Dmitry Dolgov, one of Waymo's executives, criticized the camera-only approach to developing self-driving cars. Speaking at the Y Combinator Startup School event, he stated that such a set of sensors can only provide a limited level of safety and is not suitable for fully autonomous driving.
Although Dolgov did not directly mention Tesla, his words clearly echo Elon Musk's company's strategy. Tesla relies on cameras and computer vision algorithms, having gradually abandoned lidars, radars, and ultrasonic sensors. Musk himself called lidars “expensive and unnecessary” back in 2019, arguing that the future belongs exclusively to cameras.
According to Dolgov, humans do indeed drive cars primarily using vision. However, the task of an autonomous system is significantly more complex.
If the goal is only to approximate human capabilities or create a driver assistance system, cameras are a perfectly reasonable solution.
Waymo uses several types of sensors. Cameras are responsible for object and color recognition, lidars build a three-dimensional map of the surrounding space, and radars measure the speed of moving objects. This information is then combined into a single model of the environment.
According to Dolgov, this combination is especially important in challenging conditions. For example, during tests in a dust storm in Arizona, lidar was able to detect a pedestrian that cameras could barely distinguish.