Waymo Criticizes Tesla's Method as 'False Summit'

Waymo published an article titled 'Ten AI Lessons from Over 200 Million Miles of Fully Autonomous Driving' on its official blog and X account. Written by Srikanth Thirumalai, Head of Engineering, the article presented key lessons Waymo learned during the development of autonomous driving technology, based on over 200 million miles of driving data.

In the blog post, Waymo argued that autonomous driving is impossible without multimodal sensors, emphasizing that cameras alone are not sufficient to achieve large-scale, safe fully autonomous driving. Waymo explained that each sensor has a unique role, with LiDAR identifying 3D shapes, cameras recognizing traffic lights and signs, and radar tracking speed even in adverse weather conditions.

Waymo likened attempts to achieve full self-driving by merely improving driving assistance systems to a 'False Summit,' highlighting its limitations. They argued that no matter how well FSD is refined, it will not lead to the completion of full self-driving, and only systems designed from the outset for uncrewed driving can reach the final summit.

Waymo criticized the 'black box' method, where AI decision-making processes are opaque, stating that it cannot build trust. They noted that pure neural network approaches carry a risk of failure, and Waymo operates an independent verification system to mitigate such risks. This expresses concern about potential dangers that can arise when AI produces results but cannot explain its process.

Waymo announced that it has reduced autonomous driving hardware costs to under $20,000, more than halving the price compared to the previous generation. They significantly lowered costs compared to Jaguar-based robotaxis, achieving this reduction by combining US-designed sensors and computers with ZEEKR vehicles.

Waymo has consistently emphasized the importance of using all three sensors—LiDAR, camera, and radar—arguing that autonomous driving is impossible without multimodal sensors. However, newly released vehicles show a 42% reduction in the number of sensors. This expansion of camera and AI's role could be interpreted as a shift towards Tesla's autonomous driving direction.

Currently, Waymo operates approximately 3,000 robotaxis in over 10 US cities, recording more than 500,000 paid rides per week, making it the leader in the robotaxi sector. This figure shows a significant difference compared to Tesla's estimated unsupervised robotaxi numbers.

On May 27, a Waymo robotaxi was involved in a head-on collision with another vehicle exiting a parking lot in California. The accident video showed the vehicle not stopping or slowing down immediately after the collision, continuing to move, which led to suspicions of a 'hit-and-run.' Waymo explained that it moved to exchange insurance information, but this explanation has been criticized as lacking persuasiveness.

A problem has been raised with multi-sensor-based autonomous driving systems: conflicting information from different sensors can lead to confusion in decision-making. For example, if LiDAR detects a collision but the camera reports seeing nothing, the system can become disoriented. The situation in the accident video, where all three of Waymo's sensors appeared not to detect the collision, contradicts Waymo's blog claims and has drawn criticism for lack of transparent explanation.

The ultimate winner in the robotaxi market is expected to be determined by consumer choice rather than technological debate. Actual accident rates and service prices are likely to be key factors in determining the winner in the robotaxi market. In this context, Waymo's blog post can be interpreted as an attempt to strategically convey a message ahead of Tesla's major event.