Thursday, October 19, 2017

Autonomous Vehicles: The King of LiDAR

A first rate piece on the backstory of the future.
From The Verge:

The billion-dollar widget steering the driverless car industry
No matter what it took, David Hall was going to kill that clown. He maneuvered Drillzilla for another ramming run. The robot was squat and heavy, with serrated blades coming off one end and a sharp drill whirling on the other. Across the arena, Conquering Clown awaited. It had the face of a goofy jokester, but its hands were a pair of smashing hammers and its body was equipped with a pair of circular saws.

Drillzilla managed to flank the clown, then ram it, sneaking blades under its body and lifting it up off its wheels. Its opponent was helpless, and Drillzilla pushed it onto a waiting geyser of flames. As the audience cheered, the clown’s grinning face melted away. “We were in there to make great TV,” recalls Hall with a chuckle, “and damn it, we were successful.”

The year was 2001, and this was the third round of the Robot Wars Annihilator Challenge. The show pitted homemade gladiators against one another. Hall, an eccentric inventor, was best known as the creator of a high-end subwoofer. His company, Velodyne, had around 60 employees and a few million dollars in annual sales. But Hall had grown bored with the audio industry, and was trying his hand at building robots. Perhaps, he thought, Velodyne could find a new product to manufacture. At the very least he could have some fun. 

Hall admits he “bent the rules a little” to win the competition — he camouflaged the robot’s wheels as “legs” — and, after other teams complained, the judges banned Hall’s approach for future seasons. He took that as a sign to move on. But he was eager to continue his work on vehicles in the public eye.

He got his chance in 2004, when the US Army’s research division, DARPA, held its first Grand Challenge. Teams were asked to design an automobile that could autonomously navigate its way through a 150-mile course. Hall took a bunch of the motor controls and code from his robots and got to work on a self-driving truck. 

There were 15 vehicles in that first race, all competing for a $1 million prize. Not a single one finished the course. Hall used stereo cameras to see the road and avoid obstacles. “I could see some of the road all the time, and all the road some of the time,” he says now. But the vehicles were constantly confused. “There was all kinds of weird artifacts. It would see a fence five feet in front of it and throw on the brakes.”

A few of the teams were using a technology called LIDAR, which uses laser beams to sense objects and measure their range. “I didn’t know what a LIDAR was,” admits Hall, “but Jim McBride from Ford kept bending my ear about how LIDAR was the solution for all his problems. I kinda made a mental note — maybe I’ll look into it when I’m bored. A few months later, I started looking into it, and the more I did, the more intrigued I was.”

Hall returned the next year with a custom LIDAR unit he had designed and built. Instead of putting a laser scanner on the front of his car, as most teams had done, he put it on the roof. And instead of looking forward, it scanned in all directions at once. It was a radically different approach to the technology. 

He didn’t win the race, but his design impressed competitors, and the DARPA challenge convinced a number of the participants that driverless cars were no longer an unattainable fantasy. The teams from Stanford and CMU would go on to help found the self-driving projects at Google and Uber. And as those massive corporations built out their self-driving ambitions, they turned to Velodyne for LIDAR.

At age 66, Hall finds himself in an enviable position. LIDAR has eclipsed Velodyne’s audio business; the company now has over 400 full-time employees people and generates hundreds of millions in annual revenue. Velodyne’s most recent round of funding made Hall a billionaire. And earlier this year, the company announced it was on the verge of rolling out a new kind of LIDAR, one that would make the technology radically cheaper, allowing it to move from expensive test vehicles to a standard piece of an affordable consumer sedan.

After four decades of inventing, five product categories, and a lot of wrong turns, Hall has stumbled onto a high-tech widget that is poised to become an essential piece of a trillion-dollar self-driving auto industry. Setting aside Steve Jobs’ return to Apple, it might be Silicon Valley’s greatest second act.

There are many ways of seeing the world. Humans rely on our eyes, which interpret incoming rays of light. On a sunny day, they allow us to see a richly nuanced view of the world. In total darkness, they aren’t much use. Some animals, like bats, use echolocation. While hunting small insects at night, they emit high-pitched noises, then interpret the sound waves that bounce back to get a picture of the world. 

Humans have created technologies that mimic echolocation. Sonar, used by submarines, emits a pulse of sound, then reads the waves that bounce back. Radar does the same thing, using radio waves instead. It’s found countless applications, from spotting incoming missiles to catching drivers who break the speed limit. LIDAR, short for “light detection and ranging,” adopts Radar’s approach, but uses lasers in place of radio waves. Scientists at Hughes Research Lab demonstrated the first functioning lasers in 1960, and LIDAR attempts quickly followed. Early on, it was used by government research agencies to map and measure the natural world, from cloud formations to sea floors to the surface of the Moon. 

By the late 1980s, LIDAR had found its way into autonomous cars. Researchers at Carnegie Mellon University’s Navlab used a laser scanner to help detect obstacles and determine their range back in 1989, but it was not their primary sensor. “We would see, with our scanning lasers, these very grainy updates every half second,” says Dean Pomerleau, a Navlab researcher. “A kid on a bicycle would just look like a blob.” The lab used LIDAR primarily because it was good at detecting reflective materials, like lane markings and road signs. In the late ‘90s, Mitsubishi actually tried LIDAR in its driver assistance system, but its high cost made it prohibitive. As the 21st century arrived, the auto industry moved to radar and cameras, which were much cheaper....MUCH MORE