A Waymo self-driving Jaguar I-Pace recently collided with a pedestrian on Haight St. in San Francisco. This incident occurred while a human safety driver was at the wheel, operating the vehicle in manual mode, according to techcrunch.

Waymo is developing advanced AI to simulate human decision-making for crash avoidance. Yet, this real-world incident involved a human safety driver, exposing a tension between AI capabilities and human oversight. The company's focus on replicating human responses in its virtual driver by 2026 presents a paradox.

Despite significant technological advancements in simulating human reactions, truly safe and fully autonomous vehicles still face complex real-world challenges. This necessitates robust testing and a deeper understanding of edge cases.

Inside Waymo's Human-Like Reaction Model

  • Waymo developed ReD (Reference Driver), a new computer model simulating human crash-avoidance decisions, as reported by The Verge.
  • Developed with TU Delft, ReD compares Waymo's autonomous software to human capabilities, according to Zamin Uz.
  • ReD even includes a 0.2-second pause, mimicking the human foot shifting between pedals, a detail noted by The Verge.

This integration of nuanced human reaction times into autonomous decision-making is a significant step. Waymo believes human predictability enhances safety, aiming for more human-like responses. The meticulous effort to embed human reaction delays into ReD reveals a critical insight: true autonomous safety might hinge on predictable, human-mimicking responses, rather than purely optimized, alien-fast reactions, forcing a re-evaluation of what 'better' driving truly means.

ReD in the Context of Waymo's Broader Tech

The 6th-generation Waymo Driver uses a next-gen 17-megapixel imager, providing clearer environmental data with fewer cameras than previous generations. This streamlined sensor suite optimizes efficiency and data quality, directly enhancing inputs for cognitive models like ReD. Ironically, while Waymo engineers human-like predictability into AI, the recent collision involving a human safety driver underscores that human intervention, not AI failure, remains the most unpredictable variable. This challenges the very premise of human 'oversight' as a reliable safety net, suggesting that ReD's human-mimicking responses might ultimately make AVs safer for humans, even from other human errors.