In March 2020 Waymo introduced the fifth generation of the Waymo Driver, the hardware-and-software system it puts on its vehicles. The announcement leaned on scale and experience: the design drew on “20 million self-driven miles on public roads and over 10 billion miles of simulation.” Waymo’s core engineering claim is that no single sensor type is enough, so it builds complementary sensors that work as one system.
The fifth generation centered on three custom sensor families. The redesigned lidar delivered “higher resolution across a 360 degree field of view with > 300 meter range,” split between a roof-mounted 360-degree unit and four perimeter lidars for close-in city maneuvering. The camera set added long-range optics that could pick out “pedestrians and stop signs greater than 500 meters away.” And Waymo described building “one of the world’s first imaging radar” systems for self-driving, able to flag objects that are “moving, barely moving, or stopped.” Crucially, the company said it achieved all this at “half the cost of our previous generation.”
That cost line is the real story. The hard part of robotaxis is not a one-off demo but a fleet you can afford to build and maintain, and halving sensor cost while improving capability is what makes scaling plausible. The fifth-generation Driver is the platform behind Waymo’s driverless rides in Phoenix and later cities.
For a general reader, this entry shows the unglamorous engineering that turns a science project into a business: custom silicon and sensors, redundancy across modalities, and a relentless focus on bringing the per-vehicle cost down.