/

EICPS：具身智能信息物理系统建模与仿真

EICPS: Modeling and Simulation of Embodied Intelligent Cyber-Physical Systems

本项目面向 智能无人艇（USV）等具身智能无人系统，提出 ℙ–ℰ–Φ 统一框架与 Hamiltonian 神经网络 (HNN) + 控制屏障函数 (CBF) + PDT 数学安全接口， 打通从 任务建模 → 物理建模 → 安全控制 → 实船验证 的全流程。

This project targets intelligent unmanned surface vessels (USVs) and related embodied unmanned systems. It proposes a unified ℙ–ℰ–Φ framework together with a Hamiltonian Neural Network (HNN) + Control Barrier Function (CBF) + PDT based mathematical safety interface, enabling a full pipeline from task modeling → physical modeling → safe control → real-world verification.

ℙ–ℰ–Φ Framework HNN · CBF · PDT USV Case Study

一、EICPS 理论框架

I. EICPS Theoretical Framework

1. ℙ–ℰ–Φ 三元组

1. ℙ–ℰ–Φ Triplet

ℙ 表示 建模过程（Problem / Process），ℰ 表示 具身空间（Embodied Space）， Φ 表示 数学安全接口（Safety Interface）。EICPS 将传统 CPS 控制、 具身智能与条令化任务表达统一在同一数学与工程框架中。

ℙ denotes the modeling process / problem formulation, ℰ the embodied space, and Φ the mathematical safety interface. EICPS unifies classical CPS control, embodied intelligence, and doctrine-like task descriptions within a single framework.

2. HNN + CBF + PDT 数学安全接口

2. HNN + CBF + PDT Safety Interface

HNN 学习保持结构的系统动力学，CBF 刻画安全域与约束集，PDT 提供时间一致性的决策调度。 三者级联构成 Φ 接口，使大模型 / RL 控制器在物理上可验证、在工程上可部署。

HNN learns structure-preserving dynamics, CBF encodes safety sets and constraints, and PDT provides time-consistent decision scheduling. Cascaded together as Φ, they project AI or RL controllers into physically verifiable and deployable actions.

3. 具身空间 ℰ：从状态空间到任务空间

3. Embodied Space ℰ: From State Space to Task Space

具身空间不仅包含系统状态，还显式编码任务约束、环境结构与条令规则， 为仿真训练、安全验证与行为评估提供统一载体。

The embodied space ℰ extends classical state space by explicitly encoding task constraints, environmental structure, and doctrine rules, serving as a unified carrier for simulation, safety verification, and behavior evaluation.

4. “皮层–脊髓”双层结构

4. “Cortex–Spinal” Dual-Layer Structure

上层 AI 控制器（RL / LLM / World Model）负责高层意图与策略； 下层 EICPS–Φ 负责毫秒级安全反射与物理一致性控制， 形成人工智能时代的“数字脊髓”。

Higher-level AI controllers (RL / LLM / world models) handle intent and strategy, while the lower EICPS–Φ layer ensures millisecond-level safety reflex and physical consistency, acting as a “digital spinal cord” for embodied systems.

二、EICPS–MSI 软件实现

II. EICPS–MSI Software Implementation

1. Modelica/MWORKS 物理建模

1. Physical Modeling with Modelica/MWORKS

基于 MWORKS 平台构建无人艇船体、水动力学、执行机构等物理模型，形成可复用的 Modelica 组件库，为 EICPS 动力学建模提供工程基础。

Physical models of USV hulls, hydrodynamics, and actuators are built on the MWORKS/Modelica platform, forming reusable component libraries that ground the EICPS dynamic modeling in real engineering practice.

2. HNN/CBF/PDT 模块化接口

2. Modular HNN/CBF/PDT Interfaces

将 HNN、CBF 与 PDT 封装为统一 Φ 接口模块，可与现有控制器无缝对接， 支持 C++/Python 等多语言实现，适配在线仿真与现场部署。

HNN, CBF, and PDT are encapsulated as a unified Φ interface module, which plugs into existing controllers and supports C++/Python implementations for both online simulation and field deployment.

3. 条令–任务–场景库

3. Doctrine–Task–Scenario Library

通过 ETL/ESML 语言对条令、任务与场景进行结构化描述，构建可复用的任务库， 支持多类型无人系统在统一具身空间中的训练与评测。

ETL/ESML languages describe doctrines, tasks, and scenarios in a structured way, forming reusable task libraries that support training and evaluation of various unmanned systems in a common embodied space.

4. USV–EICPS 教学与实验平台

4. USV–EICPS Teaching & Experiment Platform

以智能无人艇为核心案例，构建教学与科研一体化平台， 从理论推导、代码实现到仿真与实船试验，形成可复用的课程与实验资源。

Using intelligent USVs as a core case, EICPS provides an integrated teaching and research platform, covering theory, implementation, simulation, and sea trials, and producing reusable course and lab materials.

三、实验与应用案例

III. Experiments & Applications

1. 智能无人艇海缆检测

1. Intelligent USV for Subsea Cable Inspection

在复杂海况与港口环境下，对海底电缆开展自主巡检任务。 USV–EICPS 框架实现了从任务建模、路径规划、安全控制到实船测试的闭环。

Autonomous cable inspection missions are conducted under complex sea states and harbor conditions. The USV–EICPS framework closes the loop from task modeling and path planning to safety control and real-vessel experiments.

2. 复杂环境中的避碰与回避动作

2. Collision Avoidance in Complex Environments

针对窄水道、港区交汇等复杂场景，利用 CBF + PDT 实现对大模型 / RL 控制器输出的 安全投影，保证满足 COLREGs 及工程安全约束。

For narrow channels and congested harbor scenarios, CBF + PDT is used to safely project AI/RL controller outputs while respecting COLREGs and engineering safety constraints.

3. 虚实融合仿真对比实验

3. Hybrid-Reality Simulation Studies

在虚拟场景中快速迭代策略，并与实船试验数据进行对比，分析 EICPS 安全接口对 轨迹质量、控制稳定性与安全裕度的影响。

Policies are rapidly iterated in virtual environments and compared against real-ship data, analyzing how the EICPS safety interface affects trajectory quality, control stability, and safety margins.

4. 安全消融实验

4. Safety Ablation Studies

对比 “无 CBF / 无 HNN / 无 PDT” 等多种配置，定量评估各模块对碰撞率、 安全约束违例次数与控制输入平滑性的贡献。

Configurations such as “no CBF / no HNN / no PDT” are compared to quantify each component’s contribution to collision rate, safety-constraint violations, and control smoothness.

四、Top 10 代表论文

IV. Top 10 Representative Publications

更多论文请见：
🔗 Google Scholar： https://scholar.google.com/citations?user=vMLtpZYAAAAJ
🔗 ResearchGate： Zhiguo Zhou on ResearchGate

For a full and up-to-date list of publications, please visit:
🔗 Google Scholar: https://scholar.google.com/citations?user=vMLtpZYAAAAJ
🔗 ResearchGate: Zhiguo Zhou on ResearchGate

五、联系与合作

V. Contact & Collaboration

如果你对以下方向感兴趣，欢迎交流或合作：

We are happy to discuss and collaborate on:

  • 具身智能信息物理系统（EICPS）理论与软件实现
  • 智能无人艇 / 无人系统的安全控制与仿真
  • HNN / CBF / PDT 数学安全接口与大模型控制
  • Theory and software implementation of EICPS
  • Safe control and simulation of USVs and other unmanned systems
  • HNN / CBF / PDT based safety interfaces for large-model control

📍 Beijing Institute of Technology

🏫 北京市海淀区中关村南大街 5 号

🏫 5 Zhongguancun South Street, Haidian, Beijing

✉️ Email: your-email@bit.edu.cn

🐙 GitHub: spaitlab/EICPS

📖 Google Scholar: Zhiguo Zhou

📘 ResearchGate: Zhiguo Zhou