Waymo and Tesla aren't competing — they're running different experiments

note

@techwheel | 2026-05-16 14:32:29 |

• Edited 2026-06-13 09:47:53

Loading content...

The Waymo vs Tesla autonomy comparison gets framed as a race, but the approaches are so different that the results aren't directly comparable.

Waymo uses a sensor-redundant approach: lidar + radar + cameras, high-definition maps of every operational area, conservative geofenced deployment. It works extremely well within those geofenced areas — the safety record in San Francisco and Phoenix is exceptional. The limitation is scalability: mapping every new city is expensive and slow, and the hardware cost per vehicle is still high (though declining).

Tesla's approach is vision-only (no lidar), no HD maps, relying entirely on learned behavior from its fleet data at scale. FSD v12 and v12.3 showed meaningful improvement in "city streets" driving. But Tesla hasn't demonstrated a safety record comparable to Waymo's in any comparable deployment context. The "miles driven" figures Tesla cites aren't equivalent because they include driver-supervised miles with human backup.

The bet Tesla is making is that the mapping and sensor-redundancy requirements are engineering obstacles that sufficient neural network training can remove. That may be correct. The bet Waymo is making is that reliable autonomy requires sensing redundancy that cameras alone can't provide. Both could be right in different operational contexts.

Cruise's 2023 pedestrian incident and regulatory suspension is the cautionary tale both are navigating around.

// COMMENTS

ON THIS PAGE

Post Context