Researchers at MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL) have created a framework that lets self-driving cars navigate on roads they’ve never been on before.
Current testing of self-driving cars is limited to specific areas because companies like Google have painstakingly labeled exact 3D positions of roads, curbs, and road signs ahead of the testing. Sensors and vision algorithms on self-driving cars are used only to detect and avoid dynamic objects, such as pedestrians or other cars. With millions of miles of roads around the world, it’s impossible to do this kind of 3D mapping in order for self-driving cars to operate safely.
The MIT CSAIL MapLite framework also uses sensors, but ones that observe the road conditions. Combined with simple GPS data, the framework can “reliably detect the road more than 100 feet in advance.” CSAIL graduate student Teddy Ort said in a press statement that the framework “shows the potential of self-driving cars being able to actually handle roads beyond the small number that tech companies have mapped.”
Ort is the lead author of a paper describing the project, co-written by MIT professor Daniela Rus and PhD graduate Liam Paull, now an assistant professor at the University of Montreal. The paper (download here) will be presented later this month at the International Conference on Robotics and Automation in Brisbane, Australia. The project was supported in part by the National Science Foundation and the Toyota Research Initiative, MIT CSAIL said.
To prove that the concept work, the MIT team and Toyota Research Institute fitted a Toyota Prius with a range of lidar and inertial measurement unit (IMU) sensors. The system was able to reliably detect an unpaved country road in Devens, Mass., and “see the road” 100 feet in front of the car.
The GPS in the car provides a rough estimate of the car’s location, and the system then sets a final destination and a “local navigation goal,” which needs to be within view of the car. Perception sensors then create a path to get to that point, using lidar to estimate the location of the road’s edge. The team said MapLite can make this calculation without physical road markings “by making basic assumptions about how the road will be relatively more flat than the surrounding areas.”
The team admits that MapLite is still limited – it’s not reliable enough for mountain roads, due to possible dramatic changes in elevations. The researchers said they hope to expand the variety of roads that the vehicle could handle, with the goal of reaching comparable performance levels as systems that are highly mapped.
“I imagine that the self-driving cars of the future will always make some use of 3D maps in urban areas,” said Ort in a statement. “But when called upon to take a trip off the beaten path, these vehicles will need to be as good as humans driving on unfamiliar roads they have never seen before. We hope our work is a step in that direction.”