Elon Musk May Be Secretly Planning to Use LiDAR in Future Tesla Models [Updated]
Ignore Musk's tweets and what he's said in earnings calls, a Tesla equipped with LiDAR may be in the works.
By now, everyone in the automotive technology community knows Tesla CEO Elon Musk's opinion on using LiDAR systems in the company's vehicles: they're overkill. But most people working on autonomous driving systems disagree with him—and that includes engineers within his own company.
Light Detection and Ranging (LiDAR) scanning uses beams of light to create a highly accurate "point cloud" of objects it sees and builds a digital three-dimensional view of the surrounding area. It's what Google, HERE, and Uber are using to build the highly accurate maps self-driving cars will use for navigation, and what most manufacturers are using to ensure the safety of their autonomous driving system for occupants and pedestrians around them. But not Tesla.
Musk is alone in his pursuit of a non-LiDAR based autonomous driving platform. In earnings calls he has defended his choice to use only cameras, radars, and sensors in current versions of Autopilot. Doubling down on that position, Musk compiled a hardware package that will be installed on all new Tesla vehicles going forward. Although the system includes eight cameras, 12 ultrasonic sensors, a forward-facing radar, and enhanced processor, it still lacks the software needed to make it operate on true "autopilot." Theoretically, the programming can be downloaded at a later date through over-the-air updates, then activated if the owner elects to check the box on the $8,000 option.
But if Tesla is committed to autonomous driving without LiDAR, why have Tesla test vehicles sporting Velodyne and Quanergy LiDARs been photographed near its headquarters?
UPDATED: A spokesperson for Tesla responded with this statement: "The claim that Tesla may be planning to use LiDAR as part of its self-driving hardware suite is fundamentally untrue. We regularly test our own technologies against other sensors to calibrate our camera, sonar and radar system."
The Drive stands by its reporting.
Tesla most likely will use LiDAR systems in future models, according to a source familiar with the technology. Plans are already in motion to integrate LiDAR, so it's not clear why Musk publicly appears dead set against it on Twitter and on earnings calls. After all, the accuracy of LiDAR is empirically superior to other solutions, especially when combined with a separate radar as a failover and backup to assist in dense fog (radar can see through rain and snow just fine), and it's the approach most other automotive manufacturers are taking. The motivation is likely more financial than scientific, although not for the underlying reasons you probably think.
"He wants to add functionality to the cars he produces today, and nobody today can supply [low-cost] LiDAR until 2018," says Brad Templeton, chairman of the Electronic Frontier Foundation and a software architect who consults with a few automotive manufacturers developing autonomous vehicles.
Templeton says that cost for the component is falling dramatically and quickly. Mechanical LiDARs seen on Google's autonomous cars and Ford's test vehicles is an older-generation that at one time cost approximately $70,000. But Menlo Park-based Quanergy will likely be the first manufacturer to mass-produce solid-state LiDAR that will retail for $250. Its competitor Velodyne, which recently received a $150 million investment from Ford and Baidu, isn't much further behind—and it's only a matter of time until $10 Lidar-on-a-chip becomes a reality.
That said, low-cost LiDAR at the volumes needed for mass market cars won't be available until at least next year. For most auto-manufactures planning to release autonomous vehicles in 2020 or 2021, this timeline isn't a problem. But if an auto maker wanted to advertise full autonomous capability on a production vehicle today, the only option is to cobble together a robust system of cameras, radars, and sensors and cross your fingers that AI software capabilities catch up. That's a risky bet.
And it's a bet that is debated hotly by computer vision and software engineers on obscure technology blogs. The gist of the argument is usually, "if a system that uses cameras and radar is "good enough," why use additional technology such as LiDAR that costs more?"
"I think the question is, if we have infinite processing capability infinitely quickly, can we do the same thing that humans do with two eyeballs and a brain? Theoretically yes, but it is such a difficult job that it is practically no," says Alex Lidow, CEO of Efficient Power Conversion in El Segundo, Calif., which makes underlying technology behind some brands of the low-cost solid state LiDAR entering the market. In short, it's just not possible for what Tesla is proposing to be "good enough."
The reason is that computer vision is not yet as good as a human's. Humans use 3D and other inputs such as light, shadows and other information that we've learned over time to make decisions about objects at a distance. LiDAR can detect and build 3D images, then classify them in nanoseconds. Radar, not so much.
The computationally intensive processing required to detect objects with 100 percent accuracy isn't a problem as long as you have infinite computation ability, but that isn't the case today, says Lidow. Current capacity constraints create latency and systems prone to errors. Think: MC Escher prints. Even with all these hardware and software hurdles removed, the same approach using complex AI algorithms on neural networks would take time until it could be considered safe, and there would be fatalities along the way as the software learns, Lidow explains.
Proponents of using LiDAR in autonomous systems believe that it's not worth risking the lives of pedestrians and passengers by trading accuracy for a few hundred dollars of cost savings. Using anything but the best technologies on the market—especially with a brand like Tesla that has a reputation for being cutting edge—doesn't make financial or engineering sense. That's why every other auto maker plans on using LiDAR in their solution, and probably why the electronic vehicle manufacturer's future models reportedly will use LiDAR as well.
"They would be foolish if they’re not looking at all technologies, and Elon Musk is no fool," says Lidow.