Waymo: AI Alone Won't Solve Self-Driving, Despite Rivals' Bets

Waymo pushes back on AI shortcuts

Waymo is drawing a line in the autonomous vehicle industry's biggest debate: whether breakthrough AI models can fast-track the path to safe self-driving cars. After more than 15 years of development and 200 million driverless miles on public roads, the company's answer is a firm no.

Srikanth Thirumalai, Waymo's vice president of onboard software, told Axios that deploying autonomous vehicles at scale requires more than sophisticated AI. The company's position directly challenges competitors like Tesla, Wayve, and Waabi, which are building what they call AV 2.0 systems — end-to-end neural networks that process raw sensor data and output steering commands with minimal intermediate steps.

Why it matters

The technical approach that wins this debate will shape which companies dominate the autonomous vehicle market. If Waymo is right that there are no shortcuts, its massive head start in testing and real-world data becomes an insurmountable advantage. But if newer AI techniques prove safe enough, rivals could leapfrog years of expensive development work.

Safety versus simplicity

Waymo began as Google's self-driving project in 2009, initially relying on many specialized AI models for specific tasks like pedestrian detection or traffic light recognition. The company has since consolidated toward larger foundation models, but it has stopped short of the pure end-to-end approach some competitors embrace.

Thirumalai explained the fundamental problem: even the most advanced AI models hallucinate, producing confident but incorrect outputs. In a physical system controlling a vehicle, there's no refresh button. "You have to deal with the consequences of it," he said.

Waymo tested pure end-to-end systems internally but concluded they couldn't meet the company's safety standards at operational scale. The company published a detailed blog post Wednesday outlining 10 AI lessons from its 200 million autonomous miles, according to details first reported by Axios.

The sensor debate

Waymo also weighed in on whether cameras alone can achieve full autonomy — a direct challenge to Tesla's camera-only strategy. Thirumalai said Waymo's testing found that combining cameras, lidar, and radar produced far better visibility than eliminating any sensor type.

"The AI can only make sense of what it sees, and if you just can't see it, the AI can't do much," he noted. The same principle applies to high-definition mapping, which some newer entrants claim is unnecessary but Waymo considers essential.

Competitive positioning

Thirumalai framed the blog post as a response to narratives that could undermine public trust in autonomous vehicles. But the timing also serves Waymo's competitive interests. By arguing that safety requires extensive real-world testing and multi-sensor systems, Waymo effectively raises the bar for rivals trying to reach market with leaner approaches.

Whether rapid AI advances will validate Waymo's caution or prove there's a faster path remains an open question. For now, the company with the industry's longest track record is betting there are no silver bullets.

These details were first reported by Axios.