跳至正文 / Skip to content

一项任务,如何让设备与智能体协作。How devices and agents work on one task.

农业现场需要的,是感知环境、按权限控制设备,并确认动作后的变化。沿着这项任务,看 AIOS 如何连接现场能力、人的控制与结果反馈。A farm needs to sense its environment, control equipment with permission, and check what changed. Follow this task to see how AIOS connects field capabilities, human control, and feedback.

当前正通过具体项目验证可信协作,可用能力与接入范围按项目确认。Trusted collaboration is being validated through specific projects. Available capabilities and integration scope are confirmed per project.

菌菇培育 · 环境调控Mushroom cultivation · Environmental control

机制示意 · 非实时数据Mechanism illustration · Not live data
Reference photograph of mushroom cultivation, not an Instaon pilot site
行业资料图 · 非引态试点实拍Industry reference · Not an Instaon site
温度 · 湿度 · CO₂Temperature · Humidity · CO₂

同一片空间, 需要彼此配合的能力。One environment. Capabilities working together.

环境在变化, 设备需要配合。Conditions change. Devices need to work together.

菌菇培育需要关注温度、湿度与 CO₂。感知、通风、加湿和温控,共同服务于环境调节。Mushroom cultivation depends on temperature, humidity and CO₂. Sensing, ventilation, humidification and temperature control all contribute to the same task.

理解现场Read the environment
汇集温度、湿度与 CO₂ 等状态。Bring temperature, humidity and CO₂ into context.
共同响应Coordinate a response
让感知与设备控制围绕同一目标配合。Connect observations and device control around a shared goal.
三个观察面,同一项任务Three perspectives. One task.

试点进展:已实现环境感知、设备控制与培育结果反馈。协作图解释系统设计,不代表所有产品均已部署。Pilot progress: sensing, device control and cultivation feedback are implemented. The diagram explains system design; it does not imply deployment of every product.

图片来源与许可Image source & license

Mushroom Farm · Andrew Bossi · CC BY-SA 2.5 · 缩小与格式转换,页面裁切;非引态试点。Resized, converted and displayed cropped; not an Instaon pilot.

下一次任务,可以从已有经验开始。The next task can build on experience.

现场之外,模型、开发者与能力生态提供持续供给。只有经过验证的结果与纠正,才成为后续改进的依据。Beyond the field, models, developers, and the capability ecosystem provide ongoing supply. Verified outcomes and corrections inform subsequent improvements.

模型提供判断候选Models provide candidate judgments

天枢负责模型服务与模型包的管理和交付;灵枢、地枢及其他模型提供认知或空间理解。执行权限仍由运行产品校验。TianShu manages and delivers model services and packages. LingShu, DiShu, and other models provide cognitive or spatial understanding; runtime products still check execution permission.

新的能力可以参与New capabilities can participate

开发者通过供给入口组织、封装与测试能力;织界的职责是帮助需求发现能力,记录调用和贡献关系。Developers organize, package, and test capabilities through the supply entry. Orcha is responsible for discovery and records of calls and contributions.

反馈推动能力成长Feedback supports capability growth

元域负责保存经验、评估改进与能力版本。成功结果、失败原因和人的纠正,都需要验证后再进入复用。Realm is responsible for experience, improvement evaluation, and capability versions. Successes, failures, and human corrections need validation before reuse.

接入前,把能力边界说清楚。Make capability boundaries clear before connecting.

跨产品需要对任务、状态、权限与结果保持共同理解。这里是协作要求;具体 API、SDK 与发布状态以对应产品文档和项目验证为准。Products need a shared understanding of tasks, state, permission, and results. These are collaboration requirements; APIs, SDKs, and release status depend on product documentation and project validation.

身份与调用条件Identity and invocation conditions

说明谁提供能力、可在哪些条件下调用、哪些动作需要人的确认。Identify the capability provider, valid calling conditions, and actions that require human confirmation.

结果与失败反馈Results and failure feedback

返回实际执行状态、可核实的结果和失败原因,让下一步能够判断、恢复或交由人处理。Return execution state, verifiable outcomes, and failure reasons so the next step can assess, recover, or hand over to a person.

把你的场景或能力接进来。Bring your scenario or capability.

从一个明确任务开始,确认现场条件、可调用设备与结果标准,再讨论如何扩大协作。Start with a clear task. Establish field conditions, callable devices, and outcome criteria before expanding collaboration.