New Tech Takes Self-Driving Cars Off Road
By Mike Howie
Self-driving cars are rapidly becoming more capable and more common, thanks to high-profile efforts from the likes of Tesla, Uber, and other companies. But they’re far from perfect: many of them struggle with faded lane lines, especially in rainy conditions, and can’t navigate in areas that aren’t thoroughly mapped. And after a self-driving Uber vehicle tragically struck and killed a pedestrian in Arizona in March, it’s clear that obstacle detection and avoidance technology must significantly improve as well before these cars are ready for wide-spread adoption.
A team from MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL) has been working to create a framework that would allow self-driving cars to drive on unfamiliar and even unpaved roads without the advanced 3D maps that most driverless cars depend on. It’s called MapLite, and it uses GPS data similar to what’s widely available on Google Maps paired with sensors that evaluate road conditions. Together, they can reliably detect the road more than 100 feet in advance.
Navigating with Digital Eyes
While most self-driving cars rely heavily on maps to navigate and only use sensors to avoid cars, pedestrians, and other objects, MapLite relies on sensors for all aspects of navigation and only uses GPS to estimate the location of the car relative to its destination. The sensors generate a path, and a LIDAR system estimates where the edges of the road are.
The team tested their car on unpaved roads in Devens, Massachusetts, but they still have more work to do. The system isn’t reliable enough to tackle mountain roads yet because it doesn’t account for dramatic changes in elevation. Their next challenge is to utilize more diverse test routes, because their ultimate goal is to prove that MapLite-equipped cars can equal or improve upon the navigation performance of mapped systems but across a greater variety of roads and road conditions than is currently possible.
Because MapLite is significantly less complex than other systems and doesn’t require advanced maps, it could make self-driving cars more viable in smaller cities and even unmapped rural areas.
Discussion Questions
- What other technology might make self-driving cars safer? How would it work?
- How could self-driving cars be useful off road?
Vocabulary
- GPS
- Lidar