Waymo CEO Explains Why Camera-Only Self-Driving Falls Short

Waymo co-CEO Dmitri Dolgov provided a compelling technical argument as to why camera-only self-driving systems are insufficient for achieving full autonomy. In a talk at Y Combinator's Startup School, he highlighted that "weak sensing" ultimately creates a safety limit before reaching superhuman performance. Dolgov addressed the ongoing debate about the sensors required for autonomous driving, stating that camera-only solutions do not fully capture the solution.

While humans can drive using just their eyes, he acknowledged that this approach is reasonable to reach approximate human performance or build an assist product. However, when aiming for full autonomy and superhuman performance, Dolgov argued that relying solely on cameras results in a safety curve that levels off prematurely. He emphasized that camera, lidar, and radar are not redundant, but complementary sensing modalities that together create a superior view compared to a single sensor.

Furthermore, combining multiple sensor types safeguards against physical failures, such as a leaf or branch obstructing a camera, and helps Waymo progress along the "exponential ladder of nines" necessary for fully driverless safety. Dolgov cautioned that camera-only systems may experience rapid improvements before hitting a plateau long before the performance level required for product success.

Additionally, he challenged the notion that lidar is prohibitively expensive, asserting that hardware costs are expected to continue declining over time.