, , ,

Beyond the Mapping Frenzy: Driverless Cars Will Depend on Highly Accurate GPS

By Paula Reinman

By Ken Pesyna
Co-authored by Paula Reinman

The current money in the driverless car market is on mapping. Maps will play a key role in helping automated vehicles determine their exact location by comparing their sensor outputs (what they see) to this mapping data (what they should expect to see). With Uber spending $500M to create its own maps, Nokia’s HERE maps selling to Audi, BMW and Mercedes for $3B and Google spending undisclosed amounts to map the world, it seems like those with the best maps will win. When you dig below the mapping frenzy, though, there’s another piece of technology that has to go right for the driverless car market to happen. That’s centimeter-accurate GPS.

Pushing the Boundaries on Accuracy and Economics

Today’s autonomous and semi-autonomous vehicles stay in their lanes and change lanes by relying on a suite of sensors consisting of cameras, lasers, and radar. These sensors determine what lane the car is in and are typically sensitive enough to determine where in the lane it is. But these sensors also require visual features, especially lane markings, to operate. So when the sun is setting and blinding their view, or when it’s snowing and everything looks the same, today’s sensors often break down and are no longer able to determine their lane whereabouts. Current sensors enable a vehicle to accurately navigate without human intervention about 95% of the time in pre-mapped areas. Although this may seem extraordinary, it is simply not good enough for driverless cars. That’s where centimeter-accurate GPS comes in. Through its ability to determine where the vehicle is to within 10 centimeters (about four inches), despite poor weather or the absence of lane markings, this technology is a must to achieve the safety levels needed for so-called Level 4 automation, where the car self-drives all the time and human intervention is strictly optional.

In addition, centimeter-accurate GPS will also underpin the economics needed to scale the mapping that supports these vehicles. Today’s mapping solutions are not only expensive to acquire – they are incredibly costly to implement. Fleets of vehicles equipped with specialized sensors are driving around cities to map them. This is unsustainable since roadways are constantly changing and companies like Google, Tesla and Uber would need to re-map cities every few months to keep their maps up-to-date. Because of this, mapping will eventually become crowd-sourced: the same sensors that our future automated vehicles will use to navigate by comparing what they see to these pre-made maps will simultaneously be used to update the maps. However, stitching together maps from million of vehicles will quickly become overwhelming unless algorithms know exactly where to put the data within the larger scale map and how much to trust data from each source. With centimeter-accurate GPS, each vehicle can attach its precise position and precise orientation at the time the data was taken to its mapping data. This will make map curation significantly easier than if these data were instead only stamped with meter-level accuracy that comes from standard GPS. This will allow map curation algorithms to distinguish good data from bad data: cars with better accuracy will have their data trusted more than vehicles with less accuracy.

From the Farm to the Car

Centimeter-accurate GPS is already in use today for applications like farming and surveying, but it requires expensive, purpose-built equipment. A couple of years ago, University of Texas professor Todd Humphreys and three graduate students discovered how to make this technology run almost entirely in software and on smartphones, thereby making it available to the consumer market. Ken Pesyna was honored as a Marconi Society Young Scholar for this work and he continues to be at the forefront of this technology with his company, Radiosense. Radiosense is bringing this highly accurate GPS capability to the suite of sensors that will become common in driverless vehicles. Through its software-based solution, Radiosense is able to provide a low-cost system capable of running on a vehicle’s existing computing platform.

Getting to Centimeter Accuracy

However, as with all new technologies, there are a couple of key challenges to be solved for centimeter-accurate GPS to be brought to the mass market.

In order to know where it is, a vehicle must connect to and difference its GPS measurements from those of a known reference station. To quickly get centimeter-level accuracy, a network of these reference stations, spaced by 20-to-50 kilometers, will be required. Radiosense has partnered with The University of Texas to create a low-cost, first-of-its-kind dense reference network in Austin. About 12 stations have been deployed, at $1000 each, to cover the entire city. The build-out of these networks will likely start in metro areas, such as Austin, and expand over time.

Additionally, centimeter-accurate GPS coverage in downtown environments is more difficult because the whole sky is frequently not visible. In order to provide centimeter-accuracy when the sky is partially blocked, companies like Radiosense are testing combinations of sensors, such as vision sensors that track nearby building facades as the vehicle moves, to make up for the lack of GPS signals. These sensors help provide information about how far the vehicle has traveled during periods in which there are not enough signals to compute a standalone GPS position. This is also where highly accurate maps come into play. It is very easy for a vehicle to recognize pre-mapped points of interest in feature-rich urban areas and use these to triangulate its position.

Despite the hype, it will be a slow transition from what we have now to fully-driverless cars. While today’s vehicles can warn us if we are drifting out of our lane or getting too close to another vehicle, the next generation will pull the car back into line if the driver does not respond to an audible warning. The jury is still out on whether we’ll get to mass use of driverless cars through incremental improvements, like those that Tesla is bringing to market, or through a huge technology leap with a whole new type of vehicle, as Google is developing. Either way, the next generation of highly accurate GPS will provide technology that is imperative for this market to scale.