Self-driving cars rely on knowing their environment by reading maps. It has been thought that these maps need to be super detailed and that generating and updating these complex maps would be one of the major hurdles in developing a self-driving future.
But Teddy Ort, a graduate student in robotics at the Massachusetts Institute of Technology’s Computer Science and Artificial Intelligence Laboratory, has a new way of looking at maps that would radically reduce their complexity and size. “Maps for even a small city tend to be gigabytes; to scale to the whole country, you’d need incredibly high-speed connections and massive servers,” says Ort.
Simplified global map could fit on a flash drive
Ort says the maps we feed to self-driving cars don’t need to be as complex as previously thought.
“But for our approach, a global map could fit on a flash drive.”
For instance, self-driving cars may not need all the information for the environment like curb heights and roadside features. Instead, a self-driving car’s GPS system only needs a very minimalist map and its sensors which it would use to navigate ahead to a 'waypoint'.
So instead of understanding everything around it, the system would navigate along the road in a general direction. Much like the way human drivers navigate a new or unfamiliar road.
MIT has tested the technology, called MapLite, on quiet country roads in its home state of Massachusetts. The system will need further extensive testing in complex road environments that include multiple lanes, U-turns, and other sophisticated rules of the road.
“But this can work in general environments,” says Prof. Daniela Rus, who heads the laboratory. “It’s a general approach to building drive-by-wire autonomous systems that do not depend on dense feature maps.”
The research is being supported by the Toyota Research Institute and the lab's test car is a Toyota Prius. There are huge amounts of research going on around the world into self-driving and autonomous cars.
Each self-driving lab must tackle navigation question
Each lab is taking a different approach to mapping and navigation. Some labs have focussed on generating detailed maps that use ground-based relays to augment GPS data. Others have tried to steer away from a reliance on mapping data and are using machine learning to apply lessons gained from experience on one road to the problems posed by another road.
MIT say they are taking a unique approach to the problem by not relying solely on machine learning based on neural networks. “We do use machine learning to find what road it is,” Ort says. “But our path finding is all from a model-based approach. If it doesn’t work as we thought, we can go in and fix it.”
The team admits that although they are excited by the progress so far, they have a lot of work to do. “The main conceptual drawback is verification,” Ort says. “A detailed map means someone’s driven over it, done a fair amount of testing and shown that it’s safe—it hasn’t changed. But if you’ve never driven over it before, that’s not so. We’re working on how to verify the safety of driving on a road we’ve never seen before.”