You might think that with all their cameras and other sensors, self-driving cars can handle any road in any location. That actually isn't the case, although it soon may be, thanks to MIT's new MapLite system.
In their current state of development, self-driving cars can really only be trusted on roads in major cities. There, the cars are able to utilize pre-made 3D maps on which the exact positions of stationary items such as lanes, curbs and stop signs have already been manually labelled. The vehicles' sensors mainly only come into use for detecting and avoiding dynamic obstacles, such as pedestrians and other traffic.
In the case of little-used rural roads, the incentive just isn't there for putting in all the work to label them, so the cars can't drive on them autonomously.
That's where MapLite comes in. It uses GPS to let the car know what road it's on, providing a rough estimate of its current location, and guiding it toward its final destination. The car's LiDAR (Light Detection and Ranging) sensors, however, are used to make sense of the road immediately around it.
The system continuously creates 3D point cloud models to establish new "local navigation goals" on the road that's visible ahead, then plots the best path to reach each of those goals. Additionally, the LiDAR is used to determine where the edges of the road are, based on the assumption that the surface of the road will be smoother than the surrounding terrain.
MapLite additionally incorporates computer models that advise the car on how to handle some basic situations, such as when it's approaching intersections. The technology still can't account for all possible variables, though, including the dramatic changes in elevation that occur on mountain roads. So far, it's been successfully tested using a specially-outfitted Toyota Prius, on a number of unpaved country roads around Devens, Massachusetts.
"I imagine that the self-driving cars of the future will always make some use of 3D maps in urban areas," says graduate student Teddy Ort, lead author of a paper on the research. "But when called upon to take a trip off the beaten path, these vehicles will need to be as good as humans at driving on unfamiliar roads they have never seen before. We hope our work is a step in that direction."
There's more information in the following video.
Source: MIT Computer Science and Artificial Intelligence Laboratory
