锦绣未央

China pilots "quiet communities" to curb urban noise, disputes_我的网站

一 |     当AI新型设备层出不穷、具身智能加速落地，未来的AI操作系统及其软件栈将呈现何种形态？在WAIC 2026安谋科技发布会的压轴环节，安谋科技市场及生态副总裁梁泉给出了一个判断：操作系统正在从传统的“工具型”向“智能体”演进。    

    An aerial drone photo taken in October 2025 shows a view of the Dongshan pocket park in Guangzhou, south China's Guangdong Province. (Xinhua)When 72-year-old Shi Jikun opened his windows at home in Yuxi, southwest China's Yunnan Province, he used to hear the repeated thud of balls and blowing of whistles from an indoor sports venue near his residential community.
"During the day, I didn't dare open the windows," Shi said, recalling years when the noise disrupted his daily routine.
The sports venue has since undergone two rounds of soundproofing. Sound-absorbing materials were installed on its walls, while double-glazed soundproof windows were fitted along the side facing the residential area.
The noise has eased. "The neighborhood is quiet now, making everyday life much more comfortable," Shi noted.
Shi's experience is part of a broader move underway across China to tackle one of the less obvious costs of urbanization: the growing amount of noise packed into increasingly dense neighborhoods.
As cities expand, the likes of homes, roads, businesses, construction sites and recreational spaces are often crammed together. China has, notably, been testing a neighborhood-level response to this challenge via the development of "quiet communities," combining physical upgrades with local management and mediation.
This initiative was launched as a national pilot project in May 2023. By the end of 2025, a total of 3,272 quiet communities had been built nationwide, benefiting more than 2.4 million households, according to the Ministry of Ecology and Environment.
This approach now also has a stronger legal footing. On Saturday, China's Ecological and Environmental Code came into force, with a dedicated section on noise pollution prevention and control.
The scale of the problem is reflected in noise complaints. In 2025, cities at the prefecture level and above received about 6.83 million noise-related complaints, with social-life noise accounting for 73.2 percent.
Noise from neighbors, equipment malfunctions and disturbances from square dancing, a popular group exercise in China often held by middle-aged and older residents in public spaces, were among the most common issues raised.
China's quiet-community program is designed to address such problems before they become recurring disputes. "Quiet communities do not mean that there should be no sound," said Lu Lu, deputy head of the noise management division at the Ministry of Ecology and Environment.
The goal, Lu said, is to create a comfortable and harmonious living environment while balancing everyday activity with residents' need for quiet.
The solutions vary from place to place. In south China's Shenzhen, some newly built residential communities are incorporating noise control at the planning and design stage. Measures include low-noise pavements, vegetation buffers, separated pedestrian and vehicle routes, and automatic noise-monitoring equipment.
Older communities are using renovation projects to address noise concerns. In Xiaoshan District in east China's Hangzhou, Huiyu garden community replaced aging elevators and repaired drainage grates and other public facilities as part of its quiet-community program.
But physical upgrades alone cannot settle conflicts between neighbors.
At Fenglin Shunhe community in Weihai, eastern China's Shandong Province, residents, property managers, community officials and square-dancing groups agreed to end dancing at 8:30 p.m. The main lights in the square switch off at that time, signaling that it is time for the dancers to leave.
The key is to respond quickly to complaints and give residents and activity groups a way to negotiate solutions, said Jiang Chunyan, head of the neighborhood committee.
In another community in Shenzhen, property managers have gone further by treating renovation noise as a process to be managed proactively rather than a problem to be addressed after complaints arise.
Renovation work is divided into nine stages and mapped on a visual noise-management chart. Property managers use the chart to carry out targeted inspections as work progresses. Contractors are also required to work only during designated hours and under fully enclosed conditions.
The program still faces gaps in terms of regional coverage and long-term maintenance. Quiet communities remain concentrated in a few cities, while older neighborhoods and parts of central and western China have fewer projects. Some communities also face difficulties maintaining noise-control measures after initial upgrades are completed.
The Ministry of Ecology and Environment plans to incorporate quiet-community development into China's noise pollution prevention and control work for the 2026-2030 period. It has called for regular inspections, dynamic monitoring and follow-up visits to ensure that measures remain effective.
Zheng Shanshan, an official responsible for ecological affairs in Hangzhou's Xiangfu Subdistrict, said the next step should extend noise management from individual communities to larger areas through tools such as sound-environment maps.
"Quiet communities are only a starting point," Zheng said. "We hope to create more quiet spaces across the city."
。

二 | 作为这一趋势的回应，安谋科技牵头联合多家合作伙伴，正式发起开源AIOS联盟，共同探索新一代AI产品的基础设施。

三 |          交互范式革命：AIOS的四个核心驱动力          回顾各类计算平台发展历史，每一次计算载体跃迁，都源于“新交互×新计算”的双重驱动。PC时代用鼠标键盘，移动时代转向了多点触控，交互方式决定了人与设备打交道的方式。而AI时代，这一规律仍在延续。         梁泉将驱动AIOS变革的核心因素归纳为四个维度：          第一，交互范式升级。语音交互、生成式UI将成为常态——用户面对的将不再是固定界面，而是根据交互过程实时生成的界面。更具颠覆性的是“具身互动”：当人与机器人、机器狗打交道时，交互不再是回合制对话，而是持续感知的自然过程。         梁泉举了一个生动的例子：“你跟一只机器狗打招呼，一定要唤起它的名字吗？不是的。可能它在那边跑，你叫一声，它回头看你一眼，就应该停止了。

四 | ”这种自然交互方式，将深刻驱动AI计算平台和系统软件的大幅变革。         第二，系统架构迭代。异构计算已成为共识，但更关键的演进发生在Agentic AI的落地层面：从工具的多轮稳定调用，到长短时记忆的引入，再到基于数据流驱动感知的落地，以及任务的智能编排。         梁泉特别提到了近期业界流行的一个概念——“Orchestration”，即如何将AI任务更好地编排到虚拟化环境或实际本地运行环境中。

五 | 这一能力的成熟，将成为AIOS区别于传统操作系统的核心标志之一。

六 |          第三，端侧AI能力突破。过去业界普遍认为端侧AI永远落后云端两三代，但当端侧模型能力逐渐跨过可用性门槛后，“代际差”正在缩小。         梁泉指出，如今云端应用的Agentic能力，在端侧模型乃至MoE模型上已能达到很可用的状态；配合芯片计算架构的异构迭代，端侧AI的突破已是“近在眼前的事情”。安谋科技在IP研发和软硬件协同上的大量投入，正是在加速这一进程。         第四，端云协同深化。

七 | 当前大量AI计算发生在云端，但随着端侧能力提升，计算将更多在本地完成。

八 | 未来的模式将不再是简单的“端侧采集、云端处理”，而是端云协同更加丰富、更加动态。         重新定义AIOS：交互是核心，模型是中枢          基于以上趋势判断，梁泉给出了安谋科技对AIOS的定义：一个基于感知的生成式交互和具身互动的系统，其中交互被置于核心位置。         他解释道，如果AIOS只是作为一个计算基础设施静静地运行，那么基于现有Linux或服务器系统的架构已经在云端和边缘侧广泛落地。但如果我们要讨论的是新型边缘设备或具身产品——那些能与人持续互动的设备——交互就是整个系统的灵魂。

九 | “大家回忆PC和手机，日常打交道最多的就是交互。所以我们将交互的演进、交互范式的迭代放在系统的核心。”          在这一框架下，AI模型能力成为系统中枢，AI任务编排、传感与运控成为关键支柱。

十 | 梁泉还特别提到了对传感层面的深入思考：如何让传感真正实现自然、无缝的交流？他引用了OpenAI近期发布的GPT-Live，已能实现非常自然的语言流交互。

十一 | 但具身设备的交互更为复杂——机器人可能通过视觉感知用户的动作、情绪变化，这种持续感知与交互对系统的要求远超单纯的语言对话。         此外，记忆系统同样不可或缺，包括长短时记忆和上下文管理，同时还需兼顾隐私和规则约束。         “上面的AIOS基础设施都会有完整的重构过程，而基于现成的Linux系统可以很好地搭建这样的能力。”梁泉表示，正是这套OS产品和技术的定义，驱动了开源AIOS联盟的发起。         AIOS架构揭秘：四大引擎构筑智能体底座          演讲中，梁泉展示了一张囊括AIOS核心模块的架构图，并逐一拆解了其中的四大引擎。         模型引擎，负责模型推理和端云协同路由，贯穿从感知到交互的全过程。端侧需运行多个模型，包括语音识别、TTS文本转语音等。更为关键的是，处理自然交互不仅需要能把语言模型或多模态模型跑起来的技术架构，更需要融入交互流中的互动结构——这是新型AIOS区别于传统系统的重要模块。         Agentic引擎，涵盖多轮工具调用和任务编排，尤其强调“Loops”的概念：如何将自动任务或循环任务与复杂的终极目标相结合，将一个简单语句背后隐藏的复杂规划和拆解，落地为可执行的动作或可分解的任务。梁泉特别强调，Agentic引擎中非常重视隐私和安全特性，确保那些看似可自主执行的关键动作任务，始终在合理的边界内运行。         记忆引擎与数据流引擎，后者的重要性在具身场景中尤为突出。数据流引擎将视频流和音频流深度融合，支撑混合输入——实时的、同时的语音、视频、文本输入。梁泉判断，这些输入未来可能以全模态模型的方式呈现，也可能在早期以工程化方式通过实时处理逐步落地。但无论哪种路径，DataFlow引擎都将让异构系统以最高效的方式（如Zero Copy）处理整个数据流向，成为贯穿新型AIOS的“中枢通道”。

十二 |          那么哪种计算架构更适合AIOS呢？梁泉认为，这套AIOS最终将运行在所有搭载Arm CPU的设备上。因为，Arm CPU在端侧的占比极高，同时系统还将支持Arm Immortalis GPU、安谋科技自研的周易NPU以及众多客户的加速单元，最终以升级革新的交互方式服务于各类新型AI设备。这也是为什么安谋科技会牵头发起开源AIOS联盟的一大关键原因。         开源AIOS联盟：不止于软件，覆盖全链条开放          “我们的目标是什么？我们定义它是一个开源的AIOS联盟。”梁泉明确表示，无论是AIOS软件本身、AIOS核心的AI模型层面，还是AIOS落地的原型或概念产品，联盟都希望做到开源和开放。         技术层面，联盟将与合作伙伴共同研发可能成为核心模块的新型技术，推动技术、代码、产品文档、设计原型等全方位开源开放。

十三 | 落地层面，联盟将以实际产品覆盖端边侧大量设备及云端互动基础设施，并开展联合市场推广。

十四 |          梁泉特别介绍了安谋科技旗下的极术社区。作为实验载体，极术社区今年已发布了JishuShell（极术秀）工具，今年四季度还将推出知识库产品并计划开源相关动作。         “它看上去是一个简单的命令行工具，实际上背后是小型操作系统，或者说小型系统层面管理器。

十五 | ”它会安装能管理任务、编排任务的服务器级编排机制（针对边缘优化），并将任务分配到Native环境或Micro VM虚拟机中。“虽然我们用一行命令看似安装了一个工具，但背后是操作系统的雏形。”梁泉对JishuShell进一步解释道。         作为一个开源和开放的联盟，AIOS联盟开放的范围并不局限于技术方案。梁泉透露，团队还做了端侧模型的Benchmark套件，用于实测端侧模型在Agentic能力和长时任务上能否达到云端API的水准，该套件将在本次会议后逐步开放。         在研究合作层面，安谋科技正与港科大物理AI实验室合作发布VLA模型，并与清华智能产业研究院开展模型相关合作。未来，这些研究成果和工程成果，都将逐步开放给联盟和市场。         联盟已获多方响应，下一步将加速OS落地          梁泉透露，AIOS联盟一经推出便已获得产业链多方的积极响应。

十六 | 首批发起的合作伙伴中，既包括芯发科技、紫光展锐、辉羲科技、爱芯元智、此芯科技、瑞芯微、全志科技、芯擎科技、黑芝麻智能、后摩智能等深度参与的芯片客户，也包括做机器狗、边缘AI盒子的设备与方案商，以及端侧模型领域的领军者面壁智能等研究机构。         “发起AIOS联盟只是第一步动作，”梁泉在表示，“我们希望尽快把AIOS真正带给大家，也希望能尽快通过AIOS带来更多前沿的AI产品和设备，让这些产品和设备走向大家的生活，让更多AI的能力走进大家的生活。”          编辑：芯智讯-浪客剑。

Current article:http://ek8a.foucaorengtunzhuaikuowang.cyou/3sf5/ccihs.html

Published on:20:42:40