It found Luminar while it was still working in secrecy.
Toyota is turning to a startup that operated for years in secrecy until making its public debut in April to accelerate the development of its self-driving car program.
On Wednesday, the research arm of Toyota unveiled its next-generation self-driving test car, a vehicle it says can more accurately detect objects and roadways and, as a result, better predict the safest driving route.
This Platform 2.1 vehicle will be tested on closed courses and eventually public roads in Silicon Valley, Ann Arbor, Mich., and Cambridge, Mass. Toyota Research Institute debuted its first generation autonomous vehicle in March 2017. The quick turnaround illustrates the quickening pace of development among automakers and tech companies hoping to deploy self-driving cars.
Automakers and tech startups develop new and improved iterations of its self-driving test cars all the time. But this one has a few notable changes and a new key piece of sensing technology called LiDAR that was developed by Luminar, the Silicon Valley startup founded by Stanford dropout Austin Russell.
LiDAR, or light detection and ranging radar, measures distance using laser light to generate highly accurate 3D map of the world around the car. LiDAR is considered by many companies an essential piece of technology to safely roll out self-driving cars to the masses.
Most rely on Velodyne, the leading supplier of LiDAR that has backing from Ford and Chinese search engine Baidu. Waymo, the Google self-driving project that recently spun out to become a business under Alphabet, as well as Uber are among the few developing their own LiDAR. (The technology is at the center of a lawsuit between Waymo and Uber.)
Russell contends that companies have been wrongly focused on driving down the price of LiDAR—instead of the performance. Velodyne LiDAR systems were once more than $80,000. Its 2016 prototype cost about $8,000 and the company is trying to reduce that even further.
“There really haven’t been any advancements in LiDAR performance for over a decade now because everyone has been re-packaging the same types of off the shelf components, albeit in newer, slightly more interesting ways,” Russell told Fortune. “People have been trading off performance to get lower price points. It’s optimizing for the wrong factors here.”
Luminar built its LiDAR from the ground up (Russell says “from the chip up”) and ended up simplifying the design and improving the performance, Russell says. Luminar’s LiDAR has only single laser and single receiver and 50 times better resolution and can “see” 10 times farther than existing LiDAR platforms, Russell says. The company’s LiDAR is also scalable, meaning it will be easy to make at high volumes, keeping costs low. Russell would not reveal a price.
The image below shows Luminar’s LiDAR in action.
TRI discovered Luminar before it came out of stealth mode in April and has been able to quickly adopt Luminar’s platform, Russell said.
Luminar is also building out a factory in Florida that will start to produce 10,000 of these LiDAR units this year. The units will be used by companies working on autonomous vehicles. Luminar has partnered with four companies. Toyota is the first to be named.
The new 2.1 self-driving car embodies Toyota’s and TRI’s philosophy towards deploying autonomous vehicles. The company is taking a dual approach to autonomy that it calls “Guardian” and “Chauffeur,” both of which use the same technology stack.
While the company will develop fully autonomous cars to serve an aging population and the disabled, it’s also working on technology for regular production cars that could switch between assisted and full autonomy. Unlike other automakers that might have drivers hit a button to switch in and out of full autonomous mode, Toyota would have it operate silently in the background.
Under “guardian” mode, the human driver maintain vehicle control and the automated system would step in and take control of the car to avoid an accident. Chauffeur would take over the driving and the vehicle’s occupants would be passengers.
The 2.1 self-driving car has been built with two steering wheels and sets of brakes and pedals as a research tool to test these two approaches.
The setup is supposed to help Toyota researchers find the best ways to transfer vehicle control between the human driver and the AI-controlled autonomous car. The research arm also said it will help develop machine learning algorithms that can learn from expert human drivers and provide coaching to novice drivers.