Waymo states AI cannot provide shortcut to self-driving technology

<p>Waymo has stated that advancements in artificial intelligence (AI) do not offer a shortcut to achieving fully autonomous vehicles. The company, which has been developing self-driving technology for over 15 years and has completed 200 million driverless miles, emphasizes the importance of safety in deploying autonomous vehicles (AVs) at scale.</p><p><strong>Importance of the statement:</strong> This assertion may influence the competitive landscape of the AV industry, determining whether it remains a lengthy and costly process that benefits established players like Waymo or allows newer competitors to enter the market more quickly.</p><hr /><p><strong>Details from the interview:</strong> Srikanth Thirumalai, Waymo's vice president of onboard software, shared insights in an interview with Axios, indicating that merely improving AI is insufficient for safe AV deployment. He noted that Waymo's objective has always been to achieve "demonstrably safe AI," and expressed concerns regarding competitors' reliance on single "end-to-end" AI systems, which may introduce safety risks.</p><ul><li>Thirumalai remarked, "Even the best AI models with trillions of parameters still hallucinate... We don't have a choice to say, 'Let's hit refresh' ... There is no click reboot or reload or refresh [in] physical AI. You have to deal with the consequences of it."</li><li>In a blog post published on August 26, Thirumalai elaborated on "10 AI lessons" that Waymo has learned from its extensive testing.</li></ul><p><strong>Background:</strong> Waymo originated as Google's self-driving car project in 2009, prior to the current deep-learning advancements. Initially, the company utilized various specialized models for tasks such as pedestrian detection and traffic light recognition. Over time, it transitioned to larger foundation models, leveraging advancements in AI.</p><ul><li>Competitors like Tesla, Wayve, and Waabi are now exploring AV systems that they claim can mimic human reasoning through "end-to-end" neural networks that process raw sensor data to produce steering commands directly.</li></ul><p><strong>Waymo's findings:</strong> The company has experimented with similar tools but concluded that pure end-to-end systems do not meet its safety standards at the operational scale required.</p><ul><li>Thirumalai stated, "What we found is that just pure end-to-end systems are not able to meet our safety bar at the scale that we operate."</li></ul><p><strong>Addressing public perception:</strong> Thirumalai mentioned that the blog post was intended to counter the narrative suggesting that AI could provide a shortcut to safe self-driving, especially as public trust in the AV industry is crucial.</p><ul><li>He noted that the safety standards Waymo advocates for the industry also highlight its competitive advantage, stemming from years of testing and real-world data.</li><li>While not naming specific competitors, Waymo outlined potential risks associated with newer AV development approaches, effectively raising the bar for those companies.</li></ul><p><strong>Discussion on sensor technology:</strong> Waymo contributed to the ongoing debate regarding the sufficiency of camera-only systems for achieving full autonomy.</p><ul><li>Thirumalai explained that Waymo's training indicated that combining cameras, lidar, and radar resulted in better visibility compared to relying on any single sensor.</li><li>He stated, "The AI can only make sense of what it sees, and if you just can't see it, the AI can't do much."</li></ul><p><strong>Critique of competitors:</strong> Thirumalai's comments also served as a critique of Tesla's reliance on a camera-only, end-to-end autonomous driving system.</p><ul><li>He asserted that improved AI cannot compensate for insufficient sensing capabilities and mentioned that this also applies to high-definition mapping, which some competitors deem unnecessary.</li></ul><p><strong>Conclusion:</strong> Waymo has invested over 15 years in developing its robotaxi services and maintains that there is no single solution to achieving safe autonomy. Thirumalai's insights underscore the company's leadership position in the industry, although the ongoing advancements in AI may still reveal alternative paths to achieving full autonomy.</p>