Key Problems of Embodied Intelligence Research: Autonomous Perception, Action, and Evolution
-
摘要: 具身智能强调大脑、身体及环境三者的相互作用, 旨在基于机器与物理世界的交互, 创建软硬件结合、可自主学习进化的智能体. 当前, 机器学习、机器人学、认知科学等多学科技术的快速发展极大地推动了具身智能的研究与应用. 已有的具身智能文献更多从技术和方法分类的角度入手, 本文以具身智能在研究和应用过程中面临的关键挑战为角度切入, 分析具身智能研究的一般性框架, 围绕具身感知与执行、具身学习与进化两个方面提出具体的研究思路, 并针对其中涉及的关键问题详细梳理相关技术及研究进展. 此外, 以移动机器人、仿生机器人、平行机器人三方面应用为例, 介绍具身智能在感知与理解、控制与决策、交互与学习等方面给实际机器人系统设计带来的启发. 最后, 对具身智能的发展前景进行展望, 探索虚实融合数据智能、基础模型与基础智能、数字孪生与平行智能在其中的重要作用和应用潜力, 希望为相关领域学者和从业人员提供新的启示和思路. 论文相关项目详见
https://github.com/BUCT-IUSRC/Survey__EmbodiedAI .Abstract: Embodied intelligence aims to create intelligent entities that combine software with hardware and can learn and evolve autonomously based on the interaction between machines and the physical world, by emphasizing the interaction among the brain, body, and environment. Currently, the rapid development of multidisciplinary technologies such as machine learning, robotics, and cognitive science has greatly promoted the research and application of embodied intelligence. Existing literature mainly focus on technical and methodological classifications of embodied intelligence. From a new perspective, this paper starts from the key challenges involved in the research and application of embodied intelligence. By analyzing the general framework of embodied intelligence research, specific research ideas are proposed for the two kernel stages including embodied perception and execution, as well as embodied learning and evolution. Accordingly, detailed explanations of relevant technologies and research progress related to these key issues are provided. In addition, typical applications in mobile robots, bionic robots, and parallel robots are exploited to illustrate how embodied intelligence inspires practical robot systems design in terms of perception and understanding, control and decision-making, as well as interaction and learning. Finally, the prospects of embodied intelligence is discussed, considering the importance and application potential of virtual-real fusion data intelligence, basic modeling and foundation intelligence, digital twins and parallel intelligence. This paper is expected to provide new inspirations and ideas for scholars and practitioners in related fields. The associated project is available athttps://github.com/BUCT-IUSRC/Survey__EmbodiedAI .-
Key words:
- Embodied intelligence /
- perception and execution /
- learning and evolution /
- robot systems
-
多个独立的智能体通过信息交互以及多种方式组成多智能体系统[1−6], 旨在解决单个智能体无法解决的大规模复杂性任务. 多智能体系统能够完成在复杂网络下的信息传递, 已广泛应用在诸多重要的实际场景, 如航天器领域[7]、无人驾驶飞行器[8]和水下车辆系统[9]等. 特别地, 多智能体系统的跟踪控制策略[10−11]一直是进一步探索的热点研究方向. 例如, 文献[12]研究针对多智能体系统网络化预测PID的控制问题来达到输出信号一致性. 文献[13]针对带有非线性扰动的多智能体系统, 提出基于神经网络的自适应控制策略来实现共识控制目标. 然而, 智能体之间的信息交互严格依赖网络环境, 特别是存在大量智能体时将同时占用多个网络通信渠道, 不可避免地导致沉重的网络负担, 这一现象亟需解决.
近年来, 为缓解信息传递渠道上的通讯压力, 学者们提出事件触发策略[14−18]使控制器以非周期的方式更新. 文献[19]提出针对多智能体系统的事件触发分布式控制策略, 包括固定阈值策略、相对阈值策略和切换阈值策略. 根据这一概念, 相继研究出多种新颖的事件触发机制. 例如, 动态事件触发机制、状态触发机制和记忆事件触发机制等. 其中, 状态触发机制[20−21]引起广泛关注. 特别地, 文献[21]首次提出基于状态触发机制的非线性多智能体系统自适应一致性控制方案, 设计的事件触发机制首先将采样信号值与系统真实值构造出采样误差, 随后转换得到带有触发信号的同步误差, 并将之设计到控制输入信号中来缓解控制器−传感器渠道上的通讯压力. 值得注意的是, 现存文献中状态触发机制的阈值条件是动态变化的, 并且会采用根据分解的方式将产生的采样误差值的平方项化为常数项的方法判断下一次采样时刻. 需要进一步指出的是, 该阈值条件的设计方法严格依赖于稳定性条件, 从而限制阈值条件设计的灵活性. 但在实际系统的真实状况中, 状态信息通常不可测量, 因而会导致控制方案实施的准确性. 因此, 提出状态观测器来实现对原系统重构, 从而满足反馈控制的需要. 所以, 在大规模实际系统的控制运行中, 首先需要解决在未知状态不可测情况下系统通讯渠道负担重的问题, 并且在信息传递过程中多个通讯链路上的资源节约问题同样值得注意. 例如, 控制器−执行器环节、传感器−控制器环节、智能体与智能体之间的通讯渠道等. 因此, 如果实现同时在多个通讯渠道上节约通讯资源势必会大幅度降低整体控制系统通讯带宽占用率.
值得关注的是, 一些非线性因素[22−28]可能会导致系统性能下降或系统抖震现象发生. 模糊逻辑系统或神经网络是处理非线性项的常用近似工具. 然而在大多数情况下, 模糊逻辑系统或神经网络的逼近能力是有限的, 并且其逼近效果取决于模糊规则或神经网络节点的数量. 因此, 系统中非线性因素的存在会导致系统不稳定. 为解决上述情况并保证系统的实时性能, 文献[29]提出规定性能控制方法, 通过将跟踪误差约束在预定的范围内实现对系统瞬态性能和稳态性能的保证. 文献[30]针对非线性多智能体系统, 设计新颖的规定性能转换函数. 文献[31]利用误差转换方法和规定性能控制策略设计分布式自适应控制器来保证控制目标的有效实现. 因此可以看出, 规定性能控制方法在确保系统性能方面十分有效, 并且当发现性能指标无法满足时, 该机制可以及时采取相应的容错措施以提高系统的可靠性和安全性.
因此, 考虑到非线性多智能体系统中通讯资源负担重以及现存状态事件触发机制的阈值设计条件具有一定地局限性的双重问题, 本文展开基于混合双端事件触发机制的协同控制策略研究, 旨在提升现存状态事件触发机制阈值设计的灵活性且改善多通讯渠道的通讯压力状况, 进一步拓展多智能体系统一致性控制策略的多样性.
本文主要贡献如下:
1) 与现存结果相比[21], 所提出的状态触发机制的阈值条件可以在不使用杨氏不等式的情况下直接设计, 从而减少现有控制方案的缩放次数和保守性. 并且首次使用估计状态进行采样, 扩展了状态触发机制的应用范围. 基于新的状态触发机制与控制器触发机制, 构造新的双端分布式触发框架, 在较少的参数设计限制性下, 具有更小的通信压力.
2) 基于规定性能技术特性, 提出的自适应控制方案既减少系统不稳定情况发生的概率, 又保证多智能体一致性任务的精确度. 此外, 所设计的分布式观测器仅依赖于相对输出信息进行反馈调节来解决状态不可测问题, 具有良好的可扩展性和灵活性.
本文的组织结构安排如下. 第1节给出本文工作所需的预备知识; 第2节介绍模糊状态观测器的设计过程; 为获得预期的控制目标, 第3节提出自适应分布式控制器的设计方法、稳定性及芝诺行为分析; 第4节通过一个实际仿真例子证明所提出策略的有效性; 第5节给出本文的结论, 并对未来进行展望.
1. 预备知识
1.1 图论
针对非线性多智能体系统, 智能体之间的信息传输关系需要清晰描述. 考虑一个有向图来表明多个智能体之间的关系. 首先, 定义有向图为 $ \bar{{\cal{G}}}=({{\cal{V}}},\;\acute{{\cal{E}}},\;\bar{{\cal{A}}}) $, 其中, $ {{\cal{V}}}=\left( {1},\;\cdots,\;{N}\right) $ 为节点集合, $ \acute{{\cal{E}}}\subseteq{{\cal{V}}} \times {{\cal{V}}} $ 为边集合. 定义邻接矩阵为 $ {\bar{{\cal{A}}}}=[a_{h,\; \ell}] \in \bf{R}^{N\times N} $. 与此同时, 节点$ \ell $到节点 $ h $形成的边表示为 $ \left({{\cal{V}}}_{h},\;{{\cal{V}}}_{\ell}\right) \in \acute{{\cal{E}}} $. 当节点$ \ell $可以传输信息到节点$ h $, 即可以得到$ a_{h,\;\ell}>0 $. 否则, $ a_{h,\;\ell}=0 $. 此外, $ \check{{\cal{D}}} $ 表示度矩阵, 其中 $ \check{{\cal{D}}}=\hbox{diag}\left\{d_{1},\;\cdots,\;d_{N}\right\} $ 以及 $ d_{h} = {\sum\nolimits^N_{\ell=1}}a_{h,\;\ell} $. $ {{\cal{L}}} $ 表示拉普拉斯矩阵, 其定义为$ {{\cal{L}}}=\check{{\cal{D}}}-\bar{{\cal{A}}} $.
假设 1[32]. 将有向图$ {\cal{G}} $看作一个生成树, 需要保证任意一个节点到根节点都至少存在一条有效传输路径, 并将领导者节点$ {\bf{0}} $ 视为生成树的根.
引理1[33]. 定义$ {\cal{B}}=\hbox{diag}\left\{b_{h}\right\}\in \bf{R}^{N\times N} $, 并且使得$ b_{h}>0 $. 随后, 可以得到 $ {\cal{L}}+{\cal{B}} $是非奇异的.
1.2 问题形成
1) 智能体的动态模型. 对于$ h=1,\;2,\;\cdots,\;M $和 $ g=1,\;2,\;\cdots,\;n-1 $, 定义第$ h $个智能体的动态模型为
$$ \begin{aligned} \left\{ \begin{aligned} &\dot{x}_{h,\;g}=x_{h,\;g+1}+\zeta_{h,\;g}(\bar{x}_{h,\;g})\\ &\dot{x}_{h,\;n}=u_{h}+\zeta_{h,\;n}(\bar{x}_{h,\;n})\\ &y_{h}=x_{h,\;1} \end{aligned} \right. \end{aligned} $$ (1) 其中, $ {\bar x_{h,\;g}} = [x_{h,\;1},\;x_{h,\;2},\;\cdots,\;x_{h,\;g}]^\text{T}\in {{\bf{R}}}^{g} $ 和 $ {\bar x_{h,\;n}}= [x_{h,\;1},\;x_{h,\;2},\;\cdots,\;x_{h,\;n}]^\text{T}\in {\bf{R}}^{n} $ 代表状态向量. $ u_{h} $和$ y_{h} $分别代表控制输入信号和输出信号. $ \zeta_{h,\;g}(\bar{x}_{h,\;g}) $和$ \zeta_{h,\;n}(\bar{x}_{h,\;n}) $表示未知的光滑非线性函数.
2)控制目标. 针对非线性多智能体系统, 设计一个带有规定性能机制的自适应混合双端事件触发跟踪控制策略来保证如下的两个控制目标:
a) 保证闭环内所有信号都是半全局一致最终有界的;
b) 使跟随者的输出轨迹和领导者的输出轨迹保持一致.
引理2[33]. 定义$ \bar{s}_{.1}=({s}_{1,\;1},\;{s}_{2,\;1},\;\cdots,\;{s}_{N,\;1})^\text{T} $, $ \bar{y}=(y_{1},\;y_{2},\;\cdots,\;y_{N})^\text{T} $ 和 $ \bar{y}_{r}=({y_{r},\;y_{r},\;\cdots,\; y_{r}})^\text{T} $. 三者满足如下关系:
$$ \begin{align} \begin{aligned} ||\bar{y}-\bar{y}_{r}||\le\frac{||\bar{s}_{.1}||}{{\bar{\sigma}}({\cal{L}}+{\cal{B}})} \end{aligned} \end{align} $$ (2) 其中, $ {\bar{\sigma}}({\cal{L}}+{\cal{B}}) $为矩阵$ {\cal{L}}+{\cal{B}} $的最小奇异值.
1.3 模糊逻辑系统
使用模糊逻辑系统理论近似严格反馈多智能体系统中存在的未知非线性函数. 考虑如下模糊逻辑系统:
$$ \begin{aligned} \tilde{y}(x)=\frac{{\sum\limits_{\flat=1}^r}\bar{y}_{\flat}{\prod\limits_{\hbar=1}^{\check{n}}}{\bar{\mu}_{F^{\flat}_{\hbar}}}(x_{\hbar})}{{\sum\limits_{\flat=1}^r}[\prod\limits_{\hbar=1}^{\check{n}}{\bar{\mu}_{F^{\flat}_{\hbar}}}(x_{\hbar})]} \end{aligned} $$ (3) 其中, $ \bar{y}_{\flat}=\max_{y\in{\bf{R}}}{\bar{\mu}_{G^{\flat}}}(\tilde{y}) $.
定义模糊基函数为
$$ \begin{aligned} \psi_{\hbar}(x)=\frac{{\prod\limits_{\hbar=1}^{\check{n}}}{\bar{\mu}_{F^{\flat}_{\hbar}}}(x_{\hbar})}{{\sum\limits_{\flat=1}^{r}}[\prod\limits_{\hbar=1}^{\check{n}}{\bar{\mu}_{F^{\flat}_{\hbar}}}(x_{\hbar})]} \end{aligned} $$ (4) 并且, 定义向量 $ \zeta^\text{T}=[\bar{y}_{1},\;\bar{y}_{2},\;\cdots,\;\bar{y}_{n}]= [\zeta_{1},\; \zeta_{2},\;\cdots, \zeta_{n}] $ 和 $ \psi(x)=[\psi_{1}(x),\;\cdots,\;\psi_{n}(x)]^\text{T} $. 随后, 进一步表示模糊逻辑系统为 $ \tilde{y}(x)=\zeta^\text{T}\psi(x) $.
引理3[33]. 对于$ \varepsilon>0 $, 表示模糊逻辑系统为
$$ \begin{align} \underset{x\in\varpi}\sup\lvert \zeta(x)-{\eta}^\text{T}\psi(x) \rvert \le\varepsilon \end{align} $$ (5) 其中, $ \zeta(x) $为定义在紧集$ \varpi $上的连续函数.
1.4 规定性能函数
定义规定性能机制[30]为
$$ \begin{align} \begin{aligned} -e_{h,\;1,\;{\mathrm{min}}}(t)<e_{h,\;1}(t)<e_{h,\;1,\;{\mathrm{max}}}(t) \end{aligned} \end{align} $$ (6) 接下来, 定义规定性能上下边界为
$$ \begin{aligned} e_{h,\;1,\;{\mathrm{min}}}(t)=\;&(e_{h,\;1,\;0,\;{\mathrm{min}}}-e_{h,\;1,\;\infty,\;{\mathrm{min}}})\text{e}^{-o_{h}t}\;+\\ & e_{h,\;1,\;\infty,\;{\mathrm{min}}}\nonumber\\ e_{h,\;1,\;{\mathrm{max}}}(t)=\;&(e_{h,\;1,\;0,\;{\mathrm{max}}}-e_{h,\;1,\;\infty,\;{\mathrm{max}}})\text{e}^{-o_{h}t}\;+\\ & e_{h,\;1,\;\infty,\;{\mathrm{max}}}\nonumber \end{aligned} $$ 其中, $ -e_{h,\;1,\;{\mathrm{min}}}\;(t) $表示设计的规定范围下界, $ e_{h,\;1,\;{\mathrm{max}}}(t) $ 表示设计的规定范围上界. $ o_{h} $ 是一个可设计的常数. $ e_{h,\;1,\;0,\;{\mathrm{min}}} $, $ e_{h,\;1,\;0,\;{\mathrm{max}}} $, $ e_{h,\;1,\;\infty,\;{\mathrm{min}}} $ 和 $ e_{h,\;1,\;\infty,\;{\mathrm{max}}} $ 是正的参数. 并且, 参数需要满足$ e_{h,\;1,\;0,\;{\mathrm{min}}} > e_{h,\;1,\;\infty,\;{\mathrm{min}}} $ 和 $ e_{h,\;1,\;0,\;{\mathrm{max}}} > e_{h,\;1,\;\infty,\;{\mathrm{max}}} $.
假设2[30]. 对于智能体$ h $, 初始同步误差必须满足受限不等式$ -e_{h,\;1,\;{\mathrm{min}}}(0) < e_{h,\;1}(0) < e_{h,\;1,\;{\mathrm{max}}}(0) $.
根据所考虑的规定性能方法, 得到误差转换机制为
$$ \begin{align} \begin{aligned} e_{h,\;1}=e_{h,\;1,\;{\mathrm{max}}}\Re_{h}(s_{h,\;1}) \end{aligned} \end{align} $$ (7) 其中, $ s_{h,\;1} $ 是转换后的误差. $ \Re_{h}(s_{h,\;1}) $ 表示误差转换函数, 其是光滑且严格单调递增的, 同时满足 $ \Re_{h}(s_{h,\;1})\in(-\kappa_{h},\;1) $ 且 $ \kappa_{h}={e_{h,\;1,\;{\mathrm{max}}}(t)}/{e_{h,\;1,\;{\mathrm{min}}}(t)} $.
转换函数的表达式为
$$ \begin{align} \begin{aligned} \Re_{h}(s_{h,\;1})=\frac{\text{e}^{s_{h,\;1}}-\text{e}^{-s_{h,\;1}}}{\text{e}^{s_{h,\;1}}+{\kappa}^{-1}_{h}\text{e}^{-s_{h,\;1}}} \end{aligned} \end{align} $$ (8) 将式(8)代入式(7), 可得
$$ \begin{align} s_{h,\;1}=\frac{1}{2}\ln\left(1+\frac{e_{h,\;1}}{e_{h,\;1,\;{\mathrm{min}}}}\right)-\frac{1}{2}\ln\left(1-\frac{e_{h,\;1}}{e_{h,\;1,\;{\mathrm{max}}}}\right) \end{align} $$ (9) 注1. 由于本文考虑状态触发机制, 该机制会导致系统出现阶跃现象或抖震现象. 为克服这一现象发生, 本文采用规定性能方法来约束系统的同步误差以减少系统性能下降的情况发生.
1.5 混合双端事件触发机制
针对非线性多智能体系统, 如何有效节省通信资源是十分重要的问题, 事件触发机制可以减少通信带宽的占用. 同时, 在设计事件触发机制时, 重要的是在设计相应的阈值条件时要考虑到通信资源和跟踪性能之间的平衡.
在网络环境中进行信息交换时, 多个信息传输通道会同时进行数据传输. 基于这一考虑, 本文设计混合双端事件触发机制来同时释放控制器−执行器环节和传感器−控制器环节中通信渠道上的压力. 因此, 提出如下的混合双端事件触发机制:
$$ \left\{\begin{split} &\breve{\hat{x}}_{h,\;g}=\hat{x}_{h,\;g}(t^{x}_{h,\;k}),\;t\in[t^{x}_{h,\;k},\;t^{x}_{h,\;k+1})\\ &t^{x}_{h,\;k+1}=\inf\Bigr\{t>t^{x}_{h,\;k}:|\hat{x}_{h,\;g}(t)-\breve{\hat{x}}_{h,\;g}|\ge \\ &\qquad\qquad \nu_{h}+m_{h}\text{e}^{-b_{h}t}\Bigr\} \end{split}\right. $$ (10) 随后可得
$$ \left\{\begin{split} &\breve{u}_{h}(t)=u_{h}(t^{u}_{h,\;k}),\;t\in[t^{u}_{h,\;k},\;t^{u}_{h,\;k+1})\\ &t^{u}_{h,\;k+1}=\inf\Bigr\{t>t^{u}_{h,\;k}:|u_{h}(t)-\breve{u}_{h}(t)|\ge\\ &\qquad\qquad \rho_{h}+\mu_{h}\text{e}^{-\tau_{h}t}\Bigr\} \end{split} \right.$$ (11) 其中, $ t^{x}_{h,\;k} $ 表示系统状态的触发时刻. $ t^{u}_{h,\;k} $ 表示控制输入信号的触发时刻. $ \nu_{h} $, $ m_{h} $, $ b_{h} $, $ \rho_{h} $, $ \mu_{h} $ 和 $ \tau_{h} $ 是正的常数. 同时, 通常假设第一个触发发生在系统运行的初始时刻.
注2. 由于本文考虑未知不可测量状态问题, 提出的状态触发机制首次使用估计状态作为采样信号并构成触发误差, 拓宽了状态触发机制的应用范围. 并且, 设计的阈值条件会随着系统运行时间的变化而变化, 从而更好地平衡了系统性能和资源节约之间的关系.
2. 模糊状态观测器的设计
本节通过构造模糊状态观测器解决未知状态不可测量问题, 该观测器仅使用相对输出分布式误差信息进行反馈. 首先, 定义$ \zeta_{h}(x_{h},u_{h})=\zeta_{h,\;n}(\bar{x}_{h,\;n}) $ 且要求 $ |\zeta_{h}(x_{h},\;u_{h})|\le \bar{\zeta}_{h}(x_{h},\;u_{h}) $. 重新构造系统模型为
$$ \begin{aligned} \left\{ \begin{aligned} &\dot{x}_{h,\;g}=x_{h,\;g+1}+\zeta_{h,\;g}(\hat{\bar{x}}_{h,\;g})\\ &\dot{x}_{h,\;n}=u_{h}+\zeta_{h}(\hat{x}_{h},\;u_{h})\\ & y_{h}=x_{h,\;1} \end{aligned} \right. \end{aligned} $$ (12) 基于模糊逻辑系统理论, 考虑的多智能体系统包含非线性函数$ \zeta_{h,\;g}(\hat{\bar{x}}_{h,\;g}) $ 和 $ \zeta_{h}(\hat{x}_{h},\;u_{h}) $, 近似这两项可得:
$$ \begin{align} \begin{aligned} \hat{\zeta}_{h,\;g}(\hat{\bar{x}}_{h,\;g}|\eta_{h,\;g})=\eta^{*\text{T}}_{h,\;g}\psi_{h,\;g}(\hat{\bar{x}}_{h,\;g}) \end{aligned} \end{align} $$ (13) 其中, $ \hat{\bar{x}}_{h,\;g} $ 表示$ {\bar{x}}_{h,\;g} $的估计值.
$ \eta^{*}_{h,\;g} $ 是最优参数向量, 其可以表示为
$$ \begin{split} \eta^{*}_{h,\;g}=&\;\arg\underset{\eta_{h,\;g}\in\Omega_{h,\;g}}\min[\underset{(\bar{x}_{h,\;g},\;\hat{\bar{x}}_{h,\;g})\in U}\sup| \hat{\zeta}(\hat{\bar{x}}_{h,\;g}|\hat{\eta}_{h,\;g})\;-\\ &\zeta(\hat{\bar{x}}_{h,\;g})|] \\[-1pt]\end{split} $$ (14) 其中, $ U $ 和 $ \Omega_{h,\;g} $ 分别为 $ \hat{\bar{x}}_{h,\;g} $ 和 $ \eta_{h,\;g} $ 对应的紧集. $ \hat{\eta}_{h,\;g} $ 表示 $ \eta^{*}_{h,\;g} $ 的估计值. $ \varepsilon_{h,\;g} $ 表示模糊最小化近似误差且 $ \varepsilon_{h,\;g}=\zeta_{h,\;g}(\bar{x}_{h,\;g})-\hat{\zeta}_{h,\;g}(\hat{\bar{x}}_{h,\;g}|\eta_{h,\;g}) $. 接下来, 构建分布式状态观测器为
$$ \left\{\begin{aligned} &\dot{\hat x}_{h,\;g}=\hat{x}_{h,\;g+1}+\hat{\eta}^\text{T}_{h,\;g}\psi_{h,\;g}(\hat{\bar{x}}_{h,\;g})\;+\\ &\quad k^{*}_{h,\;g}\left(\sum_{\ell=1}^N a_{h,\;\ell}(y_{h}-y_{\ell})-\sum_{\ell=1}^N a_{h,\;\ell}(\hat{y}_{h}-y_{\ell})\right)\\ &\dot{\hat x}_{h,\;n}=u_{h}+\hat{\eta}^\text{T}_{h,\;n}\psi_{h,\;n}(\hat{\bar{x}}_{h,\;n})\;+\\ &\quad k^{*}_{h,\;n}\left(\sum_{\ell=1}^N a_{h,\;\ell}(y_{h}-y_{\ell})-\sum_{\ell=1}^N a_{h,\;\ell}(\hat{y}_{h}-y_{\ell})\right)\\ &\hat{y}_{h}=\hat{x}_{h,\;1} \\[-1pt] \end{aligned}\right. $$ (15) 其中, $ k^{*}_{h,\;g} $ 和 $ k^{*}_{h,\;n} $ 是正的常数. 此外, 相对输出误差表示为 $ \sum_{\ell=1}^N a_{h,\;\ell}(y_{h} - y_{\ell}) $. 定义 $ \Delta_{h} = \bar{x}_{h,\;n} - \hat{\bar{x}}_{h,\;n} $ 和 $ \hat{\bar{x}}_{h,\;n}=[\hat{x}_{h,\;1},\;\cdots,\;\hat{x}_{h,\;n}]^\text{T} $. 经过上述分析, 可得
$$ \begin{align} \begin{aligned} \dot{\Delta}_{h}=&\Xi_{h}\Delta_{h}+\varepsilon_{h}+\sum_{\eth=1}^{n} A_{h,\;\eth} \tilde{\eta}^\text{T}_{h,\;\eth}\psi_{h,\;\eth}(\hat{\bar{x}}_{h,\;\eth}) \end{aligned} \end{align} $$ (16) 其中, $ \varepsilon_{h} = [\varepsilon_{h,\,1},\,\cdots,\,\varepsilon_{h,\,n}]^\text{T} $, $ A_{h,\,\eth} = [0\cdots1\cdots0]_{n \times 1} $. 为了确保 $ \Xi_{h} $ 是一个严格的赫尔维兹矩阵, 选择向量 $ K_{h} $ 且 $ K_{h}=[k^{*}_{h,\;1},\;\cdots,\;k^{*}_{h,\;n}]^\text{T} $.
并且,
$$ \begin{align} \begin{aligned} A_{h}= \begin{bmatrix} -k^{*}_{h,\;1}\displaystyle\sum\limits_{\ell=1}^N a_{h,\;\ell}\\ -k^{*}_{h,\;2}\displaystyle\sum\limits_{\ell=2}^N a_{h,\;\ell}&H_{h,\;n-1}&\\ \vdots\\ -k^{*}_{h,\;n}\displaystyle\sum\limits_{\ell=n}^N a_{h,\;\ell}&...&0\nonumber \end{bmatrix} \end{aligned} \end{align} $$ 对于矩阵 $ Q_{h}=Q^\text{T}_{h}>0 $ 和矩阵 $ Y_{h}=Y^\text{T}_{h}>0 $, 满足如下关系
$$ \begin{align} \begin{aligned} A^\text{T}_{h}Y_{h}+Y_{h}A_{h}=-2Q_{h} \end{aligned} \end{align} $$ (17) 挑选如下Lyapunov函数 $ V_{h,\;0} $:
$$ \begin{align} \begin{aligned} V_{h,\;0}=\frac{1}{2}\tilde{x}^\text{T}_{h}Y_{h}\tilde{x}_{h} \end{aligned} \end{align} $$ (18) 基于上述分析, 计算$ V_{h,\;0} $的导数可以为
$$ \begin{split} \dot{V}_{h,\;0}=\;&-\Delta^\text{T}_{h}Q_{h}\Delta_{h}+\Delta^\text{T}_{h}Y_{h}\varepsilon_{h}\;+\\ &\Delta^\text{T}_{h}Y_{h}\sum_{\eth=1}^{n} A_{h,\;\eth} \tilde{\eta}^\text{T}_{h,\;\eth}\psi_{h,\;\eth} \end{split} $$ (19) 利用杨氏不等式, 可得
$$ \begin{align} \Delta^\text{T}_{h}Y_{h}\varepsilon_{h}\le\frac{1}{2}||Y_{h}||^{2}||\varepsilon^{*}_{h}||^{2}+\frac{1}{2}||\Delta_{h}||^{2} \end{align} $$ (20) $$ \begin{split} &e^\text{T}_{h}Y_{h}\sum_{\eth=1}^{n} A_{h,\;\eth} \tilde{\eta}^\text{T}_{h,\;\eth}\psi_{h,\;\eth}(\hat{\bar{x}}_{h,\;\eth})\le \\ &\qquad\frac{n}{2}||\Delta_{h}||^{2}+ \frac{1}{2}||Y_{h}||^{2}\sum_{\eth=1}^{n} A_{h,\;\eth} \tilde{\eta}^\text{T}_{h,\;\eth}\tilde{\eta}_{h,\;\eth} \end{split} $$ (21) 其中, $ \varepsilon^{*}_{h}=[\varepsilon^{*}_{h,\;1},\;\varepsilon^{*}_{h,\;2},\;\cdots,\;\varepsilon^{*}_{h,\;n}]^\text{T} $.
随后, 可得
$$ \begin{split} \dot{V}_{h,\;0}\le\;&\frac{1+n}{2}||\Delta_{h}||^{2}-\Delta^\text{T}_{h}Q_{h}\Delta_{h}+\frac{1}{2}||Y_{h}||^{2}||\varepsilon^{*}_{h}||^{2}\;+\\ &\frac{1}{2}||Y_{h}||^{2}\sum_{\eth=1}^{n} A_{h,\;\eth} \tilde{\eta}^\text{T}_{h,\;\eth}\tilde{\eta}_{h,\;\eth}\le\frac{1}{2}||Y_{h}||^{2}||\tilde{\eta}^{2}_{h,\;n}\;+\\ & \frac{1}{2}||Y_{h}||^{2}||\varepsilon^{*}_{h}||^{2}+ \xi_{0}||\Delta_{h}||^{2}\\[-1pt] \end{split} $$ (22) 其中, $ \xi_{0} = \min\left\{\tau_{{\mathrm{min}}}(Q_{h}) - \frac{1+n}{2}\right\} $ 和 $ \xi_{0} > 0 $. $ \tau_{{\mathrm{min}}}(Q_{h}) $ 是矩阵 $ Q_{h} $ 的最小特征值.
注3. 本文设计的模糊状态观测器仅依赖于智能体的相对输出信息进行反馈, 表明仅使用部分的分布式信息就可以解决未知不可测状态问题. 此外, 设计的观测器可识别严格反馈多智能体系统中的未知非线性函数.
3. 主要内容
3.1 自适应分布式控制器设计
本节给出自适应控制器的设计过程且解决非线性多智能体系统的自适应模糊跟踪控制问题. 定义局部的同步误差 $ e_{h,\;1} $ 和 $ e_{h,\;g} $ ($ g\;=\; 2,\;\cdots,\;n $) 为
$$ \left\{\begin{split} &e_{h,\;1}=\sum_{\ell=1}^Na_{h,\;\ell}(y_{h}-y_{\ell})+b_{h}(y_{h}-y_{r})\\ &e_{h,\;g}=\hat{x}_{h,\;g}-\alpha_{hf,\;g-1} \end{split}\right.$$ (23) 其中, $ \alpha_{hf,\;g-1} $ 表示滤波后的虚拟控制器.
在传统的反步法框架下, 为避免“复杂性爆炸”问题, 引入了一阶滤波器:
$$ \left\{\begin{split} &\Phi_{h,\;g-1}\dot{\alpha}_{hf,\;g-1}+{\alpha}_{hf,\;g-1}={\alpha}_{h,\;g-1}\\ &{\alpha}_{hf,\;g-1}(0)={\alpha}_{h,\;g-1}(0) \end{split}\right. $$ (24) 其中, $ \alpha_{h,\;g - 1} $ 是虚拟控制信号. 虚拟控制信号$ \alpha_{h,\;g - 1} $ 通过一阶滤波器 $ \Phi_{h,\;g-1}>0 $ 会产生一个新的信号 $ {\alpha}_{hf,\;g-1} $.
随后, 表示一阶滤波器的误差为
$$ \begin{align} \begin{aligned} \vartheta_{h,\;g-1}={\alpha}_{hf,\;g-1}-{\alpha}_{h,\;g-1} \end{aligned} \end{align} $$ (25) 基于式 (9), 计算 $ s_{h,\;1} $ 的导数为
$$ \begin{align} \begin{aligned} \dot{s}_{h,\;1}=\rho_{h}\dot{e}_{h,\;1}-\phi_{h}e_{h,\;1} \end{aligned} \end{align} $$ (26) 其中,
$$ \begin{align} \begin{aligned} \rho_{h}=\;&\frac{1}{2}\bigg(\frac{1}{e_{h,\;1,\;{\mathrm{min}}}+e_{h,\;1}}+\frac{1}{e_{h,\;1,\;{\mathrm{max}}}-e_{h,\;1}}\bigg)\\ \phi_{h}=\;&\frac{1}{2}\bigg(\frac{\dot{e}_{h,\;1,\;{\mathrm{min}}}}{e_{h,\;1,\;{\mathrm{min}}}(e_{h,\;1,\;{\mathrm{min}}}+e_{h,\;1})}\;+\\ &\frac{\dot{e}_{h,\;1,\;{\mathrm{max}}}}{e_{h,\;1,\;{\mathrm{max}}}(e_{h,\;1,\;{\mathrm{max}}}-e_{h,\;1})}\bigg)\nonumber \end{aligned} \end{align} $$ 步骤 1. 选择如下的Lyapunov函数:
$$ \begin{align} \begin{aligned} V_{h,\;1}=\frac{1}{2}s^{2}_{h,\;1}+\frac{1}{2}\eta^\text{T}_{h,\;1}\eta_{h,\;1}+\frac{1}{2}\vartheta^{2}_{h,\;1} \end{aligned} \end{align} $$ (27) 计算$ V_{h,\;1} $的导数为
$$ \begin{split} \dot{V}_{h,\;1}&=s_{h,\;1}\dot{s}_{h,\;1}-\tilde{\eta}_{h,\;1}\dot{\hat{\eta}}_{h,\;1}+\vartheta_{h,\;1}\dot{\vartheta}_{h,\;1}=\\ &s_{h,\;1}(\rho_{h}\dot{e}_{h,\;1}-\phi_{h}e_{h,\;1})-\tilde{\eta}_{h,\;1}\dot{\hat{\eta}}_{h,\;1}+\vartheta_{h,\;1}\dot{\vartheta}_{h,\;1}=\\ &s_{h,\;1}\rho_{h}\left(\sum_{\ell=1}^Na_{h,\;\ell}(\dot{y}_{h}-\dot{y}_{\ell})+b_{h}(\dot{y}_{h}-\dot{y}_{r})\right)-\\ &\phi_{h}e_{h,\;1}s_{h,\;1}+\vartheta_{h,\;1}\dot{\vartheta}_{h,\;1}-\tilde{\eta}_{h,\;1}\dot{\hat{\eta}}_{h,\;1}\\[-1pt] \end{split} $$ (28) 根据式 (15), 进一步可得
$$ \begin{split} \dot{V}_{h,\;1}=\;&s_{h,\;1}\rho_{h}\big((b_{h}+d_{h})(\hat{x}_{h,\;2}+\zeta_{h,\;1})\;-\\ & d_{h}(\hat{x}_{\ell,\;2}+\zeta_{\ell,\;1})-b_{h}\dot{y}_{r}\big)-\phi_{h}e_{h,\;1}s_{h,\;1}\;-\\ &\tilde{\eta}_{h,\;1}\dot{\hat{\eta}}_{h,\;1}+\vartheta_{h,\;1}\dot{\vartheta}_{h,\;1}=s_{h,\;1}\rho_{h}\big((b_{h}\;+\\ & d_{h})(e_{h,\;2}+\alpha_{h,\;1}+\vartheta_{h,\;1}+\zeta_{h,\;1})\;-\\ & d_{h}(\hat{x}_{\ell,\;2}+\zeta_{\ell,\;1})-b_{h}\dot{y}_{r}-\phi_{h}e_{h,\;1}\rho^{-1}_{h}\big)\;-\\ & \tilde{\eta}_{h,\;1}\dot{\hat{\eta}}_{h,\;1}+\vartheta_{h,\;1}\dot{\vartheta}_{h,\;1} \end{split} $$ (29) 其中, $ e_{h,\;2}=\hat{x}_{h,\;2}-\alpha_{hf,\;1} $ 且 $ \vartheta_{h,\;1}={\alpha}_{hf,\;1}-{\alpha}_{h,\;1} $. $ \sum_{j=1}^Na_{h,\;\ell} $ 可以由 $ d_{h} $ 来表示.
随后, 根据引理3, 可得
$$ \begin{split} &\bar{F}_{h,\;1}({x}_{h,\;1},\;{x}_{\ell,\;1})=(b_{h}+d_{h})f_{h,\;1}-d_{h}(\hat{x}_{\ell,\;2}\;+\\ &\quad f_{\ell,\;1})\nonumber-\phi_{h}e_{h,\;1}\rho^{-1}_{h}\nonumber={\eta}^{\mathrm{T}}_{h,\;1}\psi_{h,\;1}+\varepsilon_{h,\;1} \end{split} $$ 式 (29) 可进一步表示为
$$ \begin{split} \dot{V}_{h,\;1}=\;&s_{h,\;1}\rho_{h}\big((b_{h}+d_{h})(e_{h,\;2}+\alpha_{h,\;1}+\vartheta_{h,\;1})\nonumber\;-\\ & b_{h}\dot{y}_{r}+\bar{F}_{h,\;1}\big)-\tilde{\eta}_{h,\;1}\dot{\hat{\eta}}_{h,\;1}+\vartheta_{h,\;1}\dot{\vartheta}_{h,\;1}\nonumber=\\ & s_{h,\;1}\rho_{h}\big((b_{h}+d_{h})(e_{h,\;2}+\alpha_{h,\;1}+\vartheta_{h,\;1})\nonumber\;+\\ & {\eta}^{\mathrm{T}}_{h,\;1}\psi_{h,\;1}+\varepsilon_{h,\;1}-b_{h}\dot{y}_{r})\big)\nonumber\;-\\ &\tilde{\eta}_{h,\;1}\dot{\hat{\eta}}_{h,\;1}+\vartheta_{h,\;1}\dot{\vartheta}_{h,\;1} \end{split} $$ 基于杨氏不等式, 可得
$$ \begin{align} \left\{\begin{aligned} &(b_{h}+d_{h})s_{h,\;1}e_{h,\;2}\le\frac{(b_{h}+d_{h})^{2}}{2}s^{2}_{h,\;1}+\frac{1}{2}e^{2}_{h,\;2}\nonumber\\&(b_{h}+d_{h})s_{h,\;1}\vartheta_{h,\;1}\le\frac{(b_{h}+d_{h})^{2}}{2}s^{2}_{h,\;1}+\frac{1}{2}\vartheta^{2}_{h,\;1}\nonumber\\ &s_{h,\;1}\varepsilon_{h,\;1}\le\frac{1}{2}s^{2}_{h,\;1}+\frac{1}{2}\varepsilon^{2}_{h,\;1} \end{aligned}\right. \end{align} $$ 设计虚拟控制器为
$$ \begin{split} \alpha_{h,\;1}=\; &\rho^{-1}_{h}\bigg(-(b_{h}+d_{h})s_{h,\;1}-\frac{1}{2(b_{h}+d_{h})}s_{h,\;1}\;-\\ & c_{h,\;1}s_{h,\;1}+\frac{b_{h}}{b_{h}+d_{h}}\dot{y}_{r}-\frac{1}{b_{h}+d_{h}}\hat{\eta}_{h,\;1}\psi_{h,\;1}\bigg) \end{split} $$ (30) 其中, $ c_{h,\;1} $ 是设计参数. $ \hat{\eta}_{h,\;1} $ 是 $ {\eta}_{h,\;1} $的估计值.
在触发时刻, 系统会更新自适应律, 并且在触发间隔区间中保持不变. 因此, 设计自适应律为
$$ \left\{\begin{aligned} \hat{\eta}^+_{h,\;1}&=-N_{h,\;1}\hat{\eta}_{h,\;1}+s_{h,\;1}\psi_{h,\;1},&& t=t_{h,\;k}\\ \dot{\hat{\eta}}_{h,\;1}&=0,&&t\in[t_{h,\;k},\;t_{h,\;k+1}) \end{aligned} \right. $$ (31) 其中, $ N_{h,\;1} $ 是设计参数.
基于上述分析, 可知
$$ \begin{split} \dot{V}_{h,\;1}&\le-{c}_{h,\;1}s^{2}_{h,\;1}-\frac{1}{2}N_{h,\;1}\tilde{\eta}^{2}_{h,\;1}+\frac{1}{2}N_{h,\;1}{\eta}^{2}_{h,\;1}\;+\nonumber\\ &\quad\frac{1}{2}\rho_{h}\vartheta^{2}_{h,\;1}+\frac{1}{2}\varepsilon^{2}_{h,\;1}\rho_{h}+\frac{1}{2}\rho_{h}e^{2}_{h,\;2}+\vartheta_{h,\;1}\dot{\vartheta}_{h,\;1} \end{split} $$ 随后, 可得
$$ \begin{align} \begin{aligned} \dot{\vartheta}_{h,\;1}=-\frac{1}{\upsilon_{h,\;1}}{\vartheta}_{h,\;1}-\dot{\alpha}_{h,\;1} \end{aligned} \end{align} $$ (32) 其中, $ \upsilon_{h,\;1} $ 是设计常数.
根据上述分析, $ \dot{V}_{h,\;1} $ 满足如下不等式:
$$ \begin{split} \dot{V}_{h,\;1}\le &-c_{h,\;1}s^{2}_{h,\;1}-\frac{1}{2}N_{h,\;1}\tilde{\eta}^{2}_{h,\;1}+\frac{1}{2}N_{h,\;1}{\eta}^{2}_{h,\;1}\;+\\ &\frac{1}{2}\varepsilon^{2}_{h,\;1}\rho_{h}+\frac{1}{2}\rho_{h}e^{2}_{h,\;2}-\frac{1}{\upsilon_{h,\;1}}{\vartheta}^{2}_{h,\;1}\;+\\ &\frac{1}{2}\dot{\alpha}^{2}_{h,\;1}+\frac{1}{2}\rho_{h}\vartheta^{2}_{h,\;1}\\[-1pt] \end{split} $$ (33) 步骤g. $(g=2, \cdots, n-1) $选择如下的Lyapunov函数:
$$ \begin{align} \begin{aligned} V_{h,\;g}=\frac{1}{2}e^{2}_{h,\;g}+\frac{1}{2}\eta^\text{T}_{h,\;g}\eta_{h,\;g}+\frac{1}{2}\vartheta^{2}_{h,\;g} \end{aligned} \end{align} $$ (34) 计算$ V_{h,\;g} $的导数为
$$ \begin{split} \dot{V}_{h,\;g}=\;&e_{h,\;g}\dot{e}_{h,\;g}-\tilde{\eta}_{h,\;g}\dot{\hat{\eta}}_{h,\;g}+\vartheta_{h,\;g}\dot{\vartheta}_{h,\;g}=\\ &e_{h,\;g}(\dot{\hat{x}}_{h,\;g}-\dot{\alpha}_{hf,\;g-1})\;-\\ & \tilde{\eta}_{h,\;g}\dot{\hat{\eta}}_{h,\;g}+\vartheta_{h,\;g}\dot{\vartheta}_{h,\;g} \end{split} $$ (35) 基于式 (15), 式 (35)重新表示为
$$ \begin{split} \dot{V}_{h,\;g}=\;&e_{h,\;g}\big(\hat{x}_{h,\;g+1}+\eta_{h,\;g}\psi_{h,\;g}+k^{*}_{h,\;g}\big(d_{h}(y_{h}-y_{\ell})\nonumber\;-\\ & d_{h}(\hat{y}_{h}-y_{\ell}) \big) -\dot{\alpha}_{hf,\;g-1} \big) -\tilde{\eta}_{h,\;g}\dot{\hat{\eta}}_{h,\;g}\nonumber + \vartheta_{h,\;g}\dot{\vartheta}_{h,\;g} \end{split} $$ 其中, $ \hat{x}_{h,\;g+1}=e_{h,\;g+1}+\alpha_{hf,\;g} $.
随后, 可得
$$\begin{aligned} \dot{V}_{h,\;g}=\;&e_{h,\;g}\big(e_{h,\;g+1}+\alpha_{hf,\;g}+\eta_{h,\;g}\psi_{h,\;g}\;+\\ & k^{*}_{h,\;g}\big(d_{h}(y_{h}\nonumber-y_{\ell})-d_{h}(\hat{y}_{h}-y_{\ell})\big)-\dot{\alpha}_{hf,\;g-1}\big)\;-\\ & \tilde{\eta}_{h,\;g}\dot{\hat{\eta}}_{h,\;g}\nonumber+\vartheta_{h,\;g}\dot{\vartheta}_{h,\;g}\nonumber=e_{h,\;g}\big(e_{h,\;g+1}+\alpha_{h,\;g}\;+\\ & {\vartheta}_{h,\;g}+\eta_{h,\;g}\psi_{h,\;g}\nonumber+k^{*}_{h,\;g}\big(d_{h}(y_{h}-y_{\ell})\;-\\ &d_{h}(\hat{y}_{h} - y_{\ell}) \big)\nonumber-\dot{\alpha}_{hf,\;g-1}\big) -\tilde{\eta}_{h,\;g}\dot{\hat{\eta}}_{h,\;g} +\vartheta_{h,\;g}\dot{\vartheta}_{h,\;g} \end{aligned} $$ 基于杨氏不等式, 可知
$$ \begin{align} \begin{aligned} e_{h,\;g}\vartheta_{h,\;g}\le\frac{1}{2}e^{2}_{h,\;g}+\frac{1}{2}\vartheta^{2}_{h,\;g} \end{aligned} \end{align} $$ (36) 设计虚拟控制器为
$$ \begin{split} \alpha_{h,\;g}=\; &-k^{*}_{h,\;g}\big(d_{h}(y_{h}-y_{\ell})-d_{h}(\hat{y}_{h}-y_{\ell})\big)\;-\\ &\hat{\eta}_{h,\;g}\psi_{h,\;g}-\left(\frac{1}{2}+c_{h,\;g}\right)e_{h,\;g}+\dot{\alpha}_{hf,\;g-1} \end{split} $$ (37) 其中, $ c_{h,\;g} $ 是设计常数.
设计自适应律为
$$ \begin{align} \left\{ \begin{aligned} \hat{\eta}^+_{h,\;g}&=-N_{h,\;g}\hat{\eta}_{h,\;g}+e_{h,\;g}\psi_{h,\;g},&& t=t_{h,\;k}\\ \dot{\hat{\eta}}_{h,\;g}&=0,&& t\in[t_{h,\;k},\;t_{h,\;k+1}) \end{aligned} \right. \end{align} $$ (38) 并且, 可知
$$ \begin{align} \begin{aligned} \dot{\vartheta}_{h,\;g}=-\frac{1}{\upsilon_{h,\;g}}{\vartheta}_{h,\;g}-\dot{\alpha}_{h,\;g} \end{aligned} \end{align} $$ (39) 其中, $ \upsilon_{h,\;g} $ 是一个设计常数.
接下来, $ \dot{V}_{h,\;g} $ 满足如下不等式:
$$ \begin{split} \dot{V}_{h,\;g}\le&-{c}_{h,\;g}e^{2}_{h,\;g}-\frac{1}{2}N_{h,\;g}\tilde{\eta}^{2}_{h,\;g}+\frac{1}{2}N_{h,\;g}{\eta}^{2}_{h,\;g}\;+\\ &\frac{1}{2}\vartheta^{2}_{h,\;g}-\frac{1}{\upsilon_{h,\;g}}{\vartheta}^{2}_{h,\;g}+\frac{1}{2}\dot{\alpha}^{2}_{h,\;g} \\[-1pt]\end{split} $$ (40) 步骤${\boldsymbol{n}} $. 当应用一阶滤波器, 可得
$$\left\{\begin{split} &\Phi_{h,\;n-1}\dot{\alpha}_{hf,\;n-1}+{\alpha}_{hf,\;n-1}={\alpha}_{h,\;n-1}\\ &{\alpha}_{hf,\;n-1}(0)={\alpha}_{h,\;n-1}(0) \end{split}\right. $$ (41) 其中, $ \alpha_{h,\;n-1} $ 是虚拟控制信号. 虚拟控制信号$ \alpha_{h,\;n-1} $ 通过一阶滤波器 $ \Phi_{h,\;n-1}>0 $ 会产生一个新的信号 $ {\alpha}_{hf,\;n-1} $.
挑选如下的Lyapunov函数:
$$ \begin{align} \begin{aligned} V_{h,\;n}=\frac{1}{2}e^{2}_{h,\;n}+\frac{1}{2}\eta^\text{T}_{h,\;n}\eta_{h,\;n} \end{aligned} \end{align} $$ (42) 计算$ V_{h,\;n} $的导数为
$$ \begin{split} \label{lin} \begin{aligned} \dot{V}_{h,\;n}=\;&e_{h,\;n}\dot{e}_{h,\;n}-\tilde{\eta}_{h,\;n}\dot{\hat{\eta}}_{h,\;n}\nonumber=\\ &e_{h,\;n}(\dot{\hat{x}}_{h,\;n}-\dot{\alpha}_{hf,\;n-1})-\tilde{\eta}_{h,\;n}\dot{\hat{\eta}}_{h,\;n} \end{aligned} \end{split} $$ 随后, 可得
$$ \begin{split} \dot{V}_{h,\;n}=\;&e_{h,\;n}\Bigg(u_{h}+{\eta}_{h,\;n}\psi_{h,\;n}(\hat{\bar{x}}_{h,\;n})\;+\\ &k^{*}_{h,\;n}\Biggr(\sum_{\ell=1}^N a_{h,\;\ell}(y_{h}-y_{\ell})-\sum_{\ell=1}^N a_{h,\;\ell}(\hat{y}_{h}\nonumber\;-\\ &y_{\ell})\Biggr)-\dot{\alpha}_{hf,\;n-1}\Bigg)-\tilde{\eta}_{h,\;n}\dot{\hat{\eta}}_{h,\;n}\nonumber=\\ &e_{h,\;n}\Bigg(\breve{u}_{h}+(u_{h}-\breve{u}_{h})+\eta^\text{T}_{h,\;n}\psi_{h,\;n}(\hat{\bar{x}}_{h,\;n})\;+\\ & k^{*}_{h,\;n}\Biggr(\sum_{\ell=1}^N a_{h,\;\ell}(y_{h}-y_{\ell})-\sum_{j=1}^N a_{h,\;\ell}(\hat{y}_{h}\nonumber\;-\\ & y_{\ell})\Biggr)-\dot{\alpha}_{hf,\;n-1}\Bigg)-\tilde{\eta}_{h,\;n}\dot{\hat{\eta}}_{h,\;n} \end{split} $$ 设计自适应事件触发控制器 $ \breve{u}_{h} $ 为
$$ \begin{split} \breve{u}_{h}=\;&-c_{h,\;n}\breve{e}_{h,\;n}-\hat{\eta}_{h,\;n}\psi_{h,\;n}\;+\\ &\frac{\breve{\alpha}_{h,\;n-1}-\breve{\alpha}_{hf,\;n-1}}{\Phi_{h,\;n-1}}- k^{*}_{h,\;n}\Biggr(\sum_{\ell=1}^Na_{h,\;\ell}(y_{h}-y_{\ell})\;-\\ &\sum_{\ell=1}^N a_{h,\;\ell}(\hat{y}_{h}-y_{\ell})\Biggr)\\[-1pt]\end{split} $$ (43) 其中,
$$ \begin{split} &\breve{e}_{h,\;n}(t)=\breve{\hat{x}}_{h,\;n}(t)-\breve{\alpha}_{hf,\;n-1}(t)\nonumber\\ &\breve{\alpha}_{hf,\;n-1}(t)={\alpha}_{hf,\;n-1}(t^{\alpha_{f}}_{h,\;k}),\quad t\in[t^{\alpha_{f}}_{h,\;k},\;t^{\alpha_{f}}_{h,\;k+1})\nonumber \end{split} $$ $$ \begin{split} t^{k+1}_{h,\;\alpha_{f}}=\;&\;\inf\Bigr\{t>t^{k}_{h,\;\alpha_{f}}:|{\dot{\alpha}}_{hf,\;n-1}(t)\;-\\ &{\dot{\breve{\alpha}}}_{hf,\;n-1}(t)|\ge\Theta_{h,\;\alpha_{f}}\Bigr\} \end{split} $$ (44) $ \breve{\alpha}_{h,\;n-1} $ 是 $ \alpha_{h,\;n-1} $ 触发后的信号. $ \breve{\alpha}_{hf,\;n-1} $ 是 $ \alpha_{hf,\;n-1} $ 触发后的信号. 在本文考虑的状态触发机制中, $ {\alpha}_{hf,\;n-1} $ 和 $ {\alpha}_{h,\;n-1} $ 均依赖于系统的状态值.
选择自适应律为
$$ \left\{\begin{split}& \hat{\eta}^+_{h,\;n}=-N_{h,\;n}\hat{\eta}_{h,\;n}+e_{h,\;n}\psi_{h,\;n}, \;\; t=t_{h,\;k}\\ &\dot{\hat{\eta}}_{h,\;n}=0,\qquad\qquad\qquad\qquad\qquad\; t\in[t_{h,\;k},\;t_{h,\;k+1}) \end{split} \right.$$ (45) 其中, $ N_{h,\;n} $ 是设计参数.
随后, 可知
$$ \begin{split} \dot{V}_{h,\;n}=\;&e_{h,\;n}\big((u_{h}-\breve{u}_{h})+{c}_{h,\;n}(e_{h,\;n}-\breve{e}_{h,\;n})\;+\\ & \tilde{\eta}_{h,\;n}\psi_{h,\;n}\nonumber+\frac{\breve{\alpha}_{h,\;n-1}-\breve{\alpha}_{hf,\;n-1}}{\Pi_{h,\;n-1}}\;-\\&\frac{{\alpha}_{h,\;n-1}-{\alpha}_{hf,\;n-1}}{\Pi_{h,\;n-1}}\nonumber\;-\\&{c}_{h,\;n}e_{h,\;n}\big)-\tilde{\eta}_{h,\;n}\dot{\hat{\eta}}_{h,\;n} \end{split} $$ 引理4[34]. 触发误差的上界可以表示为
$$ \begin{split} &|e_{h,\;n}-\breve{e}_{h,\;n}|\le\bar{\Theta}_{h}\nonumber\\ &\qquad\quad\left|\frac{\breve{\alpha}_{h,\;n-1}\breve{\alpha}_{hf,\;n-1}}{\Phi_{h,\;n-1}}-\frac{{\alpha}_{h,\;n-1}{\alpha}_{hf,\;n-1}}{\Phi_{h,\;n-1}}\right|\le \\ &\qquad\quad \frac{\Theta_{h,\;\alpha_{f}}+\Theta_{hf,\;n-1}}{\Phi_{h,\;n-1}} \end{split} $$ 接下来, 可得
$$ \begin{split} \dot{V}_{h,\;n}\le\;&-c_{h,\;n}e^{2}_{h,\;n}-\frac{1}{2}N_{h,\;n}\tilde{\eta}^{2}_{h,\;n}\;+\\ &\frac{1}{2}N_{h,\;n}{\eta}^{2}_{h,\;n}+\varsigma_{h,\;n} \end{split} $$ (46) 其中, $ \varsigma_{h,\;n}=\bar{\Theta}^{2}_{h}+\frac{{(\Theta_{h,\;\alpha_{f}}+\Theta_{hf,\;n-1})}^{2}}{\Phi^{2}_{h,\;n-1}}+p^{2}_{h}+\rho^{2}_{h} $.
注 4. 针对控制器−执行器环节和传感器−控制器环节, 本文设计混合双端事件触发机制, 可同时缓解双信道的通讯负担. 首先, 考虑在传感器−控制器环节上设置事件触发机制. 因为每次传输闭环系统的信息时, 输入信号都是根据输出反馈结果设置的, 所以输出信号的有效更新和更新次数是需要考虑的重要环节. 另外, 考虑控制器−执行器环节的资源节约. 通过在输入信号中设置触发项并进一步设置事件采样所需的条件, 实现节省通信资源的目的. 在多智能体系统或分布式系统中, 通信资源通常是有限的. 双端事件触发机制需要在有限的通信条件下, 确保信息的及时传输和系统的协调运行. 因此, 可以通过调整参数$ \nu_{h} $, $ m_{h} $, $ b_{h} $, $ \rho_{h} $, $ \mu_{h} $, $ \tau_{h} $ 和 $ \Theta_{h,\;\alpha_{f}} $实现对系统实时性能和通讯带宽占用率的有效平衡.
3.2 稳定性分析
定理1. 在假设1和假设 2下, 针对非线性多智能体系统 (1), 考虑混合双端事件触发机制 (10), (11), (44) 和模糊状态观测器(15), 设计自适应律 (31), (38) 和 (45), 虚拟控制器 (30), (37) 和分布式控制器 (43)可以使得闭环系统内的所有信号是半全局一致最终有界的.
此外, 对于 $ \forall \chi>0 $, 设计的参数满足
$$ \begin{align} \begin{aligned} \underset{t\to\infty}\lim||y-{y}_{r}||\leq\chi \end{aligned} \end{align} $$ (47) 证明. 为证明整体闭环系统的稳定性, 选择总Lyapunov函数为
$$ \begin{align} \begin{aligned} V_{h}=\sum_{m=1}^n V_{h,\;m}+V_{h,\;0} \end{aligned} \end{align} $$ (48) 根据式 (22), (33), (40) 和 (46), 可得
$$ \begin{split} \dot{V}_{h}\le\;&\xi_{0}||\Delta_{h}||^{2} + \frac{1}{2}||Y_{h}||^{2}||\tilde{\eta}^\text{T}_{h,\;n}\tilde{\eta}_{h,\;n} + \frac{1}{2}||Y_{h}||^{2}||\varepsilon^{*}_{h}||^{2}\;+\\ &\frac{1}{2}||Y_{h}||^{2}\sum_{m=1}^{n} A_{h,\;m} \tilde{\eta}^\text{T}_{h,\;m}\tilde{\eta}_{h,\;m}+\frac{1}{2}\varepsilon^{2}_{h,\;1}\rho_{h}\;-\\ &\sum_{m=1}^{n}\frac{1}{2}N_{h,\;m}\tilde{\eta}^{2}_{h,\;m}+\sum_{m=1}^{n}\frac{1}{2}N_{h,\;m}{\eta}^{2}_{h,\;m}\;+\\ &\sum_{m=1}^{n-1}\frac{1}{2}\dot{\alpha}^{2}_{h,\;m}-\sum_{m=2}^{n-1}\left(\frac{1}{\upsilon_{h,\;1}}-\frac{1}{2}\right){\vartheta}^{2}_{h,\;m}\;-\\ & {c}_{h,\;1}s^{2}_{h,\;1}-\sum_{m=2}^{n-1} {c}_{h,\;m}e^{2}_{h,\;m}-{c}_{h,\;n}e^{2}_{h,\;n}+\varsigma_{h,\;n}\;-\\ & \left(\frac{1}{\upsilon_{h,\;1}}-\frac{1}{2}\rho_{h}\right){\vartheta}^{2}_{h,\;1}\le-\partial_{h}V_{h}+\omega_{h}\\[-1pt] \end{split} $$ (49) 其中,
$$ \begin{split} \partial_{h}=\;&\;\min\bigg\{ \bar{c}_{h,\;1},\; \bar{c}_{h,\;m},\; {c}_{h,\;n},\; \frac{N_{h,\;m}}{2},\;\\& \left(\frac{1}{\upsilon_{h,\;1}}-\frac{1}{2}\right),\; \left(\frac{1}{\upsilon_{h,\;1}}-\frac{\rho_{h}}{2}\right) \bigg\} \end{split} $$ 且
$$ \begin{split} \omega_{h}=\;&\;\sum_{m=1}^{n}\frac{1}{2}N_{h,\;m}{\eta}^{2}_{h,\;m}+\frac{1}{2}\varepsilon^{2}_{h,\;1}\rho_{h}\;+\\ &\varsigma_{h,\;n}+\sum_{m=1}^{n-1}\frac{1}{2}\dot{\alpha}^{2}_{h,\;m}\nonumber \end{split} $$ 在不等式(49)两边同时乘以 $ \text{e}^{\partial_{h}t} $, 并在定义域 $ [0,\;t] $ 上同时积分, 可得
$$ \begin{align} \begin{aligned} 0\le V(t)\le \text{e}^{-\partial_{h}t}V(0)+\frac{\omega_{h}}{\partial_{h}}(1-\text{e}^{-\partial_{h}t}) \end{aligned} \end{align} $$ (50) 根据 $ V $ 的定义和式(50), 可知
$$ \begin{align} \begin{aligned} ||s_{.1}||^2\le2\text{e}^{-\partial_{h}t}V_{0}+\frac{\omega_{h}}{\partial_{h}}(1-\text{e}^{-\partial_{h}t}) \end{aligned} \end{align} $$ (51) 针对 $ \forall\chi>0 $, 基于 $ \partial_{h} $ 和 $ \omega_{h} $ 的定义, 选择合适的参数, 可得如下关系式
$$ \begin{align} \frac{\omega_{h}}{\partial_{h}}\le\frac{\chi^{2}}{2}(\sigma({\cal{L}}+{\cal{B}}))^2 \end{align} $$ (52) 并且, 根据引理2, 当 $ t\to\infty $ 时, 不等式 (47) 成立.
□ 3.3 芝诺行为分析
本节需要证明提出的事件触发机制的间隔时间是有下界的, 即同时排除芝诺行为发生的可能性. 对于 $ \varkappa=1,\;\cdots,\;n $, 定义
$$\left\{\begin{split} &\sigma_{h,\;\varkappa}(t)=\hat{x}_{h,\;\varkappa}(t)-\breve{\hat{x}}_{h,\;\varkappa}(t),\; \quad t\in[t^{x}_{h,\;k},\;t^{x}_{h,\;k+1})\nonumber\\ &\sigma_{h,\;u}(t)=u_{h}(t)-\breve{u}_{h}(t),\; \quad\quad\quad t\in[t^{u}_{h,\;k},\;t^{u}_{h,\;k+1})\nonumber\\ &\sigma_{hf,\;\varkappa-1}(t)={\dot{\alpha}}_{hf,\;\varkappa-1}(t)-{\dot{\breve{\alpha}}}_{hf,\;\varkappa-1}(t),\;\\ &\qquad\qquad\qquad\qquad\qquad\qquad\quad\quad t\in[t^{\alpha_{f}}_{h,\;k},\;t^{\alpha_{f}}_{h,\;k+1})\nonumber \end{split}\right. $$ 计算以上变量的导数值为
$$ \begin{align} \begin{aligned} &\frac{{\rm{d}}}{{\rm{d}}t}|\sigma_{h,\;\varkappa}|=\frac{{\rm{d}}}{{\rm{d}}t}(\sigma_{h,\;\varkappa}*\sigma_{h,\;\varkappa})^{\frac{1}{2}}\nonumber=\\ &\qquad\hbox{sign}(\hat{x}_{h,\;n}-\breve{\hat{x}}_{h,\;n})\dot{\hat{x}}_{h,\;n}\le|\dot{\hat{x}}_{h,\;n}|\nonumber\\& \frac{{\rm{d}}}{{\rm{d}}t}|\sigma_{h,\;u}|=\frac{{\rm{d}}}{{\rm{d}}t}(\sigma_{h,\;u}*\sigma_{h,\;u})^{\frac{1}{2}}\nonumber=\\ &\qquad\hbox{sign}(u_{h}-\breve{u}_{h})\dot{u}_{h}\le|\dot{u}_{h}|\nonumber\\ &\frac{{\rm{d}}}{{\rm{d}}t}|\sigma_{hf,\;\varkappa-1}|=\frac{{\rm{d}}}{{\rm{d}}t}(\sigma_{hf,\;\varkappa-1}*\sigma_{hf,\;\varkappa-1})^{\frac{1}{2}}\nonumber=\\ &\qquad\hbox{sign}(\dot{\alpha}_{hf,\;g-1}-\dot{\breve{\alpha}}_{hf,\;g-1})\ddot{\alpha}_{hf,\;g-1}\nonumber \le\\ &\qquad|\ddot{\alpha}_{hf,\;g-1}| \end{aligned} \end{align} $$ 其中, $ \dot{u}_{h} $和$ \ddot{\alpha}_{hf,\;g-1} $与$ \hat{x}_{h,\;g} $, $ e_{h,\;g} $, $ \tilde{\eta}_{h,\;g} $和$ \dot{\alpha}_{h,\;g} $信号有关.
以上变量满足如下的关系式: $ |\dot{\hat{x}}_{h,\;m}|\;\le\;\digamma_{h} $, $ |\dot{u}_{h}|\le\digamma_{\breve{u}_{h}} $ 和 $ |\ddot{\digamma}_{h,\;\alpha_{f}}|\le\digamma_{h,\;\alpha_{f}} $. 值得注意的是, 在 $ t^{x}_{h,\;k} $, $ t^{u}_{h,\;k} $ 和 $ t^{\alpha_{f}}_{h,\;k} $ 时刻, 三个变量满足 $ \sigma_{h,\,\varkappa}(t^{x}_{h,\,k}) = 0 $, $ \sigma_{h,\;\varkappa}(t^{u}_{h,\;k})=0 $ 和 $ \sigma_{hf,\;\varkappa-1}(t^{\alpha_{f}}_{h,\;k})=0 $.
随后, 可得
$$\begin{split} &\lim_{t\rightarrow t^{x}_{h,\;k}}\sigma_{h,\;\varkappa}(t)=\nu_{h}+ m_{h}\text{e}^{-b_{h}t^{x}_{h,\;k}} \\ &\lim_{t\rightarrow t^{u}_{h,\;k}} \sigma_{h,\;\varkappa}(t)=\rho_{h}+\mu_{h}\text{e}^{-\tau_{h}t^{u}_{h,\;k}}\\ &\lim_{t\rightarrow t^{\alpha_{f}}_{h,\;k}} \sigma_{h,\;\varkappa}(t)=\Theta_{h,\;\alpha_{f}} \end{split} $$ 因此, 得到本文提出的事件触发机制间隔的最小界限值为
$$ \begin{align} t^{x}_{h,\;k+1}-t^{x}_{h,\;k}\ge\frac{\nu_{h}+m_{h}\text{e}^{-b_{h}t^{x}_{h,\;k}}}{\digamma_{h}} \end{align} $$ (53) $$ \begin{align} t^{u}_{h,\;k+1}-t^{u}_{h,\;k}\ge\frac{\rho_{h}+\mu_{h}\text{e}^{-\tau_{h}t^{u}_{h,\;k}}}{\digamma_{\breve{u}_{h}}} \end{align} $$ (54) $$ \begin{align} t^{\alpha_{f}}_{h,\;k+1}-t^{\alpha_{f}}_{h,\;k}\ge\frac{\Theta_{h,\;\alpha_{f}}}{\digamma_{h,\;\alpha_{f}}} \end{align} $$ (55) 通过以上分析可以得出, 本文提出的事件触发机制不会发生芝诺行为.
4. 仿真结果
一些仿真结果验证了本文控制方案的有效性.
图1是本文所考虑的通讯拓扑结构, 其表明4个跟随者与1个领导者之间的信息传输关系. 基于图1, 邻接矩阵 $ \bar{{\cal{A}}} $ 和拉普拉斯矩阵 $ {\cal{L}} $ 表示如下:
$$ \begin{align} \begin{aligned} \bar{{\cal{A}}}= \begin{bmatrix} 0&0&0&1\\ 1&0&0&0\\ 0&1&0&0\\ 0&1&1&0\nonumber \end{bmatrix} ,\; {\cal{L}}= \begin{bmatrix} \;\;\,1&\;\;\;\,0&\;\;\,0&-1\\ -1&\;\;1&\;\;\,0&\;\;0\\ \,\;\;0&-1&\;\;\,1&\;\;0\\ \;\;\,0&-1&-1&\;\;2\nonumber \end{bmatrix} \end{aligned} \end{align} $$ 对于 $ h=1,\;2,\;3,\;4 $, 挑选一组强阻尼系统, 其动态模型为
$$ \begin{align} \begin{aligned} \Lambda_{h}=\frac{\pi_{h}}{M_{h}L^{2}_{h}}\Lambda_{h}-\frac{v_{h}\bar{g}}{L_{h}}\sin(\Lambda_{h})+u_{h} \end{aligned} \end{align} $$ (56) 其中, $ L_{h} $ 是摆长. $ \Lambda_{h} $ 是从垂直向下的位置逆时针测量的杆的角度. $ u_{h} $ 表示传动转矩. $ \bar{g} $ 为重力加速度. $ v_{h} $ 定义为恢复转矩系数. $ \pi $ 为阻尼系数. $ M_{h} $ 为钟摆的质量.
针对$ h = 1,\;2,\;3,\;4 $, 设$ x_{h,\;1} = \Lambda_{h} $ 和 $ {x}_{h,\;2}=\dot{\Lambda}_{h} $, 其动力学方程可转化为
$$ \left\{\begin{aligned} &\dot{x}_{h,\;1}=x_{h,\;2}\nonumber\\ &\dot{x}_{h,\;2}=u_{h}-\frac{\pi_{h}}{M_{h}L^{2}_{h}}x_{h,\;2}-\frac{v_{h}\bar{g}}{L_{h}}\sin(x_{h,\;1})\nonumber\\ & y_{h}=x_{h,\;1} \end{aligned} \right. $$ 强阻尼系统的初始值矩阵选择为 $ x_{h}(0) = [0.1,\; 0.1\;]^\text{T} $. 自适应参数的初始值为 $ \hat{\eta}\;_{h,\;1}\; (0) \;=\; 0.1 $ 和 $ \hat{\eta}\;_{h,\,2} (0) = 0.1 $. 挑选参数值分别为 $ e_{h,\;1,\;0,\;\min}=0.5 $, $ e_{h,\;1,\;\infty,\;\max}=0.5 $, $ e_{h,\;1,\;\infty,\;\min}=0.3 $, $ e_{h,\;1,\;0,\;\max}=0.5 $, $ o_{h}=-1 $, $ c_{h,\;1}=c_{h,\;2}=50 $, $ \phi_{h,\;2}=2 $, $ N_{h,\;1}= N_{h,\;2}= 0.1 $, $ p_{h}=0.1 $, $ m_{h}=0.1 $, $ b_{h}=0.1 $, $ \rho_{h}=10 $, $ \mu_{h}=2 $, $ \tau_{h}=2, $ $ M_{h}=1, $ $ L_{h}=1, $ $ \bar{g}=9.8 $, $ v_{h}=\frac{1}{9.8} $ 和 $ \pi=-0.25 $.
图2表明4个跟随者的输出轨迹与既定的领导者轨迹是一致的, 设计的控制算法可使强阻尼系统的输出稳定于给定的参考信号. 基于此, 可以看出考虑的控制目标得到实现, 控制策略有效地实现分布式强阻尼系统的一致性目标.
图 2 4个跟随者和1个领导者的输出轨迹Fig. 2 The output trajectories of the four followers and one leader在图3中, 将一致性误差的值域限制在预设的范围中, 误差输出轨迹小于设计的规定性能预设边界$ (-0.5,\;0.3) $, 表明良好的规定性能效果. 图4为在双端触发框架下的控制输入曲线, 从图4中可以看出输入信号$ u_{h} $是有界的.
图5和图6凸显了模糊逻辑系统权重参数的变化过程, 证明本文所有自适应律参数是有界的, 并根据上文展示的控制性能, 说明本文控制框架针对实际系统具备有效性与适用性. $ \hat{\eta}_{h,\;g} $ 表示模糊逻辑系统的权重, 且用来调整模糊逻辑系统对非线性函数$ \zeta_{h,\;g}(\bar{x}_{h,\;g}) $ 的逼近效果.
图 5 自适应律参数$\hat{\eta}_{h,\;1}$的轨迹Fig. 5 The trajectories of the adaptive law parameters $\hat{\eta}_{h,\;1}$
图 6 自适应律参数$\hat{\eta}_{h,\;2}$的轨迹Fig. 6 The trajectories of the adaptive law parameters $\hat{\eta}_{h,\;2}$图7为观测误差的数值变化轨迹. 同时, 依据前文的设计过程, 可知考虑的观测器仅使用相对输出信息进行反馈, 能够大幅度提高观测器的实用性. 图8展示了所提出事件触发机制的触发间隔. 同时, 也说明所提出的混合双端事件触发机制具有可节省控制器环节通信资源的优势. 以智能体1为例, 正常迭代次数为
3000 次, 通过本文事件触发的设计后, 控制器的更新次数为908 次, 节省了69.7%的通讯资源. 针对多智能体系统, 智能体之间在通讯网络下进行信息传输, 当智能体数量增多时, 必会造成一定程度的通讯压力, 这足以可见本文设计事件触发机制的重要性. 并且, 针对多渠道通讯网络, 本文同时降低了控制器−执行器环节和传感器−控制器环节的通讯负担.
图 7 观测误差$\Delta_{h,\;2} $的变化情况Fig. 7 The changes in observation errors $\Delta_{h,\;2}$5. 结束语
本文研究了双端事件触发自适应模糊跟踪控制问题. 针对控制器−执行器和传感器−控制器环节, 提出基于状态触发机制和控制器触发机制的混合双端分布式事件触发机制, 并且设计一种改进的状态触发机制, 首次将估计的状态信号作为触发信号来达到节约通讯资源的目的. 最终, 一些仿真结果证明了提出控制方案的有效性. 在未来的研究工作中, 我们将致力于探索电力系统控制需求, 并将多种事件触发控制策略融合实际系统的需要, 以满足智能化、高效化与绿色化的能源转型目标.
-
图 3 “感知−模拟−执行”一体化机制框架
Fig. 3 The framework of the integrated perception-simulation-execution mechanism
图 10 NerveNet从每个节点的观测向量中获取信息, 通过多次计算相邻节点间的信息更新节点的隐藏状态, 最后在输出模型中收集每个控制器的输出形成优化策略[95]
Fig. 10 NerveNet obtains the information from the observation vectors of each node, updates the hidden states of the nodes by calculating the information between adjacent nodes multiple times, and finally collects the output of each controller in the output model to form an optimization strategy[95]
图 12 具身智能增强的机器人系统研究框架
Fig. 12 The research framework of robot systems with enhanced embodied intelligence
图 13 具身智能增强的自动驾驶系统框架
Fig. 13 The framework of auto drive system with enhanced embodied intelligence
表 1 具身智能研究现状
Table 1 The current status of embodied intelligence research
名称 年份 特点 优劣 BigDog 2009 由波士顿动力公司制造, 能够在崎岖不平的地形上行走, 并保持稳定, 展示了在复杂环境中移动的能力 具有强大的越野能力和高负载能力, 能适应复杂环境, 但采用噪音较大的内燃机动力源 Atlas 2013 由波士顿动力公司制造, 具备高度灵活性和稳定性的人形机器人, 能够进行跑步、跳跃和攀爬等复杂动作, 标志着人形机器人在运动控制和灵活性方面的显著进步 具备高度灵活性和稳定性, 能够执行复杂动作, 但开发和制造成本较高 DQN算法 2014 DeepMind公司开发的DQN (Deep Q-Network)算法首次将深度学习与强化学习相结合, 使智能体在多种视频游戏中超越人类表现, 这一算法为具身智能提供了新的学习和决策方法 可在无监督环境中通过与环境的互动进行学习, 提高了适应性, 但需要大量数据和计算资源进行训练, 运行成本高 AlphaGo 2016 DeepMind的AlphaGo战胜了世界围棋冠军李世石, 这一里程碑事件展示了智能体在复杂策略游戏中的超人表现, 推动了具身智能在复杂决策问题上的研究 结合深度学习和蒙特卡罗树搜索, 实现高效决策和自我优化, 但高计算成本和领域局限性限制了其广泛应用的可能 Walker 2018 优必选公司发布了Walker机器人, 这是一款双足仿人服务机器人, 展示了在家居和服务领域的应用潜力 具备双足行走能力和多功能性, 但高成本和续航时间有限,限制了长时间工作和普及应用 Stretch 2021 波士顿动力公司推出的Stretch机器人, 专为仓库操作设计, 展示了在物流和仓储领域的巨大应用前景 专为仓库操作设计, 提升了仓库内搬运任务的效率, 但泛化到其他领域工作的能力较低 Optimus 2024 特斯拉公司发布了Optimus人形机器人, 旨在解决劳动力短缺问题, 展示了未来具身智能在生产和日常生活中的广泛应用潜力 具备高度自主性和广泛应用前景, 但高成本和复杂技术性限制了普及性 
下载: 导出CSV
-
[1] 张钹, 朱军, 苏航. 迈向第三代人工智能. 中国科学: 信息科学, 2020, 50(9): 1281−1302 doi: 10.1360/SSI-2020-0204Zhang Bo, Zhu Jun, Su Hang. Toward the third generation of artificial intelligence. Scientia Sinica Informationis, 2020, 50(9): 1281−1302 doi: 10.1360/SSI-2020-0204 [2] Jin D D, Zhang L. Embodied intelligence weaves a better future. Nature Machine Intelligence, 2020, 2(11): 663−664 doi: 10.1038/s42256-020-00250-6 [3] Gupta A, Savarese S, Ganguli S, Li F F. Embodied intelligence via learning and evolution. Nature Communications, 2021, 12(1): Article No. 5721 doi: 10.1038/s41467-021-25874-z [4] Turing A M. Computing machinery and intelligence. Creative Computing, 1980, 6(1): 44−53 [5] Howard D, Eiben A E, Kennedy D F, Mouret J B, Valencia P, Winkler D. Evolving embodied intelligence from materials to machines. Nature Machine Intelligence, 2019, 1(1): 12−19 doi: 10.1038/s42256-018-0009-9 [6] Shen T Y, Sun J L, Kong S H, Wang Y T, Li J J, Li X, et al. The journey/DAO/TAO of embodied intelligence: From large models to foundation intelligence and parallel intelligence. IEEE/CAA Journal of Automatica Sinica, 2024, 11(6): 1313−1316 doi: 10.1109/JAS.2024.124407 [7] 沈甜雨, 李志伟, 范丽丽, 张庭祯, 唐丹丹, 周美华, 等. 具身智能驾驶: 概念、方法、现状与展望. 智能科学与技术学报, 2024, 6(1): 17−32 doi: 10.11959/j.issn.2096-6652.202404Shen Tian-Yu, Li Zhi-Wei, Fan Li-Li, Zhang Ting-Zhen, Tang Dan-Dan, Zhou Mei-Hua, et al. Embodied intelligent driving: Concept, methods, the state of the art and beyond. Chinese Journal of Intelligent Science and Technology, 2024, 6(1): 17−32 doi: 10.11959/j.issn.2096-6652.202404 [8] Ichter B, Brohan A, Chebotar Y, Finn C, Hausman K, Herzog A, et al. Do as I can, not as I say: Grounding language in robotic affordances. In: Proceedings of the 6th Conference on Robot Learning. Auckland, New Zealand: PMLR, 2023. 287–318 [9] Shah D, Osiński B, Ichter B, Levine S. LM-Nav: Robotic navigation with large pre-trained models of language, vision, and action. In: Proceedings of the 6th Conference on Robot Learning. Auckland, New Zealand: PMLR, 2023. 492–504 [10] Qiao H, Zhong S L, Chen Z Y, Wang H Z. Improving performance of robots using human-inspired approaches: A survey. Science China Information Sciences, 2022, 65(12): Article No. 221201 doi: 10.1007/s11432-022-3606-1 [11] Cao L B. AI robots and humanoid AI: Review, perspectives and directions. arXiv preprint arXiv: 2405.15775, 2024.Cao L B. AI robots and humanoid AI: Review, perspectives and directions. arXiv preprint arXiv: 2405.15775, 2024. [12] Duan J F, Yu S, Tan H L, Zhu H Y, Tan C. A survey of embodied AI: From simulators to research tasks. IEEE Transactions on Emerging Topics in Computational Intelligence, 2022, 6(2): 230−244 doi: 10.1109/TETCI.2022.3141105 [13] 刘华平, 郭迪, 孙富春, 张新钰. 基于形态的具身智能研究: 历史回顾与前沿进展. 自动化学报, 2023, 49(6): 1131−1154Liu Hua-Ping, Guo Di, Sun Fu-Chun, Zhang Xin-Yu. Morphology-based embodied intelligence: Historical retrospect and research progress. Acta Automatica Sinica, 2023, 49(6): 1131−1154 [14] Minsky M. Society of Mind. New York: Simon and Schuster, 1988. [15] Varela F J, Thompson E, Rosch E. The Embodied Mind: Cognitive Science and Human Experience. Cambridge: MIT Press, 1993. [16] Pfeifer R, Bongard J. How the Body Shapes the Way We Think: A New View of Intelligence. Cambridge: MIT Press, 2006. [17] Deisenroth M P, Fox D, Rasmussen C E. Gaussian processes for data-efficient learning in robotics and control. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2015, 37(2): 408−423 doi: 10.1109/TPAMI.2013.218 [18] Hu Y H, Yang J Z, Chen L, Li K Y, Sima C, Zhu X Z. Planning-oriented autonomous driving. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Vancouver, Canada: IEEE, 2023. 17853–17862 [19] Hu C Y, Zheng H, Li K, Xu J Y, Mao W B, Luo M C, et al. FusionFormer: A multi-sensory fusion in bird＇s-eye-view and temporal consistent Transformer for 3D objection. arXiv preprint arXiv: 2309.05257, 2023.Hu C Y, Zheng H, Li K, Xu J Y, Mao W B, Luo M C, et al. FusionFormer: A multi-sensory fusion in bird's-eye-view and temporal consistent Transformer for 3D objection. arXiv preprint arXiv: 2309.05257, 2023. [20] Yin T W, Zhou X Y, Krähenbühl P. Multimodal virtual point 3D detection. arXiv preprint arXiv: 2111.06881, 2021.Yin T W, Zhou X Y, Krähenbühl P. Multimodal virtual point 3D detection. arXiv preprint arXiv: 2111.06881, 2021. [21] Wu X P, Peng L, Yang H H, Xie L, Huang C X, Deng C Q. Sparse fuse dense: Towards high quality 3D detection with depth completion. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). New Orleans, USA: IEEE, 2022. 5408–5417 [22] Wu H, Wen C L, Shi S S, Li X, Wang C. Virtual sparse convolution for multimodal 3D object detection. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Vancouver, Canada: IEEE, 2023. 21653–21662 [23] Liu Z J, Tang H T, Amini A, Yang X Y, Mao H Z, Rus D L. BEVFusion: Multi-task multi-sensor fusion with unified bird's-eye view representation. In: Proceedings of the IEEE International Conference on Robotics and Automation (ICRA). London, UK: IEEE, 2023. 2774–2781 [24] Wei M, Li J C, Kang H Y, Huang Y J, Lu J G. BEV-CFKT: A LiDAR-camera cross-modality-interaction fusion and knowledge transfer framework with Transformer for BEV 3D object detection. Neurocomputing, 2024, 582: Article No. 127527 doi: 10.1016/j.neucom.2024.127527 [25] Drews F, Feng D, Faion F, Rosenbaum L, Ulrich M, Gläser C. DeepFusion: A robust and modular 3D object detector for LiDARs, cameras and radars. In: Proceedings of the IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS). Kyoto, Japan: IEEE, 2022. 560–567 [26] Xie B Q, Yang Z M, Yang L, Wei A L, Weng X X, Li B. AMMF: Attention-based multi-phase multi-task fusion for small contour object 3D detection. IEEE Transactions on Intelligent Transportation Systems, 2023, 24(2): 1692−1701 [27] Chiu H K, Li J, Ambruş R, Bohg J. Probabilistic 3D multi-modal, multi-object tracking for autonomous driving. In: Proceedings of the IEEE International Conference on Robotics and Automation (ICRA). Xi'an, China: IEEE, 2021. 14227–14233 [28] Tian Y L, Zhang X J, Wang X, Xu J T, Wang J G, Ai R. ACF-Net: Asymmetric cascade fusion for 3D detection with LiDAR point clouds and images. IEEE Transactions on Intelligent Vehicles, 2024, 9(2): 3360−3371 doi: 10.1109/TIV.2023.3341223 [29] Zhang P, Zhang B, Zhang T, Chen D, Wang Y, Wen F. Prototypical pseudo label denoising and target structure learning for domain adaptive semantic segmentation. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Nashville, USA: IEEE, 2021. 12409−12419 [30] Lee S, Cho S, Im S. DRANet: Disentangling representation and adaptation networks for unsupervised cross-domain adaptation. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Nashville, USA: IEEE, 2021. 15247–15256 [31] Oza P, Sindagi V A, Vs V, Patel V M. Unsupervised domain adaptation of object detectors: A survey. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2024, 46(6): 4018−4040 doi: 10.1109/TPAMI.2022.3217046 [32] Ganin Y, Ustinova E, Ajakan H, Germain P, Larochelle H, Laviolette F, et al. Domain-adversarial training of neural networks. Journal of Machine Learning Research, 2016, 17(59): 1−35 [33] Cai Q, Pan Y W, Ngo C W, Tian X M, Duan L Y, Yao T. Exploring object relation in mean teacher for cross-domain detection. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Long Beach, USA: IEEE, 2019. 11449–11458 [34] Li C X, Chan S H, Chen Y T. DROID: Driver-centric risk object identification. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2023, 45(11): 13683−13698 [35] Cheng Z Y, Lu J, Ding H, Li Y X, Bai H J, Zhang W H. A superposition assessment framework of multi-source traffic risks for mega-events using risk field model and time-series generative adversarial networks. IEEE Transactions on Intelligent Transportation Systems, 2023, 24(11): 12736−12753 doi: 10.1109/TITS.2023.3290165 [36] Wang X W, Alonso-Mora J, Wang M. Probabilistic risk metric for highway driving leveraging multi-modal trajectory predictions. IEEE Transactions on Intelligent Transportation Systems, 2022, 23(10): 19399−19412 doi: 10.1109/TITS.2022.3164469 [37] Caleffi F, Anzanello M J, Cybis H B B. A multivariate-based conflict prediction model for a Brazilian freeway. Accident Analysis & Prevention, 2017, 98: 295−302 [38] Rombach R, Blattmann A, Lorenz D, Esser P, Ommer B. High-resolution image synthesis with latent diffusion models. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). New Orleans, USA: IEEE, 2022. 10674–10685 [39] Saharia C, Chan W, Saxena S, Lit L, Whang J, Denton E, et al. Photorealistic text-to-image diffusion models with deep language understanding. In: Proceedings of the 36th International Conference on Neural Information Processing Systems. New Orleans, USA: Curran Associates Inc., 2022. Article No. 2643 [40] Ramesh A, Pavlov M, Goh G, Gray S, Voss C, Radford A, et al. Zero-shot text-to-image generation. arXiv preprint arXiv: 2102.12092, 2021.Ramesh A, Pavlov M, Goh G, Gray S, Voss C, Radford A, et al. Zero-shot text-to-image generation. arXiv preprint arXiv: 2102.12092, 2021. [41] Ramesh A, Dhariwal P, Nichol A, Chu C, Chen M. Hierarchical text-conditional image generation with CLIP latents. arXiv preprint arXiv: 2204.06125, 2022.Ramesh A, Dhariwal P, Nichol A, Chu C, Chen M. Hierarchical text-conditional image generation with CLIP latents. arXiv preprint arXiv: 2204.06125, 2022. [42] Reed S, Zolna K, Parisotto E, Colmenarejo S G, Novikov A, Barth-Maron G, et al. A generalist agent. arXiv preprint arXiv: 2205.06175, 2022. [43] Niemeyer M, Geiger A. GIRAFFE: Representing scenes as compositional generative neural feature fields. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Nashville, USA: IEEE, 2021. 11448–11459 [44] Abou-Chakra J, Rana K, Dayoub F, Sünderhauf N. Physically embodied Gaussian splatting: Embedding physical priors into a visual 3D world model for robotics [Online], available: https://eprints.qut.edu.au/247354/, November 9, 2024Abou-Chakra J, Rana K, Dayoub F, Sünderhauf N. Physically embodied Gaussian splatting: Embedding physical priors into a visual 3D world model for robotics [Online], available: https://eprints.qut.edu.au/247354/, November 9, 2024 [45] Hart P E, Nilsson N J, Raphael B. A formal basis for the heuristic determination of minimum cost paths. IEEE Transactions on Systems Science and Cybernetics, 1968, 4(2): 100−107 doi: 10.1109/TSSC.1968.300136 [46] Karaman S, Frazzoli E. Sampling-based algorithms for optimal motion planning. The International Journal of Robotics Research, 2011, 30(7): 846−894 doi: 10.1177/0278364911406761 [47] Koenig S, Likhachev M. D* lite. In: Proceedings of the 18th National Conference on Artificial Intelligence. Edmonton, Canada: AAAI, 2002. 476–483 [48] Holland J H. Adaptation in Natural and Artificial Systems: An Introductory Analysis With Applications to Biology, Control, and Artificial Intelligence. Cambridge: MIT Press, 1992. [49] Kennedy J, Eberhart R. Particle swarm optimization. In: Proceedings of the International Conference on Neural Networks. Perth, Australia: IEEE, 1995. 1942–1948 [50] Dolgov D, Thrun S, Montemerlo M, Diebel J. Practical search techniques in path planning for autonomous driving. Ann Arbor, 2008, 1001(48105): 18−80 [51] Webb D J, van den Berg J. Kinodynamic RRT*: Asymptotically optimal motion planning for robots with linear dynamics. In: Proceedings of the IEEE International Conference on Robotics and Automation. Karlsruhe, Germany: IEEE, 2013. 5054–5061 [52] Konda V R, Tsitsiklis J N. Onactor-critic algorithms. SIAM Journal on Control and Optimization, 2003, 42(4): 1143−1166 doi: 10.1137/S0363012901385691 [53] Watkins C J C H, Dayan P. Technical note: Q-learning. Machine Learning, 1992, 8(3−4): 279−292 doi: 10.1007/BF00992698 [54] Mnih V, Badia A P, Mirza M, Graves A, Lillicrap T, Harley T, et al. Asynchronous methods for deep reinforcement learning. In: Proceedings of the 33rd International Conference on Machine Learning. New York, USA: JMLR.org, 2016. 1928–1937 [55] Deisenroth M P, Rasmussen C E. PILCO: A model-based and data-efficient approach to policy search. In: Proceedings of the 28th International Conference on Machine Learning. Bellevue, USA: Omnipress, 2011. 465–472 [56] Ho J, Ermon S. Generative adversarial imitation learning. In: Proceedings of the 30th International Conference on Neural Information Processing Systems. Barcelona, Spain: Curran Associates Inc., 2016. 4572–4580 [57] Borase R P, Maghade D K, Sondkar S Y, Pawar S N. A review of PID control, tuning methods and applications. International Journal of Dynamics and Control, 2021, 9(2): 818−827 doi: 10.1007/s40435-020-00665-4 [58] Dörfler F, Tesi P, de Persis C. On the role of regularization in direct data-driven LQR control. In: Proceedings of the 61st Conference on Decision and Control (CDC). Cancun, Mexico: IEEE, 2022. 1091–1098 [59] Berberich J, Koch A, Scherer C W, Allgöwer F. Robust data-driven state-feedback design. In: Proceedings of the American Control Conference (ACC). Denver, USA: IEEE, 2020. 1532–1538 [60] Nubert J, Köhler J, Berenz V, Allgower F, Trimpe S. Safe and fast tracking on a robot manipulator: Robust MPC and neural network control. IEEE Robotics and Automation Letters, 2020, 5(2): 3050−3057 doi: 10.1109/LRA.2020.2975727 [61] Liu Y J, Zhao W, Liu L, Li D P, Tong S C, Chen C L P. Adaptive neural network control for a class of nonlinear systems with function constraints on states. IEEE Transactions on Neural Networks and Learning Systems, 2023, 34(6): 2732−2741 doi: 10.1109/TNNLS.2021.3107600 [62] Phan D, Bab-Hadiashar A, Fayyazi M, Hoseinnezhad R, Jazar R N, Khayyam H. Interval type 2 fuzzy logic control for energy management of hybrid electric autonomous vehicles. IEEE Transactions on Intelligent Vehicles, 2021, 6(2): 210−220 doi: 10.1109/TIV.2020.3011954 [63] Omidvar M N, Li X D, Mei Y, Yao X. Cooperative co-evolution with differential grouping for large scale optimization. IEEE Transactions on Evolutionary Computation, 2014, 18(3): 378−393 doi: 10.1109/TEVC.2013.2281543 [64] Gad A G. Particle swarm optimization algorithm and its applications: A systematic review. Archives of Computational Methods in Engineering, 2022, 29(5): 2531−2561 doi: 10.1007/s11831-021-09694-4 [65] Foderaro G, Ferrari S, Wettergren T A. Distributed optimal control for multi-agent trajectory optimization. Automatica, 2014, 50(1): 149−154 doi: 10.1016/j.automatica.2013.09.014 [66] Bellemare M G, Dabney W, Munos R. A distributional perspective on reinforcement learning. In: Proceedings of the 34th International Conference on Machine Learning. Sydney, Australia: PMLR, 2017. 449–458 [67] Dong G W, Li H Y, Ma H, Lu R Q. Finite-time consensus tracking neural network FTC of multi-agent systems. IEEE Transactions on Neural Networks and Learning Systems, 2021, 32(2): 653−662 doi: 10.1109/TNNLS.2020.2978898 [68] Wang Q Y, Liu K X, Wang X, Wu L L, Lü J H. Leader-following consensus of multi-agent systems under antagonistic networks. Neurocomputing, 2020, 413: 339−347 doi: 10.1016/j.neucom.2020.07.006 [69] Cai Z H, Wang L H, Zhao J, Wu K, Wang Y X. Virtual target guidance-based distributed model predictive control for formation control of multiple UAVs. Chinese Journal of Aeronautics, 2020, 33(3): 1037−1056 doi: 10.1016/j.cja.2019.07.016 [70] Harvey I, Husbands P, Cliff D, Thompson A, Jakobi N. Evolutionary robotics: The Sussex approach. Robotics and Autonomous Systems, 1997, 20(2−4): 205−224 doi: 10.1016/S0921-8890(96)00067-X [71] Kirkpatrick J, Pascanu R, Rabinowitz N, Veness J, Desjardins G, Rusu A A, et al. Overcoming catastrophic forgetting in neural networks. Proceedings of the National Academy of Sciences of the United States of America, 2017, 114(13): 3521−3526 [72] Liu X L, Masana M, Herranz L, van de Weijer J, López A M, Bagdanov A D. Rotate your networks: Better weight consolidation and less catastrophic forgetting. In: Proceedings of the 24th International Conference on Pattern Recognition (ICPR). Beijing, China: IEEE, 2018. 2262–2268 [73] Hsu Y C, Liu Y C, Ramasamy A, Kira Z. Re-evaluating continual learning scenarios: A categorization and case for strong baselines. arXiv preprint arXiv: 1810.12488, 2019.Hsu Y C, Liu Y C, Ramasamy A, Kira Z. Re-evaluating continual learning scenarios: A categorization and case for strong baselines. arXiv preprint arXiv: 1810.12488, 2019. [74] Hinton G, Vinyals O, Dean J. Distilling the knowledge in a neural network. arXiv preprint arXiv: 1503.02531, 2015.Hinton G, Vinyals O, Dean J. Distilling the knowledge in a neural network. arXiv preprint arXiv: 1503.02531, 2015. [75] Li Z Z, Hoiem D. Learning without forgetting. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2018, 40(12): 2935−2947 doi: 10.1109/TPAMI.2017.2773081 [76] Castro F M, Marín-Jiménez M J, Guil N, Schmid C, Alahari K. End-to-end incremental learning. In: Proceedings of the 15th European Conference on Computer Vision. Munich, Germany: Springer, 2018. 241–257 [77] Zhang J T, Zhang J, Ghosh S, Li D W, Tasci S, Heck L. Class-incremental learning via deep model consolidation. In: Proceedings of the IEEE Winter Conference on Applications of Computer Vision (WACV). Snowmass, USA: IEEE, 2020. 1120–1129 [78] Douillard A, Cord M, Ollion C, Robert T, Valle E. PODNet: Pooled outputs distillation for small-tasks incremental learning. In: Proceedings of the 16th European Conference on Computer Vision. Glasgow, UK: Springer, 2020. 86–102Douillard A, Cord M, Ollion C, Robert T, Valle E. PODNet: Pooled outputs distillation for small-tasks incremental learning. In: Proceedings of the 16th European Conference on Computer Vision. Glasgow, UK: Springer, 2020. 86–102 [79] 朱飞, 张煦尧, 刘成林. 类别增量学习研究进展和性能评价. 自动化学报, 2023, 49(3): 635−660Zhu Fei, Zhang Xu-Yao, Liu Cheng-Lin. Class incremental learning: A review and performance evaluation. Acta Automatica Sinica, 2023, 49(3): 635−660 [80] Tao X Y, Chang X Y, Hong X P, Wei X, Gong Y H. Topology-preserving class-incremental learning. In: Proceedings of the 16th European Conference on Computer Vision. Glasgow, UK: Springer, 2020. 254–270 [81] Rebuff S A, Kolesnikov A, Sperl G, Lampert C H. ICaRL: Incremental classifier and representation learning. In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR). Honolulu, USA: IEEE, 2017. 5533–5542 [82] Goodfellow I, Pouget-Abadie J, Mirza M, Xu B, Warde-Farley D, Ozair S, et al. Generative adversarial networks. Communications of the ACM, 2020, 63(11): 139−144 doi: 10.1145/3422622 [83] Pellegrini L, Graffieti G, Lomonaco V, Maltoni D. Latent replay for real-time continual learning. In: Proceedings of the IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS). Las Vegas, USA: IEEE, 2020. 10203–10209 [84] Wu Y, Chen Y P, Wang L J, Ye Y C, Liu Z C, Guo Y D. Large scale incremental learning. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Long Beach, USA: IEEE, 2019. 374–382 [85] Serra J, Suris D, Miron M, Karatzoglou A. Overcoming catastrophic forgetting with hard attention to the task. In: Proceedings of the 35th International Conference on Machine Learning. Stockholm, Sweden: PMLR, 2018. 4548–4557 [86] Zhu K, Zhai W, Cao Y, Luo J B, Zha Z J. Self-sustaining representation expansion for non-exemplar class-incremental learning. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). New Orleans, USA: IEEE, 2022. 9286–9295 [87] Schrauwen B, Verstraeten D, van Campenhout J. An overview of reservoir computing: Theory, applications and implementations. In: Proceedings of the 15th European Symposium on Artificial Neural Networks. Bruges, Belgium: ESANN, 2007. 471–482 [88] Hauser H, Ijspeert A J, Füchslin R M, Pfeifer R, Maass W. The role of feedback in morphological computation with compliant bodies. Biological Cybernetics, 2012, 106(10): 595−613 doi: 10.1007/s00422-012-0516-4 [89] Caluwaerts K, Despraz J, Işçen A, Sabelhaus A P, Bruce J, Schrauwen B, et al. Design and control of compliant tensegrity robots through simulation and hardware validation. Journal of the Royal Society Interface, 2014, 11(98): Article No. 20140520 doi: 10.1098/rsif.2014.0520 [90] Degrave J, Caluwaerts K, Dambre J, Wyffels F. Developing an embodied gait on a compliant quadrupedal robot. In: Proceedings of the IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS). Hamburg, Germany: IEEE, 2015. 4486–4491 [91] Rückert E A, Neumann G. Stochastic optimal control methods for investigating the power of morphological computation. Artificial Life, 2013, 19(1): 115−131 doi: 10.1162/ARTL_a_00085 [92] Pervan A, Murphey T D. Algorithmic design for embodied intelligence in synthetic cells. IEEE Transactions on Automation Science and Engineering, 2021, 18(3): 864−875 doi: 10.1109/TASE.2020.3042492 [93] Zhu Y K, Mottaghi R, Kolve E, Lim J J, Gupta A, Li F F. Target-driven visual navigation in indoor scenes using deep reinforcement learning. In: Proceedings of the IEEE International Conference on Robotics and Automation (ICRA). Singapore: IEEE, 2017. 3357–3364Zhu Y K, Mottaghi R, Kolve E, Lim J J, Gupta A, Li F F. Target-driven visual navigation in indoor scenes using deep reinforcement learning. In: Proceedings of the IEEE International Conference on Robotics and Automation (ICRA). Singapore: IEEE, 2017. 3357–3364 [94] Chen T, Murali A, Gupta A. Hardware conditioned policies for multi-robot transfer learning. In: Proceedings of the 32nd International Conference on Neural Information Processing Systems. Montréal, Canada: Curran Associates Inc., 2018. 9355–9366 [95] Wang T W, Liao R J, Ba J, Fidler S. NerveNet: Learning structured policy with graph neural networks. In: Proceedings of the 6th International Conference on Learning Representations. Vancouver, Canada: OpenReview.net, 2018. [96] Blake C, Kurin V, Igl M, Whiteson S. Snowflake: Scaling GNNs to high-dimensional continuous control via parameter freezing. arXiv preprint arXiv: 2103.01009, 2021.Blake C, Kurin V, Igl M, Whiteson S. Snowflake: Scaling GNNs to high-dimensional continuous control via parameter freezing. arXiv preprint arXiv: 2103.01009, 2021. [97] Pathak D, Lu C, Darrell T, Isola P, Efros A A. Learning to control self-assembling morphologies: A study of generalization via modularity. In: Proceedings of the 33rd International Conference on Neural Information Processing Systems. Vancouver, Canada: Curran Associates Inc., 2019. Article No. 206 [98] Luck K S, Amor H B, Calandra R. Data-efficient co-adaptation of morphology and behaviour with deep reinforcement learning. In: Proceedings of the Conference on Robot Learning. Osaka, Japan: PMLR, 2020. 854–869 [99] Ha D. Reinforcement learning for improving agent design. Artificial Life, 2019, 25(4): 352−365 doi: 10.1162/artl_a_00301 [100] Nguyen T T, Nguyen N D, Vamplew P, Nahavandi S, Dazeley R, Lim C P. A multi-objective deep reinforcement learning framework. Engineering Applications of Artificial Intelligence, 2020, 96: Article No. 103915 doi: 10.1016/j.engappai.2020.103915 [101] Clouse J A. Learning from an automated training agent [Online], available: https://citeseerx.ist.psu.edu/document?repid=rep1&type=pdf&doi=a46bc2e3a41aa419895ecb766900d7ba878aac1e, November 9, 2024 [102] Price B, Boutilier C. Accelerating reinforcement learning through implicit imitation. Journal of Artificial Intelligence Research, 2003, 19: 569−629 doi: 10.1613/jair.898 [103] van Hasselt H. Double Q-learning. In: Proceedings of the 23rd International Conference on Neural Information Processing Systems. Vancouver, Canada: Curran Associates Inc., 2010. 2613–2621 [104] Sorokin I, Seleznev A, Pavlov M, Fedorov A, Ignateva A. Deep attention recurrent Q-network. arXiv preprint arXiv: 1512.01693, 2015.Sorokin I, Seleznev A, Pavlov M, Fedorov A, Ignateva A. Deep attention recurrent Q-network. arXiv preprint arXiv: 1512.01693, 2015. [105] Bowling M, Veloso M. Multiagent learning using a variable learning rate. Artificial Intelligence, 2002, 136(2): 215−250 doi: 10.1016/S0004-3702(02)00121-2 [106] Lauer M, Riedmiller M A. An algorithm for distributed reinforcement learning in cooperative multi-agent systems. In: Proceedings of the 17th International Conference on Machine Learning. Stanford, USA: Morgan Kaufmann, 2000. 535–542 [107] Leibo J Z, Zambaldi V, Lanctot M, Marecki J, Graepel T. Multi-agent reinforcement learning in sequential social dilemmas. In: Proceedings of the 16th Conference on Autonomous Agents and MultiAgent Systems. São Paulo, Brazil: International Foundation for Autonomous Agents and Multiagent Systems, 2017. 464–473 [108] Macenski S, Foote T, Gerkey B, Lalancette C, Woodall W. Robot operating system 2: Design, architecture, and uses in the wild. Science Robotics, 2022, 7(66): Article No. eabm6074 doi: 10.1126/scirobotics.abm6074 [109] Huang W L, Wang C, Zhang R H, Li Y Z, Wu J J, Li F F. VoxPoser: Composable 3D value maps for robotic manipulation with language models. In: Proceedings of the 7th Conference on Robot Learning. Atlanta, USA: PMLR, 2023. 540–562 [110] Li C S, Zhang R H, Wong J, Gokmen C, Srivastava S, Martín-Martín R, et al. BEHAVIOR-1K: A human-centered, embodied AI benchmark with 1000 everyday activities and realistic simulation. arXiv preprint arXiv: 2403.09227, 2024.Li C S, Zhang R H, Wong J, Gokmen C, Srivastava S, Martín-Martín R, et al. BEHAVIOR-1K: A human-centered, embodied AI benchmark with 1000 everyday activities and realistic simulation. arXiv preprint arXiv: 2403.09227, 2024. [111] Li B, Zhang Y M, Zhang T T, Acarman T, Ouyang Y K, Li L, et al. Embodied footprints: A safety-guaranteed collision-avoidance model for numerical optimization-based trajectory planning. IEEE Transactions on Intelligent Transportation Systems, 2024, 25(2): 2046−2060 doi: 10.1109/TITS.2023.3316175 [112] Ma C L, Provost J. Design-to-test approach for programmable controllers in safety-critical automation systems. IEEE Transactions on Industrial Informatics, 2020, 16(10): 6499−6508 doi: 10.1109/TII.2020.2968480 [113] Cai L, Zhou C L, Wang Y Q, Wang H, Liu B Y. Binocular vision-based pole-shaped obstacle detection and ranging study. Applied Sciences, 2023, 13(23): Article No. 12617 doi: 10.3390/app132312617 [114] Hawke J, E H, Badrinarayanan V, Kendall A. Reimagining an autonomous vehicle. arXiv preprint arXiv: 2108.05805, 2021.Hawke J, E H, Badrinarayanan V, Kendall A. Reimagining an autonomous vehicle. arXiv preprint arXiv: 2108.05805, 2021. [115] Sun P, Kretzschmar H, Dotiwalla X, Chouard A, Patnaik V, Tsui P. Scalability in perception for autonomous driving: Waymo open dataset. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Seattle, USA: IEEE, 2020. 2443–2451 [116] Chen L, Wu P H, Chitta K, Jaeger B, Geiger A, Li H Y. End-to-end autonomous driving: Challenges and frontiers. arXiv preprint arXiv: 2306.16927, 2024.Chen L, Wu P H, Chitta K, Jaeger B, Geiger A, Li H Y. End-to-end autonomous driving: Challenges and frontiers. arXiv preprint arXiv: 2306.16927, 2024. [117] Fu Z P, Zhao T Z, Finn C. Mobile ALOHA: Learning bimanual mobile manipulation with low-cost whole-body teleoperation. arXiv preprint arXiv: 2401.02117, 2024.Fu Z P, Zhao T Z, Finn C. Mobile ALOHA: Learning bimanual mobile manipulation with low-cost whole-body teleoperation. arXiv preprint arXiv: 2401.02117, 2024. [118] Fager P, Calzavara M, Sgarbossa F. Modelling time efficiency of cobot-supported kit preparation. The International Journal of Advanced Manufacturing Technology, 2020, 106(5): 2227−2241 [119] 兰沣卜, 赵文博, 朱凯, 张涛. 基于具身智能的移动操作机器人系统发展研究. 中国工程科学, 2024, 26(1): 139−148 doi: 10.15302/J-SSCAE-2024.01.010Lan Feng-Bo, Zhao Wen-Bo, Zhu Kai, Zhang Tao. Development of mobile manipulator robot systems with embodied intelligence. Strategic Study of CAE, 2024, 26(1): 139−148 doi: 10.15302/J-SSCAE-2024.01.010 [120] Zhao H R, Pan F X, Ping H Q Y, Zhou Y M. Agent as cerebrum, controller as cerebellum: Implementing an embodied LMM-based agent on drones. arXiv preprint arXiv: 2311.15033, 2023.Zhao H R, Pan F X, Ping H Q Y, Zhou Y M. Agent as cerebrum, controller as cerebellum: Implementing an embodied LMM-based agent on drones. arXiv preprint arXiv: 2311.15033, 2023. [121] Wang J, Wu Z X, Zhang Y, Kong S H, Tan M, Yu J Z. Integrated tracking control of an underwater bionic robot based on multimodal motions. IEEE Transactions on Systems, Man, and Cybernetics: Systems, 2024, 54(3): 1599−1610 doi: 10.1109/TSMC.2023.3328010 [122] Egan D, Cosker D, McDonnell R. NeuroDog: Quadruped embodiment using neural networks. Proceedings of the ACM on Computer Graphics and Interactive Techniques, 2023, 6(3): Article No. 38 [123] Tong Y C, Liu H T, Zhang Z T. Advancements in humanoid robots: A comprehensive review and future prospects. IEEE/CAA Journal of Automatica Sinica, 2024, 11(2): 301−328 doi: 10.1109/JAS.2023.124140 [124] Haarnoja T, Moran B, Lever G, Huang S H, Tirumala D, Humplik J, et al. Learning agile soccer skills for a bipedal robot with deep reinforcement learning. Science Robotics, 2024, 9(89): Article No. eadi8022 doi: 10.1126/scirobotics.adi8022 [125] Dong H W, Liu Y, Chu T, Saddik A E. Bringing robots home: The rise of AI robots in consumer electronics. arXiv preprint arXiv: 2403.14449, 2024.Dong H W, Liu Y, Chu T, Saddik A E. Bringing robots home: The rise of AI robots in consumer electronics. arXiv preprint arXiv: 2403.14449, 2024. [126] Haldar A I, Pagar N D. Predictive control of zero moment point (ZMP) for terrain robot kinematics. Materials Today: Proceedings, 2023, 80: 122−127 doi: 10.1016/j.matpr.2022.10.286 [127] Qiao H, Chen J H, Huang X. A survey of brain-inspired intelligent robots: Integration of vision, decision, motion control, and musculoskeletal systems. IEEE Transactions on Cybernetics, 2022, 52(10): 11267−11280 doi: 10.1109/TCYB.2021.3071312 [128] Qiao H, Wu Y X, Zhong S L, Yin P J, Chen J H. Brain-inspired intelligent robotics: Theoretical analysis and systematic application. Machine Intelligence Research, 2023, 20(1): 1−18 doi: 10.1007/s11633-022-1390-8 [129] Sun Y L, Zong C J, Pancheri F, Chen T, Lueth T C. Design of topology optimized compliant legs for bio-inspired quadruped robots. Scientific Reports, 2023, 13(1): Article No. 4875 doi: 10.1038/s41598-023-32106-5 [130] Taheri H, Mozayani N. A study on quadruped mobile robots. Mechanism and Machine Theory, 2023, 190: Article No. 105448 doi: 10.1016/j.mechmachtheory.2023.105448 [131] Idée A, Mosca M, Pin D. Skin barrier reinforcement effect assessment of a spot-on based on natural ingredients in a dog model of tape stripping. Veterinary Sciences, 2022, 9(8): Article No. 390 doi: 10.3390/vetsci9080390 [132] Yang C Y, Yuan K, Zhu Q G, Yu W M, Li Z B. Multi-expert learning of adaptive legged locomotion. Science Robotics, 2020, 5(49): Article No. eabb2174 doi: 10.1126/scirobotics.abb2174 [133] Dai B L, Khorrambakht R, Krishnamurthy P, Khorrami F. Sailing through point clouds: Safe navigation using point cloud based control barrier functions. IEEE Robotics and Automation Letters, 2024, 9(9): 7731−7738 doi: 10.1109/LRA.2024.3431870 [134] Huang K, Yang B Y, Gao W. Modality plug-and-play: Elastic modality adaptation in multimodal LLMs for embodied AI. arXiv preprint arXiv: 2312.07886, 2023.Huang K, Yang B Y, Gao W. Modality plug-and-play: Elastic modality adaptation in multimodal LLMs for embodied AI. arXiv preprint arXiv: 2312.07886, 2023. [135] Mutlu R, Alici G, Li W H. Three-dimensional kinematic modeling of helix-forming lamina-emergent soft smart actuators based on electroactive polymers. IEEE Transactions on Systems, Man, and Cybernetics: Systems, 2017, 47(9): 2562−2573 [136] Goncalves A, Kuppuswamy N, Beaulieu A, Uttamchandani A, Tsui K M, Alspach A. Punyo-1: Soft tactile-sensing upper-body robot for large object manipulation and physical human interaction. In: Proceedings of the 5th International Conference on Soft Robotics (RoboSoft). Edinburgh, UK: IEEE, 2022. 844–851 [137] Becker K, Teeple C, Charles N, Jung Y, Baum D, Weaver J C, et al. Active entanglement enables stochastic, topological grasping. Proceedings of the National Academy of Sciences of the United States of America, 2022, 119(42): Article No. e2209819119 [138] Xie Z X, Yuan F Y, Liu J Q, Tian L F, Chen B H, Fu Z Q, et al. Octopus-inspired sensorized soft arm for environmental interaction. Science Robotics, 2023, 8(84): Article No. eadh7852 doi: 10.1126/scirobotics.adh7852 [139] Mengaldo G, Renda F, Brunton S L, Bächer M, Calisti M, Duriez C, et al. A concise guide to modelling the physics of embodied intelligence in soft robotics. Nature Reviews Physics, 2022, 4(9): 595−610 doi: 10.1038/s42254-022-00481-z [140] 白天翔, 王帅, 沈震, 曹东璞, 郑南宁, 王飞跃. 平行机器人与平行无人系统: 框架、结构、过程、平台及其应用. 自动化学报, 2017, 43(2): 161−175Bai Tian-Xiang, Wang Shuai, Shen Zhen, Cao Dong-Pu, Zheng Nan-Ning, Wang Fei-Yue. Parallel robotics and parallel unmanned systems: Framework, structure, process, platform and applications. Acta Automatica Sinica, 2017, 43(2): 161−175 [141] 王飞跃. 机器人的未来发展: 从工业自动化到知识自动化. 科技导报, 2015, 33(21): 39−44Wang Fei-Yue. On future development of robotics: From industrial automation to knowledge automation. Science & Technology Review, 2015, 33(21): 39−44 [142] 王飞跃. 软件定义的系统与知识自动化: 从牛顿到默顿的平行升华. 自动化学报, 2015, 41(1): 1−8Wang Fei-Yue. Software-defined systems and knowledge automation: A parallel paradigm shift from Newton to Merton. Acta Automatica Sinica, 2015, 41(1): 1−8 [143] Driess D, Xia F, Sajjadi M S M, Lynch C, Chowdhery A, Ichter B, et al. PaLM-E: An embodied multimodal language model. In: Proceedings of the 40th International Conference on Machine Learning. Honolulu, USA: PMLR, 2023. 8469–8488 [144] Shridhar M, Thomason J, Gordon D, Bisk Y, Han W, Mottaghi R. ALFRED: A benchmark for interpreting grounded instructions for everyday tasks. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Seattle, USA: IEEE, 2020. 10737–10746 [145] Weihs L, Deitke M, Kembhavi A, Mottaghi R. Visual room rearrangement. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Nashville, USA: IEEE, 2021. 5918–5927 [146] Batra D, Chang A X, Chernova S, Davison A J, Deng J, Koltun V, et al. Rearrangement: A challenge for embodied AI. arXiv preprint arXiv: 2011.01975, 2020.Batra D, Chang A X, Chernova S, Davison A J, Deng J, Koltun V, et al. Rearrangement: A challenge for embodied AI. arXiv preprint arXiv: 2011.01975, 2020. [147] Shen B K, Xia F, Li C S, Martín-Martín R, Fan L X, Wang G Z. iGibson 1.0: A simulation environment for interactive tasks in large realistic scenes. In: Proceedings of the IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS). Prague, Czech Republic: IEEE, 2021. 7520–7527 [148] Li C S, Xia F, Martín-Martín R, Lingelbach M, Srivastava S, Shen B K, et al. iGibson 2.0: Object-centric simulation for robot learning of everyday household tasks. In: Proceedings of the 5th Conference on Robot Learning. London, UK: PMLR, 2021. 455–465 [149] Savva M, Kadian A, Maksymets O, Zhao Y L, Wijmans E, Jain B. Habitat: A platform for embodied AI research. In: Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV). Seoul, South Korea: IEEE, 2019. 9338–9346 [150] Ramakrishnan S K, Gokaslan A, Wijmans E, Maksymets O, Clegg A, Turner J, et al. Habitat-Matterport 3D dataset (HM3D): 1000 large-scale 3D environments for embodied AI. arXiv preprint arXiv: 2109.08238, 2021.Ramakrishnan S K, Gokaslan A, Wijmans E, Maksymets O, Clegg A, Turner J, et al. Habitat-Matterport 3D dataset (HM3D): 1000 large-scale 3D environments for embodied AI. arXiv preprint arXiv: 2109.08238, 2021. [151] Wani S, Patel S, Jain U, Chang A X, Savva M. Multi ON: Benchmarking semantic map memory using multi-object navigation. In: Proceedings of the 34th International Conference on Neural Information Processing Systems. Vancouver, Canada: Curran Associates Inc., 2020. Article No. 813 [152] Srivastava S, Li C S, Lingelbach M, Martín-Martín R, Xia F, Vainio K E, et al. BEHAVIOR: Benchmark for everyday household activities in virtual, interactive, and ecological environments. In: Proceedings of the 5th Conference on Robot Learning. London, UK: PMLR, 2022. 477–490 [153] Kolve E, Mottaghi R, Han W, VanderBilt E, Weihs L, Herrasti A, et al. AI2-THOR: An interactive 3D environment for visual AI. arXiv preprint arXiv: 1712.05474, 2022.Kolve E, Mottaghi R, Han W, VanderBilt E, Weihs L, Herrasti A, et al. AI2-THOR: An interactive 3D environment for visual AI. arXiv preprint arXiv: 1712.05474, 2022. [154] Gan C, Schwartz J, Alter S, Mrowca D, Schrimpf M, Traer J, et al. ThreeDWorld: A platform for interactive multi-modal physical simulation. arXiv preprint arXiv: 2007.04954, 2021.Gan C, Schwartz J, Alter S, Mrowca D, Schrimpf M, Traer J, et al. ThreeDWorld: A platform for interactive multi-modal physical simulation. arXiv preprint arXiv: 2007.04954, 2021. [155] Gan C, Zhou S Y, Schwartz J, Alter S, Bhandwaldar A, Gutfreund D. The ThreeDWorld transport challenge: A visually guided task-and-motion planning benchmark towards physically realistic embodied AI. In: Proceedings of the International Conference on Robotics and Automation (ICRA). Philadelphia, USA: IEEE, 2022. 8847–8854 [156] Szot A, Clegg A, Undersander E, Wijmans E, Zhao Y L, Turner J, et al. Habitat 2.0: Training home assistants to rearrange their habitat. arXiv preprint arXiv: 2106.14405, 2022.Szot A, Clegg A, Undersander E, Wijmans E, Zhao Y L, Turner J, et al. Habitat 2.0: Training home assistants to rearrange their habitat. arXiv preprint arXiv: 2106.14405, 2022. [157] Bi K F, Xie L X, Zhang H H, Chen X, Gu X T, Tian Q. Accurate medium-range global weather forecasting with 3D neural networks. Nature, 2023, 619(7970): 533−538 doi: 10.1038/s41586-023-06185-3 [158] Wang F Y. The emergence of intelligent enterprises: From CPS to CPSS. IEEE Intelligent Systems, 2010, 25(4): 85−88 doi: 10.1109/MIS.2010.104 [159] Zhang J J, Wang F Y, Wang X, Xiong G, Zhu F H, Lv Y S, et al. Cyber-physical-social systems: The state of the art and perspectives. IEEE Transactions on Computational Social Systems, 2018, 5(3): 829−840 doi: 10.1109/TCSS.2018.2861224 [160] Wang F Y, Wang X, Li L X, Li L. Steps toward parallel intelligence. IEEE/CAA Journal of Automatica Sinica, 2016, 3(4): 345−348 doi: 10.1109/JAS.2016.7510067 [161] Wang F Y. Parallel intelligence in metaverses: Welcome to Hanoi! IEEE Intelligent Systems, 2022, 37(1): 16−20 doi: 10.1109/MIS.2022.3154541 [162] Wang X, Yang J, Han J P, Wang W, Wang F Y. Metaverses and DeMetaverses: From digital twins in CPS to parallel intelligence in CPSS. IEEE Intelligent Systems, 2022, 37(4): 97−102 doi: 10.1109/MIS.2022.3196592 [163] Wang F Y. Parallel control and management for intelligent transportation systems: Concepts, architectures, and applications. IEEE Transactions on Intelligent Transportation Systems, 2010, 11(3): 630−638 doi: 10.1109/TITS.2010.2060218 [164] Wang Z R, Lv C, Wang F Y. A new era of intelligent vehicles and intelligent transportation systems: Digital twins and parallel intelligence. IEEE Transactions on Intelligent Vehicles, 2023, 8(4): 2619−2627 doi: 10.1109/TIV.2023.3264812 [165] Yang J, Wang X X, Zhao Y D. Parallel manufacturing for industrial metaverses: A new paradigm in smart manufacturing. IEEE/CAA Journal of Automatica Sinica, 2022, 9(12): 2063−2070 doi: 10.1109/JAS.2022.106097 [166] Liu Y H, Sun B Y, Tian Y L, Wang X X, Zhu Y, Huai R X, et al. Software-defined active lidars for autonomous driving: A parallel intelligence-based adaptive model. IEEE Transactions on Intelligent Vehicles, 2023, 8(8): 4047−4056 doi: 10.1109/TIV.2023.3289540 [167] Wang S, Wang J, Wang X, Qiu T Y, Yuan Y, Ouyang L W, et al. Blockchain-powered parallel healthcare systems based on the ACP approach. IEEE Transactions on Computational Social Systems, 2018, 5(4): 942−950 doi: 10.1109/TCSS.2018.2865526 [168] Wang X J, Kang M Z, Sun H Q, de Reffye P, Wang F Y. DeCASA in AgriVerse: Parallel agriculture for smart villages in metaverses. IEEE/CAA Journal of Automatica Sinica, 2022, 9(12): 2055−2062 doi: 10.1109/JAS.2022.106103 [169] Wang X X, Li J J, Fan L L, Wang Y T, Li Y K. Advancing vehicular healthcare: The DAO-based parallel maintenance for intelligent vehicles. IEEE Transactions on Intelligent Vehicles, 2023, 8(12): 4671−4673 doi: 10.1109/TIV.2023.3341855 [170] Wang X X, Yang J, Wang Y T, Miao Q H, Wang F Y, Zhao A J, et al. Steps toward Industry 5.0: Building “6S” parallel industries with cyber-physical-social intelligence. IEEE/CAA Journal of Automatica Sinica, 2023, 10(8): 1692−1703 doi: 10.1109/JAS.2023.123753 [171] 王飞跃. 关于复杂系统研究的计算理论与方法. 中国基础科学, 2004, 6(5): 3−10 doi: 10.3969/j.issn.1009-2412.2004.05.001Wang Fei-Yue. Computational theory and method on complex system. China Basic Science, 2004, 6(5): 3−10 doi: 10.3969/j.issn.1009-2412.2004.05.001 [172] Zhao Y, Zhu Z Q, Chen B, Qiu S H, Huang J C, Lu X, et al. Toward parallel intelligence: An interdisciplinary solution for complex systems. The Innovation, 2023, 4(6): Article No. 100521 doi: 10.1016/j.xinn.2023.100521 [173] Li X, Ye P J, Li J J, Liu Z M, Cao L B, Wang F Y. From features engineering to scenarios engineering for trustworthy AI: I&I, C&C, and V&V. IEEE Intelligent Systems, 2022, 37(4): 18−26 doi: 10.1109/MIS.2022.3197950 [174] Li X, Tian Y L, Ye P J, Duan H B, Wang F Y. A novel scenarios engineering methodology for foundation models in metaverse. IEEE Transactions on Systems, Man, and Cybernetics: Systems, 2023, 53(4): 2148−2159 doi: 10.1109/TSMC.2022.3228594 [175] 杨静, 王晓, 王雨桐, 刘忠民, 李小双, 王飞跃. 平行智能与CPSS: 三十年发展的回顾与展望. 自动化学报, 2023, 49(3): 614−634Yang Jing, Wang Xiao, Wang Yu-Tong, Liu Zhong-Min, Li Xiao-Shuang, Wang Fei-Yue. Parallel intelligence and CPSS in 30 years: An ACP approach. Acta Automatica Sinica, 2023, 49(3): 614−634 期刊类型引用(1)
1. 金立民，王海超，谷江春，徐以涛，丁国如. 低空具身智能频谱管控技术研究. 数据采集与处理. 2025(01): 45-55 . 
百度学术其他类型引用(0)
-
下载:
计量
- 文章访问数: 2043
- HTML全文浏览量: 1449
- PDF下载量: 891
- 被引次数: 1
