Waymo has developed a computational model, termed the Reference Driver (ReD), designed to simulate human cognitive responses during traffic conflicts to establish a standardized safety benchmark for autonomous driving systems.

Waymo 開發了一款名為「參考駕駛員」(Reference Driver, ReD) 的計算模型，旨在模擬人類在交通衝突期間的認知反應，以為自動駕駛系統建立標準化的安全基準。

The development of the Reference Driver occurred through a collaboration between Waymo and the Delft University of Technology, with the resulting research published in Nature Communications. This model utilizes a neuroscience framework known as active inference, which posits that human cognition operates to minimize surprise and unpredictability. Unlike prior iterations of safety modeling that focused primarily on reactive, last-second maneuvers, the ReD model incorporates proactive avoidance by continuously calculating potential risks.

參考駕駛員是由 Waymo 與台夫特理工大學合作開發的，研究結果發表於《自然-通訊》(Nature Communications)。此模型利用了一種稱為「主動推理」(active inference) 的神經科學框架，該框架假設人類認知的運作是以將驚訝感與不可預測性降至最低為目標。不同於先前主要關注最後一秒反應式操縱的安全模型，ReD 模型透過持續計算潛在風險，將主動避險納入其中。

Technically, the model replicates specific human physiological and cognitive limitations. These include the phenomenon of 'looming' for longitudinal threat assessment, a bias toward traffic norm adherence, and a 0.2-second latency period when transitioning between acceleration and braking pedals. Such specifications allow for a more precise simulation of how a competent human driver perceives and reacts to environmental stressors.

在技術層面，該模型複製了人類特定的生理與認知限制。這些限制包括用於縱向威脅評估的「逼近」(looming) 現象、傾向遵守交通規範的偏差，以及在加速踏板與煞車踏板之間轉換時 0.2 秒的延遲。這些規格使得模型能更精確地模擬一名合格的人類駕駛員如何感知並反應環境壓力。

From a strategic positioning perspective, Waymo intends for this model to function as a behavioral analogue to the physical crash test dummies utilized in traditional automotive safety evaluations. The company is currently engaging with regulators and standards organizations, such as the SAE, to foster a rapprochement between industry practices and scientifically grounded safety definitions. To facilitate this objective, the research code has been released under an academic, non-commercial license.

從戰略定位來看，Waymo 打算讓此模型作為一種行為類比，如同傳統汽車安全評估中使用的物理碰撞測試 dummy 人。該公司目前正與監管機構及 SAE 等標準組織接洽，以促成產業實務與基於科學的安全定義之間的趨同。為實現此目標，研究代碼已根據學術非商業許可證發布。

The Reference Driver provides a scalable, open-source framework for comparing autonomous vehicle performance against human behavioral standards in complex crash scenarios.

參考駕駛員提供了一個可擴展且開源的框架，可用於在複雜的碰撞情境中，將自動駕駛汽車的表現與人類行為標準進行比較。