Waymo CEO John Krafcik is delivering the keynote address at the North American Auto Show’s Automobili-D conference Sunday, and he’s revealing a lot about the new Alphabet company’s progress and tech. Waymo, which began life as Google’s self-driving car project, has built all of its own sensors for use on its latest test vehicles, self-driving Pacifica minivans. Use of in-house sensors is huge, since it means Waymo can likely lay claim to controlling more of the self-driving car technology stack than any other player in operation today.
Much like Apple can do a lot more than most of tis competitors because it owns and builds both the software and the hardware for its product suite, Waymo can claim tighter integration between its sensor hardware, sensor fusion software, image recognition and other aspects of its self-driving system. Waymo also claims individual performance benefits in each of its new sensors, including vision cameras, radars and LiDAR, saying each provides better resolution, sensing distance and accuracy than the hardware it has been using on prior vehicles.
Krafcik also says that its decision to build out all of its sensor suites in-house has a significant talent benefit, since it means that it’ll now have deeply knowledgeable experts in each key sensing area, who can work closely with resources on the AI engineering side to ensure the best possible use of the gathered data by the car’s in-vehicle computational systems. Plus, development cycles should get shorter between generations of software and hardware thanks to their ability to control both.
Waymo also created two new categories of LiDAR in its effort to develop its own sensors. Previous vehicles used a single top-mounted medium-range LiDAR, but new short and long range sensing units included in this package mean that the surround view the car sees is now complete, and more granular detail detection both up close and at a distance.
Maybe most important from the perspective of Waymo’s ability to woo automaker partners is that building its own sensors gives the company significant cost advantages down the road. After the initial cost of research and development, cost reduction on a per unit basis should grow significantly over time, leading to Waymo’s ability to get a lot closer to reaching prices car OEMs are more comfortable at scale, more quickly. Krafcik says the cost of a single top-end LiDAR, which was $75,000 when Google began experimenting with self-driving vehicles, has dropped by 90 percent thanks to its efforts to build its own.
Waymo will field Pacificas with this homegrown sensor suite on board beginning in Arizona and California later on this month, Krafcik announced on Sunday. Throughout its time as Google’s self-driving car project and more recently as Waymo, the company has racked up almost 2.5 million miles driven, with most of those accumulated on city streets. It just take them just eight more months to pass the 3 million mile threshold. Volume is key for autonomous tech not only for training systems, but also for helping prove to regulators and industry stakeholders that th tech is actually safe.
Speaking of safety, Waymo says it’s had a four-fold improvement in performance of its vehicles in the last year, as measured by the number of disengagements (when a safety driver has to stop the self-driving software and take over manual control). In 2015, the number of disengagements was around 0.8 times per thousand miles in California, and currently, Waymo’s only seeing around 0.2 disengagements per thousand miles travelled.