NIO Achieves Cross-Platform Intelligence Driving Software Deployment
文｜肖漫 编辑｜李勤 6月18日，蔚来同时向两代平台车型（包含8款NT2.0平台车型、4款NT2.5平台车型，以及6款NT3.0车型）推送了最新版的世界模型，这意味着，蔚来现在能让同一套复杂的智驾代码，现在能跑在不同代际的芯片上。 软件迭代节奏被硬件绑架曾是一个困扰行业的难题。很多车企无法在不同版本、配置的车型上迭代同一款软件，这带来的结果是，很长时间内只有使用最新版本硬件的车才能用上最好的软件，老车主被背刺。 蔚来部署推理跨平台兼容 任少卿的团队在 2020 年就在思考如何解决这个问题。蔚来的做法是搭建一套AI Infra——自研工具链铺平不同芯片间的鸿沟，用AI编译器提升车辆的处理速度，用 AI Agent 自动化串联全流程。 当时业内主流的做法还是用英伟达的工具进行上层部署。但当时的蔚来判断，车端芯片的工程架构会继续快速迭代，主流的架构仅能用 3-5 年。基于这一判断，蔚来决定只保留最底层的硬件接口层（如 CUDA），在此之上全面自研上层部署软件，包括推理引擎、部署框架。 另外，和大多数自研芯片的主机厂一样，蔚来也自研了编译器，实现了自动算
NIO has successfully deployed its latest world model software across multiple vehicle generations, including NT2.0, NT2.5, and NT3.0 platforms. This breakthrough allows a single set of intelligent driving code to operate on different chip architectures, overcoming a long-standing industry challenge where software updates were often tied to new hardware. NIO's self-developed AI infrastructure, including its inference engine, deployment framework, and AI compiler, enables rapid deployment and improved processing speeds. The company utilizes a 'shadow mode' to test new models in real-world conditions without interfering with driving, collecting valuable data for continuous improvement. This data-driven approach, focusing on 'corner cases' identified in shadow mode, significantly enhances the value of training data.
NIO's achievement in cross-platform intelligent driving software deployment represents a significant advancement in automotive AI, potentially setting a new standard for software updates and user experience.
📌 Kaynak
Bu haber XML kaynağından derlenmiştir. Tamamı için orijinal habere gidin.
Orijinal haberi oku →