娱乐百分百

2026世界机器人大会｜破解世界模型落地的技术与数据难题_我的网站

legal high

A | 　　世界模型作为人工智能领域的前沿方向，是具备对物理世界因果理解与未来推演能力的智能系统，是机器人“看懂世界”和“改变世界”的关键技术底座。当大语言模型完成数字世界的认知跃迁，行业开始期待世界模型打通虚拟与现实的边界，让具身智能真正落地到生产生活中。　　8月21日，在2026世界机器人大会主论坛，中国电子学会世界模型专家委员会正式成立，王坚院士受聘担任主任委员，并主持“世界模型到服务人类的现实距离”主题圆桌，八位来自产业一线的专委会委员共同剖析技术路线、产业痛点与落地路径。

2026年世界机器人大会图片来自大会官方供图

　　本次圆桌汇聚大晓机器人王晓刚、奥比中光黄源浩、无界动力张玉峰、跨维智能贾奎、智澄AI胡鲁辉、飞渡科技宋彬、蚂蚁灵波科技沈宇军、极佳视界朱政，覆盖传感器硬件、机器人本体、空间数字孪生、AI基础模型等产业链关键环节。对话从技术路径切入，厘清世界模型与VLA、大语言模型的本质差异。　　机器人是AI实现闭环落地的最好场景　　在物理世界中，谈及人形机器人，就绕不开通用人工智能向物理世界赋能这一命题，“机器人，是人工智能实现闭环、完成落地的最佳场景之一。”智澄AI创始人胡鲁辉表示，通用人工智能实现闭环落地，其核心正是世界模型与机器人的深度结合。　　大语言模型、多模态大模型大多停留在数字空间的信息处理与生成，输入输出均以文本、图像等数字媒介为主，难以和物理世界产生真实交互反馈；而机器人作为实体载体，打通了“感知—理解—决策—行动—环境反馈”的完整链路，让AI真正走出屏幕接入现实物理世界。　　“相较于大语言模型、VLA模型，世界模型的能力上限高出很多，当然，它对数据的需求也十分庞大。”跨维智能创始人贾奎认为，理想的世界模型不仅能生成机器人末端的有效执行轨迹，还可以预测机器人、其他物理智能体作用于环境之后，环境会发生怎样真实的动态改变。 New York-traded West Texas Intermediate (WTI) settled down 97 cents, or almost 1%, at $98.53. The intraday low was $95.17, a bottom not seen since the week ended April 8. In just two sessions, the US crude benchmark had lost almost $15 or 14%.,Citigroup says WTI could collapse to $65 a barrel by the end of this year and slump to $45 by end-2023 if a demand-crippling recession hits.,London-traded Brent crude, settled at $100.69, down $2.08 or 2.02%. Earlier, it broke below the $100 mark the first time since April 25, hitting a session low of $98.59. Like WTI, Brent has also lost almost $15, or 13%, over two days of trading.,West May Use Proposed Russian Oil Price Cap Scheme for Gas Price, Von Der Leyen SaysYesterday, 12:20 GMT,A cluster of economic data of late has suggested that the United States may be headed for an economic slowdown - possibly a recession - as the Federal Reserve embarks on the most aggressive rate hikes in a generation to tame inflation roaring at 40-year highs.,A closely-watched gauge of the US services sector fell to its lowest in 20 months in June.,Separately, the US Labor Department signaled that the red-hot labor market may be starting to cool. Job openings, as measured by its monthly survey, fell in May, to a level of 11.254 million that is still high by historical comparison.,The job openings data came ahead of Friday’s more important June non-farm payrolls report, which is expected to show a smaller jobs growth compared with May. Economists say some 268,000 payrolls were probably added last month - versus the 390,000 in May - holding unemployment at 3.6% for a third straight month. A jobless rate of 4% or below is seen by the Federal Reserve as full employment.,“While our view remains that higher consumer prices are required to balance the oil market this summer, we acknowledge that significant and large shocks continue to distort fundamentals,” Goldman Sachs said in an energy market outlook.,Blockage of Russian Bond Payments to Erode Dollar Dominance, French Economist Says27 June, 15:29 GMT,A rocketing dollar ahead of more rate hikes by the Federal Reserve, or Fed, has also shaken the roots of the energy rally seen since the start of the Russia-Ukraine conflict in February and the resulting sanctions on Moscow.,The Dollar Index, which pits the greenback against six major currencies, continued its ascent from overnight, leaping to above 107, its highest since December 2002. The dollar has rallied with few stops since November last year on bets of aggressive rate hikes by the Federal Reserve, or Fed, which has just started delivering on those expectations.,The drone of recession talk is expected to get louder across the United States after the Atlanta division of the Federal Reserve forecast last week a second straight quarter of economic decline for the year. The Commerce Department officially reported a 1.6% economic contraction in the first quarter.,The Federal Reserve, in minutes of its June policy meeting released on Wednesday, said it saw a real danger of high inflation becoming entrenched in the US economy and the only way to balance runaway prices with growth was to have appropriate interest rate hikes.,The central bank is expected to push ahead with another three-quarter percentage point interest rates hike this month, potentially taking key lending rates to between 2.25% and 2.5% range from the 0 - 0.25% range they stood at in February.。　　现场专家普遍认为，世界模型集理解、生成、预测能力于一体，可以在执行动作前推演多种行为带来的未来状态。

B | 当前行业已经形成多条并行技术路线：有的企业走“空间世界模型+强化学习”，在高维隐空间做时空推理，不追求复刻原子级物理细节，效仿人类依靠“大致正确”的认知完成任务；有的企业则聚焦构建三维几何物理正确的可交互环境，将世界模型视作生成式AI“皇冠上的明珠”。

C | 　　飞渡科技联合创始人宋彬梳理了当前行业中的技术流派，并归纳出世界模型的四大核心能力：理解生成世界、模拟仿真推演、因果推理、任务规划执行。他打比方称，世界模型底座如同九年义务教育，先搭建通用能力，再面向垂直场景做专项训练。　　世界模型距离落地还有多远？　　从产业视角来看，世界模型正迎来发展拐点。极佳视界朱政分享了当前行业的冷暖变化，三年前资本市场尚难以理解世界模型的价值，如今行业热度快速提升，公司在12个月内估值增长了50倍。但热度之下更需要理性，蚂蚁灵波科技沈宇军提醒，单纯追求逼真的视频生成没有意义，世界模型的价值最终必须落地于机器人真实作业能力，脱离物理实操的时序建模终究是空中楼阁。　　此外，与会专家们还进一步抛出核心命题：若要真正服务人类，世界模型还需要补齐哪些能力，距离大规模落地还有多远？　　大晓机器人董事长王晓刚将具身世界模型拆解为“生成智能、物理智能、认知智能”三大支柱，其中物理智能是当前最大短板。模型可以生成画面优美的视频，却未必掌握真实世界的物理规则，当前行业极度稀缺包含几何、相机位置姿势等第一视角的交互数据。　　世界模型的落地是一套围绕模型、数据、传感器硬件、机器人本体、安全体系建设的系统性工程。无界动力创始人张玉峰提出务实落地思路，“工业场景练技能、商业场景练泛化”，优先在工业、连锁商业等可控场景跑通商业价值，依靠真机部署形成数据闭环迭代，再逐步挑战家庭这种高复杂度场景。

D | 同时人机共生场景下，不能完全依赖模型保障安全，必须配套独立的安全冗余与熔断机制。　　工程层面的矛盾同样突出。沈宇军指出机器人独特的现实约束：机器人无法像车辆一样边运行边发电，高算力带来的功耗压力，成为智能能力向终端迁移的一道硬门槛。宋彬则表示，工业场景追求长时序、高确定性输出，多物理场耦合推演仍是巨大难题，乐观估计仍需要3-5年持续攻坚。胡鲁辉补充道，人形机器人进入家庭，除技术难题外，隐私、伦理同样需要用技术手段去应对解决。　　此外，高质量数据供给不足，是当前行业面临瓶颈的原因之一。与互联网唾手可得的海量文本语料不同，具身智能真实交互数据积累几乎从零开始。实验室剧本化采集的数据效果有限，高价值数据需要来自真实生产场景，既要采集成功样本，失败案例同样具备极高训练价值。黄源浩呼吁行业建立统一的数据标准，实现多传感器毫秒级时间同步，推动数据跨本体、跨模型复用。　　面对高昂的数据采集成本，行业探索多元破局路径：一方面深度联动制造业、服务业，从真实作业过程中沉淀真实数据；另一方面依靠世界模型自身的仿真生成能力，合成现实中罕见的极端样本，降低对真机采集的依赖。贾奎提出了Physicaltoken经济学理念，采用“人类真实经验预训练+生成仿真合成数据后训练”的组合路线。宋彬则呼吁政府、行业联盟应当共建开放基础数据集，推动数据要素市场化流转。沈宇军则提出，不能盲目追求数据规模，厘清任务所需精度下限、优化数据分布，远比堆砌数量更加关键。