, and testing environments necessary to make self-driving cars possible—deserve just as much credit.