Self-driving cars are useless without specialized maps—this invention could free them
New software could make robotic vehicles much better at handling long drives into countryside.
Some background: Autonomous cars currently undergoing testing drive only on streets with meticulously labeled maps that take up huge amounts of storage space. “Maps for even a small city tend to be gigabytes; to scale to the whole country, you’d need incredibly high-speed connections and massive servers,” says Teddy Ort, a grad student at MIT’s computer science and artificial intelligence lab (CSAIL).
The solution: A new, more adaptable approach by the CSAIL team, called MapLite, uses simple GPS data to plot a path to the vehicle’s destination, and lidar sensors to navigate along the way. The sensors estimate curb distances and the road conditions more than 100 feet ahead of the vehicle.
Why it matters: Labor-intensive mapping systems are not a scalable solution. Autonomous cars will need to function in areas with poor-quality or unlit roads on which they don’t have detailed data. While MapLite has only been tested in rural Massachusetts so far, it could make robotic cars more capable drivers.