具身智能体与技能概述
目的与范围
具身智能体与技能层为 RoboFrame 提供编排、规划、安全校验与执行原语。它将自然语言意图、结构化 agent 计划与程序化工作流桥接为在 ROS 2 控制器上运行的、经校验且安全的物理操作。本页面是 agent 层的高层概览，串联 launch 构建器、运行时 agent 节点、技能目录、命令行接口与安全强制护栏。
深入的技术细节请参阅相应子页面：
来源：src/embodied_bringup/README.md:1-42, src/embodied_agent/README.md:1-20
架构总览：从自然语言到代码实体
具身系统通过一条由专用 ROS 2 节点组成的结构化流水线，把高层用户命令或 Hermes 驱动的工作流翻译为具体的机器人动作。
Title: Natural Language Space to Code Entity Space — Agent & Plan Lifecycle
graph TD
subgraph "Natural Language Space"
NL["NaturalLanguageCommand"] -->|Hermes / CLI| CLI["robot_skill_cli"]
end
subgraph "Code Entity Space"
CLI -->|"/embodied/plan_agent_command"| APN["agent_plan_node"]
APN -->|"/embodied/validate_skill"| SGN["safety_guard_node"]
APN -->|"/embodied/execute_skill"| SXN["skill_executor_node"]
SXN -->|"/embodied/execute_primitive"| PCN["PrimitiveCommand"]
end
来源：[src/embodied_bringup/README.md:1-60](https://atomgit.com/openeuler/IB_Robot/blob/openEuler-26.09/src/embodied_bringup/README.md#L1-L60), [src/embodied_agent/README.md:21-40](https://atomgit.com/openeuler/IB_Robot/blob/openEuler-26.09/src/embodied_agent/README.md#L21-L40)
7.1 具身流水线与 Agent 节点
embodied_bringup 包通过 embodied_pipeline.launch.py 协调运行时初始化，支持 hermes（默认）与 agent 两种入口模式 src/embodied_bringup/README.md:1-43。它管理 agent_plan_node、safety_guard_node 与 skill_executor_node 的生命周期，并在激活与运动相关的子图之前强制控制器就绪屏障 src/embodied_bringup/README.md:1-47, src/embodied_bringup/launch/embodied_pipeline.launch.py:53-88。
关于 launch 参数、节点拓扑与服务绑定的详细信息，见具身流水线与 Agent 节点。
来源：[src/embodied_bringup/README.md:1-47](https://atomgit.com/openeuler/IB_Robot/blob/openEuler-26.09/src/embodied_bringup/README.md#L1-L47), [src/embodied_bringup/launch/embodied_pipeline.launch.py:53-88](https://atomgit.com/openeuler/IB_Robot/blob/openEuler-26.09/src/embodied_bringup/launch/embodied_pipeline.launch.py#L53-L88)
7.2 技能目录与技能库
skill_library 包承载 skill_executor_node，作为执行核心把高层技能翻译为带版本的原语序列 src/skill_library/README.md:1-12。它将机器人配置文件与技能 manifest 编译为不可变的 RuntimeSnapshot 结构，其中包含严格的能力摘要、工作空间限制与机器人上下文 src/skill_library/README.md:34-50。
Title: Code Entity Space — Skill Resolution and Safety Validation Flow
graph TD
subgraph "Execution Gateway"
SKC["SkillCommand"] -->|"/embodied/execute_skill"| SXN["skill_executor_node"]
end
subgraph "Validation & Snapshot"
SXN -->|"/embodied/validate_skill"| SGN["safety_guard_node"]
SGN -->|"/embodied/get_skill_snapshot"| SCC["SkillCatalogCompiler"]
end
subgraph "Physical Control"
SXN -->|"/embodied/execute_primitive"| PCN["PrimitiveCommand"]
PCN -->|Trajectory Follow| RMC["ros2_control / controllers"]
end
来源：[src/skill_library/README.md:24-64](https://atomgit.com/openeuler/IB_Robot/blob/openEuler-26.09/src/skill_library/README.md#L24-L64), [src/safety_guard/README.md:24-50](https://atomgit.com/openeuler/IB_Robot/blob/openEuler-26.09/src/safety_guard/README.md#L24-L50)
关于技能 manifest、性能剖析、解析模板与自定义执行器的完整文档，见技能目录与技能库。
来源：[src/skill_library/README.md:1-64](https://atomgit.com/openeuler/IB_Robot/blob/openEuler-26.09/src/skill_library/README.md#L1-L64), [src/skill_library/skill_library/skill_executor_node.py:1-75](https://atomgit.com/openeuler/IB_Robot/blob/openEuler-26.09/src/skill_library/skill_library/skill_executor_node.py#L1-L75)
7.3 robot-skill CLI 与 Hermes
与具身运行时的交互通过 robot_skill_cli 工具套件和 Hermes 语音/生命周期集成框架管理。这些工具提供交互式命令行控制、工作流分发、原始 ROS 阻塞钩子以及语音驱动的对话管理，且不绕过核心安全与执行网关 src/embodied_bringup/README.md:64-78, src/embodied_agent/README.md:21-68。
详细用法、命令结构与语音桥接配置，见 robot-skill CLI 与 Hermes。
来源：[src/embodied_bringup/README.md:64-78](https://atomgit.com/openeuler/IB_Robot/blob/openEuler-26.09/src/embodied_bringup/README.md#L64-L78), [src/embodied_agent/README.md:21-68](https://atomgit.com/openeuler/IB_Robot/blob/openEuler-26.09/src/embodied_agent/README.md#L21-L68)
7.4 安全护栏
safety_guard 包通过 safety_guard_node 提供只读的预检校验服务（/embodied/validate_skill 与 /embodied/validate_primitive）src/safety_guard/README.md:1-80。它从 skill_library 网关同步精确的已验证快照，并在本地内存中强制严格白名单、工作空间边界、关节限位检查与分发绑定校验，不产生 ROS I/O 延迟 src/safety_guard/README.md:24-70。
校验规则、边界约束与错误代码的完整规范，见安全护栏。