MIT Team Develops Mapless Navigation for Autonomous Cars

Researchers at the Massachusetts Institute of Technology’s Computer Science and Artificial Intelligence Lab are developing software that would enable autonomous vehicles to operate on unmarked roads without requiring intensive 3D mapping of the area.

Most self-driving car systems use high-definition maps that detail the road geometry and surrounding area. This allows a vehicle’s position to be pinpointed within a lane, which is augmented by a network of onboard sensors to help a vehicle avoid obstacles and stay on course.
 

 MIT’s MapLite technology relies mainly on sensors (with some help from GPS), thus eliminating the time-consuming and expensive mapping process.

Such approaches haven’t been used in the past because they lacked the accuracy of 3D maps. But the MIT team claims MapLite can match the performance of a map-based system by using sophisticated modeling software, which categorizes a vehicle’s surroundings to determine how it should be controlled.

The researchers say their technology is in the early stages of development. They acknowledge it still needs improvement in some areas, such as the ability to navigate up a mountain road with dramatic changes in elevation. The team will present its research later this month at a robotics conference in Brisbane, Australia.