本周在北京举行的世界机器人大会上，人形机器人展示了演奏摇滚乐、拳击和指挥交通等多种能力，吸引了大批观众。中国已将机器人产业列为战略重点和未来经济增长来源，汽车制造商、手机厂商和新创机器人企业纷纷涌入该领域。大会为期五天，吸引了超过三百家参展商，充分体现了中国在全球人形机器人生产中的主导地位。（关键数字：300）
尽管硬体技术进步惊人——宇树科技展示了据称能跳得比任何人更高、跑得比博尔特更快的机器人——但该公司创办人王兴兴指出，人形机器人的「ChatGPT时刻」仍需数年才能到来。他认为最大的瓶颈在于具身智能的泛化能力不足，需要收集更多物理AI模型所需的数据。业界对这一突破时刻何时到来尚未达成共识，预测从一年到十年不等。（关键数字：4, 7, two years）
目前人形机器人的训练数据收集面临成本高昂和品质参差不齐的挑战，模型参数规模也远未达到实现通用任务所需的水平。摩根士丹利预计中国今年将出货五万台人形机器人，但商业化部署仍然有限，主要集中在政府资助的训练项目和娱乐研究领域。分析师认为，未来一年的重点将是工厂的付费部署，而家用机器人仍遥遥无期。（关键数字：50,000, 2025）
The World Robot Conference in Beijing this week showcased China's dominance in humanoid robot production, featuring machines that played in rock bands, boxed, and directed traffic before enthusiastic crowds. Chinese policymakers have designated robotics as a strategic priority, mobilizing the world's largest manufacturing sector behind its development. The five-day event drew over 300 exhibitors, including carmakers, mobile phone manufacturers, and dedicated robotics startups, underscoring the scale of investment pouring into the industry.
Despite spectacular hardware advances — including robots claimed to outjump any human and outrun Usain Bolt — Unitree founder Wang Xingxing cautioned that the industry's 'ChatGPT moment' of true autonomous intelligence remains years away. He identified the generalisation of embodied intelligence as the biggest global bottleneck, requiring far more data for physical AI models. While some competitors predicted a breakthrough within one to two years, analysts noted that current models run on only 4 to 7 billion parameters, far short of the hundreds of billions likely needed for broad generalised task performance.
Training data collection remains a key challenge, with methods ranging from human operators wearing sensors to simulation software, each carrying significant cost or quality trade-offs. Morgan Stanley expects China to ship 50,000 humanoid robots this year, up from thousands in 2025, though commercial deployments in factories and services remain scarce. Industry observers believe progress will be incremental rather than a sudden breakthrough, with the near-term focus on paid factory-floor deployments rather than household applications.