Overview
- Waymo co‑CEO Dmitri Dolgov made the case this week at Y Combinator that camera-only systems hit a capability ceiling and cannot be trusted to reach the extreme reliability required for fully driverless (Level 4) service.
- Waymo runs a multimodal perception stack in which each sensor—high-resolution cameras, lidar for direct 3D structure, and radar for velocity and through-obstruction sensing—has its own encoder and is fused into a single scene view to improve detection in darkness, glare, dust and fog.
- The companies’ field data sharpen the contrast: Waymo reports about 220 million rider-only miles and says its system shows far fewer serious-injury crashes than human drivers, while Tesla reports roughly 380,000 driverless miles with public reporting that shows higher crash rates per mile.
- Regulators are watching: the NHTSA has opened an engineering analysis into Tesla’s camera-based FSD degradation-detection system after identifying nine crashes where the system may not have detected degraded visibility or warned a driver in time.
- The debate has commercial stakes because falling lidar costs and high multimodal mileage at firms like Motional strengthen the case for sensor fusion, and the outcome could shape safety standards, deployment timing for robotaxis and what consumers face on the road.