Behind the Tech that Navigates Self-Driving Cars on Unmapped Roads
May 15, 2018—Researchers at MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL) have developed a new technology for self-driving vehicles that allows the car to navigate on rural roads.
The researchers partnered with the Toyota Research Institute and tested the technology on country roads in Devens, Mass. The test drive was done with a Toyota Prius.
MapLite represents a first step towards self-driving cars without using 3-D maps. The software enables cars to navigate on roads they’ve never been on before by using GPS and sensors.
Companies like Google have only tested their fleets in major cities, where the city has labeled exact positions of things like curbs, lanes and stop signs, said Teddy Ort, lead author and MIT CSAIL graduate student.
MapLite can detect the road 100 feet in front of the vehicle. The GPS data is similar to the data found on Google Maps and a series of sensors looks at the road conditions.
Ort said one of the sacrifices in giving up 3-D maps is losing the opportunity to perform through map-based testing , which enables the car to drive on the entire map system and verify that the system is working well on the grid.
While 3-D maps might never be completely eliminated in self-driving cars in the metro areas, Ort said the system could prove useful to determine where 3-D map systems need to be updated. It could also act as a critical layer of safety in the event that a vehicle is unable to localize itself on a map.
The team is still researching how this technology might be verified on open roads that haven’t been seen before, he said. There is no immediate timeline for deploying the technology into commercial vehicles.