AI researchers from MIT have reportedly developed a self-driving system that could allow autonomous cars to navigate roads without the need for a map.
Researchers from MIT’s Computer Science and Artificial Intelligence Laboratory designed a self-driving system that could allow driverless vehicles to navigate unpaved roads through basic GPS data and sensor technology.
Dubbed Maplite, the system will enable self-driving cars to drive on roads they’ve never been on before. The new system uses the combined technology of Google Maps GPS data for navigation along with LIDAR and IMU sensors for determining distances.
This self-driving system could potentially pave the way for better and safer self-driving experience on country roads.
“We were realizing how limited today’s self-driving cars are in terms of where they can actually drive,” Teddy Ort, a graduate student working on the MIT CSAIL, told Digital Trends.
“Companies like Google only test in big cities where they’ve labeled the exact positions of things like lanes and stop signs. These same cars wouldn’t have success on roads that are unpaved, unlit, or unreliably marked. This is a problem. While urban areas already have multiple forms of transportation for non-drivers, this isn’t true for rural areas. If you live outside the city and can’t drive, you don’t have many options.”
Today’s self-driving systems rely heavily on 3D maps which use sensors and vision algorithms for specific navigation purposes. Maplite, according to Ort, uses sensors instead for all parts of navigation.
The researchers tested the self-driving system on a Toyota Prius and equipped it with LIDAR, sensors, and the Maplite system. The self-driving car was able to navigate multiple rural roads in Devens, Massachusets by identifying the path 100 feet ahead.
Ort also explained that their system differs from other mapless methods that utilize machine learning to train the system on data of roads. Maplite works by attempting to develop models for different scenarios that autonomous vehicles could potentially encounter and inform it of its actions.
“At the end of the day, we want to be able to ask the car questions like ‘how many roads are merging at this intersection?” Ort went on to say. “By using modeling techniques, if the system doesn’t work or is involved in an accident, we can better understand why.”