Autonomous Driving Surpassing Another Roadblock
MIT research has AI emulating driver behavior to improve technology
Despite the progress made in driverless vehicular technology of late, the artificial intelligence level still has a long way to go before an autonomous car can perform like an actual human being is behind the wheel.
For openers, safety remains a going concern, with government bodies like the National Highway Safety Traffic Administration declaring earlier this year that casualties could mount on American roads if autonomous vehicular technology is rushed to hastily to market. Then there’s the question of how these cars can still navigate accurately, especially on roadways where distinct markers like road signs and lane stripes are absent.
Researchers at MIT claim they’ve come closer to solving those problems with a notion to add elements associated with human reasoning to the code that drives artificial intelligence. According to a story that appeared recently in Design News, the new technology being developed enables a computer on board an autonomous vehicle to learn how humans navigate roadways by studying how a person maneuvers a steering wheel and operate the rest of the car. By translating those behaviors into code built into AI, the car should be able to find its own direction without any involvement from its passengers.
“Our objective is to achieve autonomous navigation that is robust for driving in new environments,” said Daniela Rus, a professor specializing in electrical engineering and computer science.
Pending the successful implementation of this feature down the road, car rental firms stand to benefit from such an amenity. A client could easily use an app to not only book and pay for a car, but also wait for the rented vehicle to drive over to the customer unassisted.
So far, programming autonomous vehicles require a painstaking and detailed scanning and analysis of roadways, which is not only time consuming, but incredibly complex as well. The MIT brain trust sidesteps that by further developing a navigation system the institution already invented. In this case, the system takes a point-A-to-point-B approach, with the latter being a destination never before used by the autonomous vehicle. While digital maps are still necessary to determine how to reach a destination, the team put more focus on adding a wide range of probabilities of driver behavior that’s accessible at any moment by the system.
“With our system, you don’t need to train on every road beforehand,” said MIT graduate student Alexander Amini, a team member on the project. “You can download a new map for the car to navigate through roads it has never seen before.”