The Wild West of Automotive Lidar

The city of Houghton is in the far north of Michigan's upper peninsula, along the southern shore of Lake Superior. It's famous for two things: the notable engineering school, Michigan Technical University, and being two miles past the end of the Earth. It's more than 200 miles away from the closest freeway, and averages 250 inches of snowfall per year.

Jeremy Bos, assistant professor of electrical and computer engineering at Michigan Tech, finds this environment ideal for research on autonomous vehicles (AV). He's working to resolve corner cases—those situations that are rare and difficult to plan for, but do happen. Like epic amounts of snow.

"What is a normal condition up here in the UP—snowbanks, covered roads, etc., —that's a once in 100-year event in Chicago," says Bos. "But it does happen in places where you might actually want to field AVs. If AVs replace taxis in all major urban centers and they can't handle a snowstorm, what good are they?"

In addition to studying corner cases, Houghton is a good test case for unstructured environments, because few manmade features are present outside of town. "The problem in an unstructured environment is that one tree looks very much like another. You don't have these features like walls or corners or roadways or lane markings in order to localize yourself and say ‘I'm here,'" he says.

Lidar is one of several technologies that enable autonomous vehicles. Bos's group at Michigan Tech has been testing and evaluating various automotive lidars to figure out the minimum information needed relative to an unstructured environment to accomplish autonomous driving tasks. It's a challenge that Bos has to figure out as he goes, because there are no standard measurements available for lidar, which means that neither lidar suppliers nor auto manufacturers have a way to compare lidar products.

Bos's work testing lidars put him in contact with Paul McManamon, lidar expert, author, and president of Exciting Technology, LLC. "It's a wild west right now," says McManamon. "You can't compare between one [lidar] and the other. And no one tells you the performance. They won't tell you how it works, and they won't tell you the performance."

Frustrated by this lack of transparency, McManamon asked Bos to help him develop a series of multi-vendor benchmarking tests so that auto lidar companies will have something to compare against. This inaugural event which was planned to take place in Anaheim, California, during SPIE Defense + Commercial Sensing in April, will have to be rescheduled due to the COVID-19 outbreak and resulting restrictions.

The plan is to set up targets in a large parking lot to test key lidar capabilities: range of effective object detection; resolution, which is how accurately the lidar identifies and classifies objects; and reflectance confusion, meaning how well the lidar can see something in the presence of bright objects, like reflective signs or bright sun.

Developing a set of standards for automotive lidar will be no easy task. Different lidar companies rely on different types of lidar technology, including the standard spindle-type, which spins to give a 360-degree view; solid-state lidar, which has a fixed field of view, but no moving parts, and can use either MEMS or optical phased arrays to steer the beams; and flash lidar, which spreads a flash of light over a large field of view before the signal bounces back to a detector.

Moreover, automotive lidar has different range requirements depending on use case. If a car is driving on the autobahn, for example, then the car needs to be equipped with 250-meter lidar. But other applications, like congested city driving, really only need lidar to detect objects 50 meters away or less. "Everyone claims to have the best thing imaginable, but there are specific use cases for these lidars, and automotive manufacturers want to know that they work," says Bos.

These variations, along with closely guarded lidar IP, have made technology standards slow to develop. However, lidar is a key enabler in a growing market for autonomous vehicles, which means that measuring lidar performance will have a direct impact on the safety of those vehicles.

Automotive lidar may be a wild west, but McManamon and Bos know that the railroad is coming. The development of standard lidar measurements will help lay the first tracks.

Gwen Weerts is the managing editor of Photonics Focus.