最后是合成数据驱动、产品闭环飞轮。诺因以自研合成数据为核心，构建面向具身操控的训练体系：通过可规模化的任务生成、动作 / 轨迹生成与筛选机制，持续产出覆盖长尾场景的高质量训练数据，用于训练具备更强泛化能力的具身大模型。相较于高度依赖示教与真实采集的路径，公司更侧重 “可控、可扩展、可迭代” 的合成数据管线，并将产品与真机运行中的反馈、失败样例与关键场景抽象回流到数据生成与评测体系中，形成 “产品反馈 → 合成增强 → 训练迭代 → 体验改进” 的闭环飞轮。依托高质量合成数据管线，持续驱动模型能力提升，形成难以复制的自我进化体系，筑牢长期技术壁垒。该路线工程门槛较高，诺因已跑通关键环节，并在具身操控与任务泛化上形成可持续的增益与验证体系。
王天：这些是穿戴类消费电子产品的共性问题，就像用户希望手机屏幕大、重量轻、续航久一样，我们也希望产品越轻越好。但C端外骨骼产品的核心指标，很大程度上依赖于供应链，在行业技术路线没有颠覆性改变的情况下，各家产品的重量差异并不大，无非是百克级的差别。
，这一点在PDF资料中也有详细论述
However, in 1991, Friedmann, now twenty-two, robbed a comic-book store and then tried to steal trading cards from a Kroger supermarket. When a security guard confronted him, he brandished a pistol, ran outside, and drove off. Police followed him, and he crashed. The arresting officer recovered the pistol, then discovered a razor blade made to resemble a business card in Friedmann’s wallet. In his shirt pocket, the officer found a pen that was actually a stabbing pick. A knife was tucked in his shoe and handcuff keys were hidden on his belt and under a shoe sole. He had a portable cassette player with a hidden compartment for a gun. When the officer asked Friedmann how he’d stashed the weapons so cunningly, he said that he’d learned to do it while jailed in the Criminal Justice Center.
Qwen 3.5在2026年2月发布，这是千问系列的最新旗舰版本。它采用了混合架构，结合了Gated DeltaNet和稀疏混合专家（MoE）两种技术。
career, culture, engineering, leadership, teams