物理人工智能，正式走出实验室
“物理人工智能的 ChatGPT 时刻已经到来——机器开始理解、推理并在现实世界中行动。无人驾驶出租车是首批受益者之一。”
在今年 CES 主旨演讲上，英伟达 CEO 黄仁勋给出了一个极具标志性的判断。
“The ChatGPT moment for Physical AI has arrived—machines are now able to understand, reason, and act in the real world. Robotaxis will be among the first beneficiaries.”
This was the key message delivered by NVIDIA CEO Jensen Huang at CES.
英伟达开源 Alpamayo 1，推理型自动驾驶登场
当地时间 1 月 5 日，黄仁勋宣布开源视觉-语言-行动（VLA）模型 Alpamayo 1，并已在 Hugging Face 上发布。
该模型专为增强自动驾驶决策能力而设计，可被汽车开发者用于调优、蒸馏，成为完整辅助驾驶技术栈的核心基础。
On January 5, NVIDIA officially open-sourced its Vision-Language-Action (VLA) model, Alpamayo 1, now available on Hugging Face.
The model is designed to enhance autonomous driving decision-making and can be fine-tuned or distilled into the core of an ADAS stack.
100 亿参数，把“为什么这样开”说清楚
在技术层面，Alpamayo 1 基于 100 亿参数架构，最大特点在于：
它不仅能给出决策结果，还能清晰展示每一个驾驶决策背后的推理逻辑。
Built on a 10-billion-parameter architecture, Alpamayo 1 does more than output driving actions.
It explicitly explains the reasoning behind each decision—an essential step toward trustworthy autonomy.
VLA 路线 vs 世界模型：两种理解世界的方式
所谓 VLA 模型，本质上是通过视觉感知、语言理解和动作生成的串联，用语义化的语言 token 来表达现实世界。
而在自动驾驶领域，另一条路线是“世界模型”，用连续状态变量和物理关系描述世界，特斯拉便是代表之一。
VLA models connect perception, language understanding, and action generation, representing the world through semantic tokens.
In contrast, world-model approaches—championed by Tesla—describe reality using continuous states and physical relationships, treating the car as a robot on wheels.
三件套：模型、仿真、数据，构建自强化闭环
除 Alpamayo 1 外，英伟达还同步推出了两项关键工具：
AlpaSim —— 完全开源的端到端高保真自动驾驶仿真框架；
物理 AI 开放数据集 —— 超 1700 小时真实驾驶数据，覆盖多区域、多场景。
Alongside Alpamayo 1, NVIDIA released two complementary tools:
AlpaSim, an open-source end-to-end high-fidelity driving simulator, and
a Physical AI open dataset containing over 1,700 hours of driving data across diverse environments.
推理型自动驾驶，直指 L4
英伟达表示，这三项工具将形成一个自我强化的开发闭环，专门服务于“推理型自动驾驶堆栈”的构建。
目前，包括 Lucid、捷豹路虎、Uber、伯克利 DeepDrive 在内的多家机构，已明确表达了对 Alpamayo 的浓厚兴趣。
NVIDIA claims these tools together create a self-reinforcing development loop for reasoning-based autonomous driving stacks.
Companies like Lucid, Jaguar Land Rover, Uber, and Berkeley DeepDrive have already shown strong interest in building L4 systems on this foundation.
自动驾驶出租车，被重新点燃
花旗分析师在最新报告中指出，英伟达的新生态有望显著推动自动驾驶出租车的发展。
预计全球 Robotaxi 市场规模将从 2025 年的 44.3 亿美元，增长至 2034 年的 1889.1 亿美元。
According to Citi analysts, NVIDIA’s new ecosystem could reignite demand for robotaxis.
The global robotaxi market is projected to expand from $4.43 billion in 2025 to $188.91 billion by 2034.
资本市场，已经先行反应
受消息刺激，A 股运输服务及智能驾驶板块集体走强，大众交通、浙江世宝、德赛西威等多股涨停。
港股方面，小马智行、文远知行创上市新高，小鹏汽车、地平线机器人等智驾概念股同步上涨。
Markets reacted swiftly.
In China’s A-share market, transport and smart driving stocks surged, while in Hong Kong, autonomous driving leaders hit new highs.
中国车企，早已提前布局 VLA
从国内视角看，在英伟达开源 Alpamayo 1 之前，理想、小鹏、小米等车企已率先布局 VLA 自动驾驶模型。
这一趋势仍在加速扩大。
From a China perspective, companies like Li Auto, XPeng, and Xiaomi had already invested heavily in VLA-based autonomy before NVIDIA’s announcement.
The momentum continues to accelerate.
小米、小鹏，走向“技术外溢”
去年 11 月，小米披露全年 AI 研发投入超 70 亿元，并持续加码 VLA 项目；
何小鹏也宣布，将在今年开源小鹏第二代 VLA 模型，大众汽车将成为首批合作伙伴。
Xiaomi revealed over RMB 7 billion in annual AI R&D spending, with VLA as a key focus.
XPeng also announced it would open-source its second-generation VLA model this year, with Volkswagen as the first customer.
L3 政策加速，智驾进入“平权时代”
国元证券指出，当前智能驾驶正沿着“端到端”“智驾平权”快速推进。
高速与城区 NOA 功能正在下探至 20 万元以下车型，技术门槛持续降低。
According to Guoyuan Securities, autonomous driving is accelerating toward end-to-end architectures and “driving intelligence democratization.”
NOA features are rapidly reaching vehicles priced below RMB 200,000.
2026：智能化，成为第二增长曲线
方正证券认为，即便 2026 年行业补贴继续退坡，汽车板块仍具备结构性机会。
自动驾驶、AI 算力、机器人，将成为第二甚至第三增长曲线。
Founder Securities believes that even as subsidies fade in 2026, structural opportunities remain.
Autonomous driving, AI computing power, and robotics are emerging as the next major growth engines.
尾声：不是有没有智能驾驶，而是谁先理解世界
当自动驾驶进入“推理时代”，竞争不再是传感器数量或算力堆叠，
而是谁能更好地理解世界、解释决策，并在现实中安全行动。
As autonomous driving enters the era of reasoning, the competition is no longer about sensors or brute-force compute.
It’s about who can truly understand the world, explain decisions, and act safely in reality.
#西亚跨境视角 #品牌出海 #品牌运营 #跨境电商 #品牌 #出海 #汽车出口 #出口贸易 #传媒 #IP孵化 #社媒运营 #市场 #商品 #产品 #跨境电商运营 #外贸知识 #tk #跨境 #跨境电商 #tiktok #跨境电商运营 #跨境人 #跨境运营 #跨境独立站#汽车
