Over the past few years, autonomous vehicle technology and self-driving systems have reached new levels of sophistication and an experimental scanning system could lead to improved safety levels in poor weather.
Self-driving capabilities are measured on a scale of levels 0 - 5; 0 being no automation and level 5 considered to be a true, driverless vehicle that does not need a human to operate successfully and safely.
There are no commercially viable level 5 systems, yet, but companies worldwide including Tesla, Google's Waymo, Uber, and General Motors are developing a range of self-driving or driver "assistance" technologies generally fitting in the level 2 - 4 categories.
There are a number of roadblocks that vendors are faced with; object detection including hazards such as pedestrians; map accuracy, and how vehicles should operate in poor weather or road conditions.
Self-driving systems often use LIDAR sensors and cameras to build real-time maps of a vehicle's surroundings through data gathered from lasers, sensors, and camera feeds. However, when the weather is poor -- such as in heavy rain or snow -- these information streams can be disrupted and thereby impact the maps created and the safety level of a self-driving vehicle.
Lane markings may be missed; traffic signs might be blanketed by snow, and rain could cause camera malfunctions. While a human driver may already be familiar with the road they are traveling on and change their behavior in light of poor weather conditions, autonomous vehicles have to rely on real-time maps to operate.
It only takes a small margin of error caused by rain, snow, or fog to spell disaster and that goes for either a self-driving system or human driver. However, a solution may have been found in a place you would not expect -- under the road.
On Monday, academics from the Massachusetts Institute of Technology (MIT)'s Computer Science and Artificial Intelligence Lab (CSAIL) published research on how to make driverless cars safer, and their method bypasses cameras and LIDAR completely.
The system harnesses existing technologies known as "ground-penetrating radar" (GPR) to send electromagnetic pulses underground. The radar measures the road's combination of soil, roots, and rock, creating an alternative 'map' of the ground's composition.
The map, made up of underground fingerprints, can help orient a car, no matter the weather conditions.
The team used a GPR system developed at the MIT Lincoln Laboratory called localizing ground-penetrating radar (LGPR). During tests taking place over the course of six months, CSAIL found that the margin of error was roughly an inch off in snow, in comparison to clear weather.
The LGPR actually had more trouble in heavy rain, accounting for a rough margin of error of 5.5 inches.
The team says this is likely due to ground composition changes when water soaks the road. However, the team did not have to take the wheel once during the test period.
"If you or I grabbed a shovel and dug it into the ground, all we're going to see is a bunch of dirt," says CSAIL Ph.D. student Teddy Ort, lead author on a new paper describing the project. "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."
CSAIL's instrumentation was only tested on a closed country road and at slow speeds and so is not roadworthy at present. The hardware, too, is six feet wide and would need a serious overhaul before it would be small enough to integrate with a standard vehicle. However, the project highlights an important point: thinking outside of the box when it comes to mapping systems can lead to technologies able to augment self-driving car safety.