News Posts

MIT Invented A Tool That Allows Driverless Cars To Navigate Rural Roads Without A Map

Share on FacebookTweet about this on Twitter

Google has spent the last 13 years mapping every corner and crevice of the world.

Car makers haven’t got nearly as long a lead time to perfect the maps that will keep driverless cars from sliding into ditches or hitting misplaced medians if they want to meet their optimistic deadlines.

This is especially true in rural areas where mapping efforts tend to come last due to smaller demand versus cities.

It’s also a more complicated task, due to a lack of infrastructure (i.e. curbs, barriers, and signage) that computers would normally use as reference points.

That’s why a student at MIT’s Computer Science and Artificial Intelligence Lab (CSAIL) is developing new technology, called MapLite, that eliminates the need for maps in self-driving car technology altogether.




This could more easily enable a fleet-sharing model that connects carless rural residents and would facilitate intercity trips that run through rural areas.

In a paper posted online on May 7 by CSAIL and project partner Toyota, 30-year-old PhD candidate Teddy Ort—along with co-authors Liam Paull and Daniela Rus—detail how using LIDAR and GPS together can enable self-driving cars to navigate on rural roads without having a detailed map to guide them.

The team was able to drive down a number of unpaved roads in rural Massachusetts and reliably scan the road for curves and obstacles up to 100 feet ahead, according to the paper.

Our method makes no assumptions about road markings and only minimal assumptions about road geometry,” wrote the authors in their paper.

Once the technology is perfected, proponents argue that autonomous cars could also help improve safety on rural roads by reducing the number of impaired and drowsy drivers, eliminating speeding, and detecting and reacting to obstacles even on pitch-black roads.

Ort’s algorithm isn’t commercializable yet; he hasn’t yet tested his algorithm in a wide variety of road conditions and elevations.

Still, if only from an economic perspective it’s clear repeatedly visually capturing millions of miles of roads to train cars how to drive autonomously isn’t going to be winning mapping technology for AVs; it’s just not feasible for most organizations.

Whether it’s Ort’s work, or end-to-end machine learning, or some other technology that wins the navigation race for autonomous vehicles, it’s important to remember that maps are first and foremost a visual tool to aid sighted people in figuring out where to go.

Like humans, a car may not necessarily need to “see” to get to where it’s going—it just needs to sharpen its other senses.

Please like, share and tweet this article.

Pass it on: Popular Science

Leave a Reply

Your email address will not be published. Required fields are marked *