Waymo’s co‑CEO Dmitri Dolgov was stunned when a San Francisco robotaxi predicted a pedestrian hidden behind a bus and swerved to avoid a collision. The moment exposed a leap in perception that hinges on artificial‑intelligence models rather than solely on sensors. This breakthrough underpins Waymo’s fifth‑generation platform.

The shift began with massive simulation, turning raw radar, lidar and camera data into virtual cities where edge cases could be tested safely. In 2020, Waymo rolled out its fifth‑generation Driver, powered by a foundation model that fuses perception, planning and a critic that rates performance.

Yet Waymo’s heavy hardware stack faces rivals that strip lidar and rely on cheaper sensor suites. Companies like Waabi and Wayve claim their end‑to‑end deep‑learning models generalise beyond geo‑fenced maps, offering a lower‑cost path to mass deployment. Their strategy hinges on fewer cameras, simpler software and a single‑platform architecture that can scale across cities without bespoke mapping.

Waymo now operates in ten US cities, targeting one million paid weekly rides by year‑end and projecting 7 % of the US rideshare market by 2030. As competitors like Zoox, Tesla and WeRide expand, the race to lower cost and broader coverage will dictate who dominates the urban automation economy.