MIT Tackles Autonomous Driving Challenges With New GPS System | Tech News
Self-driving cars are quickly moving from the realm of science fiction to science fact, but if Uber’s recent fatal accident is any indication, the technology isn’t quite ready yet.
One of the biggest things holding these cars back is mapping — most of the testing programs in place right now use hand-labeled maps that upload into the car’s computer to tell it where every single landmark, lane and streetlight are in relation to the car’s position.
Researchers at MIT might change this soon with a new program that utilizes something you probably already use every day — GPS.
How is this different from the programming currently used in autonomous cars and how will this improve the functionality of self-driving cars in the future?
Hand-Drawn Maps Are Lacking
There’s a reason that most of the autonomous car tests take place in big cities. Sure, they’re populated but they’re also well-lit and well mapped, making it easier for the self-driving cars to navigate even the most crowded street.
Even with these facts working in their favor, the programming used in Google and Uber test cars is reliant on an almost archaic practice — hand-drawn maps.
For these vehicles to successfully navigate a city street or neighborhood, programmers have to hand-label everything on the map that could potentially affect the vehicle’s ability to drive. This includes buildings, parking lots, street lamps, telephone poles and the dividing lane lines on each street in the test area.
This takes a lot of time and effort — and prevents the car from exploring anywhere outside of the test area. A self-driving car cannot be counted on to safely navigate outside of its programmed test area. That leaves a lot of the country unable to support self-driving cars.
Don’t count on your self-driving car to drive you to your remote ski retreat or vacation home, unless you want to take the wheel yourself.
MIT’s answer to these hand-drawn maps is to eliminate the need for them altogether by using a system they’ve dubbed MapLite.
This system uses standard GPS together with LIDAR and IMU sensors on the car itself to create a real-time mapping service that rivals the hand-drawn maps in accuracy with one obvious benefit — it can be used anywhere that the car has a GPS signal, which these days is nearly everywhere in the world.
This map-less type of programming hasn’t been used in the past because of concerns that the GPS maps aren’t as accurate as the hand-drawn ones, but GPS technology has advanced to the point where you can easily find a destination nearly anywhere — and accurate step by step instructions on how to reach it.
MapLite takes this to the next level. Instead of relying on maps for navigation and only using sensors for avoiding obstacles like current self-driving cars do, it flips the algorithms on their heads, using sensors for everything including navigation.
The trick here is to give self-driving cars the ability to drive safely on unfamiliar roads, instead of defaulting to driver control the moment the car leaves a mapped area.
The Future of Self-Driving Cars
What does this new programming mean for the future of self-driving cars?
Not much, as of yet. It’s not perfect, and it still can’t be used in hilly or mountainous terrain because it cannot properly read large changes in elevation. Once it is ready to go, though, it could change the way we look at self-driving cars.
The entire autonomous car industry is still moving forward, in spite of the recent hiccups. Uber’s self-driving test cars are back on the road in San Francisco, though they haven’t yet gotten the green light to continue their testing in Tempe, Arizona where the fatal accident occurred.
If this new program turns out successfully, it could turn self-driving cars into a viable mode of transportation across the country, regardless of the local population or whether or not the area has been hand-mapped.
Turning GPS into something that can allow self-driving cars to navigate will be a game changer for the industry.
Written by Kayla Matthews, Productivity Bytes.