Wayve 公司 CEO Alex Kendall 分享其端到端 AI 方法如何在 500 多个城市、多种车辆中实现零样本驾驶，目标是将无人驾驶技术推广至全球。
Alex Kendall, CEO of Wayve, shares how his company's end-to-end AI approach enables zero-shot driving in over 500 cities across diverse vehicles, aiming to scale driverless technology globally.
要点 · TL;DR
端到端学习是自动驾驶的制胜方法。 End-to-end learning is the winning approach for autonomous driving.
内置传感器的量产车将超越机器人出租车。 Mass-market vehicles with built-in sensors will outscale robotaxis.
安全由最小风险策略保障，而非电车难题。 Safety is ensured by minimal risk maneuvers, not trolley problems.
核心观点 · Key points
自动驾驶本质上是一个AI问题，最好通过端到端学习来解决。 Autonomous driving is fundamentally an AI problem best solved with end-to-end learning.
扩展通用AI驾驶系统需要海量数据、算力和用于仿真的世界模型。 Scaling a generalizable AI driver requires massive data, compute, and world models for simulation.
未来是内置传感器的量产车，而非昂贵的改装车。 The future is mass-market vehicles with built-in sensors, not expensive retrofits.
安全至上；最小风险操作总能提供第三个好的选择。 Safety is paramount; minimal risk maneuvers always provide a third good option.
语言集成提升了驾驶模型的表征、推理和个性化能力。 Language integration improves representation, reasoning, and personalization in driving models.
对于深度学习，基础设施和数据质量比算法创新更为关键。 Infrastructure and data quality are more critical than algorithmic innovation for deep learning.
反共识 · Contrarian takes
端到端学习十年前被嘲笑，但现在已成为制胜方法。 End-to-end learning was laughed at a decade ago but is now the winning approach.
无需预先训练数据，即可在500多个城市实现零样本泛化。 Zero-shot generalization across 500+ cities is possible without prior training data.
短期内，消费级车辆将远超无人驾驶出租车的规模。 Consumer vehicles will outscale robotaxis in the short and medium term.
电车难题是理论练习；实际系统总有第三个选择。 The trolley problem is a theoretical exercise; practical systems always have a third option.
优先关注基础设施而非算法是一个后来才学到的重要教训。 Focusing on infrastructure over algorithms was a key lesson learned late.
自动驾驶通过消除人为错误，可将事故率降至接近零。 Autonomous driving can reduce accidents to near zero by eliminating human error.
本期章节 · Chapters（共 21）
引言与Wave起源Introduction and Wave's Origin Story
个人背景与领导力Personal Background and Leadership
构建首个端到端原型Building the First End-to-End Prototype
从博士到创立WayveFrom PhD to Founding Wayve
首次突破：10次干预实现车道保持First Breakthrough: 10 Interventions for Lane Following
在伦敦扩展与学习Scaling Up and Learning in London
竞争与自动驾驶格局Competition and the AV Landscape
商业验证与规模化Commercial Validation and Scaling
语言集成于驾驶AILanguage Integration in Driving AI
自动驾驶汽车个性化Personalization in self-driving cars
公众接受度与类人驾驶Public acceptance and human-like driving
技术授权决策Decision to license technology
当前重点与市场优先事项Current focus and market priorities
中国购车首要原因：自动驾驶Autonomy as a top purchase reason in China
安全原则与数据策略Safety principles and data strategy
应对驾驶行为规范Dealing with driving behavior norms
联合创始人过渡与CEO历程Co-founder transition and CEO journey