LGPR gives self-driving cars a new set of eyes

localizing ground-penetrating radar

MIT was able to use LGPR to allow self-driving cars to “see underground”; to give them the ability to map and position themselves correctly even in poor visibility conditions, like snow, fog or heavy rain.

Generally, self-driving cars today use a combination of cameras, LiDAR and radars that, when analyzed by artificial intelligence, allow them to see their surroundings and avoid obstacles. But what about conditions that obscure cameras – essentially blinding the AI? What would stop the car from finding itself on the wrong side of the road or accidentally switching lanes?

Enter LGPR, or Localizing Ground Penetration Radar. This technology uses radar to visualize the underground area beneath the road’s surface and map the soil, rock patterns, pipes, roots, minerals, etc to a depth of up to ten meters.

Simulation demonstrating how LGPR allows self-driving cars to map the road's subsurface
Photo: MIT Lincoln Laboratory

LGPR allows self-driving cars to position themselves correctly on the road and stay in their lane even when lane markers are completely invisible, which is often the case when it snows,

MIT CSAIL PhD student Teddy Ort spoke to Engadget about the tech: “If you or I grabbed a shovel and dug it into the ground, all we’re going to see is a bunch of dirt. But LGPR can quantify the specific elements there and compare that to the map it’s already created, so that it knows exactly where it is, without needing cameras or lasers.”

As Ort said, this technology only works if the vehicle’s AI system already has a pre-recorded map of the subsurface; meaning LGPR doesn’t work on roads it hasn’t already scanned. This is great for your daily commute to work or school, provided you were able to do at least one trip before the snowfall season came around, but don’t expect to rely on LGPR on a road trip.

Is it scalable?

That is of course unless LGPR grows popular enough to warrant the mapping of the world’s road subsurface. This may seem ridiculous, but it’s easy to forget that Google Maps only launched in 2005. It took them almost a decade to map enough of the world to become reliable. It’s also likely that Google would adopt the tech and use their Maps infrastructure to grow and distribute subsurface maps.

Ort’s team has been working on this technology since 2016. On February 24 2020, MITCSAIL’s YouTube channel published a new video showcasing their technology in a self-driving Toyota Prius. The video also linked to their paper.

Ort’s team plans to publish a paper on their project in the Robotics and Automation Letters journal later this month. They also need to continue refining the hardware. It is still somewhat unwieldy at six feet wide and not very viable to latch onto your vehicle.

The intent is eventually to have it small enough to build into the car itself. LGPR is one more tool that self-driving cars can use to detect their environment and navigate within it.