工業(yè)機(jī)器人執(zhí)行系統(tǒng)設(shè)計(jì)含4張CAD圖,工業(yè),機(jī)器人,執(zhí)行,履行,系統(tǒng),設(shè)計(jì),cad Abstract According to the application requirements of the mobile robot，the general human friendly task planning and execution system ( HFPES) model and the architecture is studied and established in dynamic intelligent environment，including the establishment of the MAS model for distributed robot system． By adding centralized structure to the distributd system and introducing composite agent and mobile agent concept, the multi-agent structure of distributed robot system with high reliability is es- tablished. The hierarchical mixed-layer structure of tehe robot agent modeling is proposed according to the robot agent in the system，including task planning layer，task allocation layer，IAAs manage- ment layer，autonomous planning layer and component management layer．The method of multi-a- gent negotiation is employed to accomplish task planning and assigning which effectively resolves the problems brought by the environment dynamic unpredictability． Through the task allocation and im- plementation methods，the intelligent robot system can run autonomously and effectively in compo- nent environment,w hich highly enhances the system's stability and reliability． Keywords: mobile robot task planning execution system multi-Agent Introduction In the future, mobile robot will become a convenient and friendly assistant for human beings. First of all, it needs to be able to adapt to the harmonious coexistence with human beings in the highly dynamic application environment, which plays a key role in the robot being able to carry out tasks according to the needs of tasks in the known application environment model At the same time, in order to effectively reduce the cost of robot and improve the ability of robot to perform tasks, the combination of intelligent environment or intelligent space and robot has become an inevitable trendThe research of robot planning mainly lies in that the robot is in all kinds of noisy environment models, and all kinds of perceptual information are uncertain to a certain extent, so the robot needs to integrate perception and execution to carry out direct planning. The research of Hoffman and breazel shows that the robot recognizes and infers human behavior and machines Based on this, human beings can plan and execute their own tasks, which can effectively improve the ability of human-computer cooperation and interaction. However, intelligent environment can monitor, identify and speculate the human behaviors of service objects in its space through its complex joint perception and execution ability. Hatp (human aware task Planner) and Galindo (human-computer joint task Planner) proposed by monteuil et al In the man-machine cooperation planning method proposed by Mausam, the behavior habit of the service object is not considered, but the behavior planning of the service object is simply predicted and provided to the robot planner. The co MDPs (concurrent mark ov processes) method proposed by Mausam is a fully observable MDPs (Markov decision processes) At the same time, it proposes how to calculate a series of planning behaviors that can be safely executed at the same time and have a certain duration. However, this method only aims at a single agent of robot and the execution time of random behaviors greatly increases the calculation burden, which is not suitable for robot in-line planning tasks In view of the above applications, this paper studies in the traditional deterministic task planning and execution of mobile robots, considering the dynamic environment such as the existence of service objects and human beings, and making full use of intelligent service components, through robot task planning and execution and online re planning execution, to achieve optimized service functions. 1. Distributed mobile robot system architecture In order to reflect the initiative of robot, this system adopts a basic architecture of robot centered service component-based mobile robot system, as shown in Figure 1 The system consists of two parts: mobile robot and environment intelligent node. They are in the same environment, and they are connected with each other through network. The network here can be wireless or wired, including LAN and Internet Therefore, the environment here includes the local environment and the Internet environment, and the environmental intelligent node also includes the local intelligent node and the internet intelligent node. The environmental intelligent node is actually each auxiliary resource device, which is distributed in the space, can be embedded in the environment fixedly, and can also have certain mobility, For example, mobile car, mobile manipulator and so on, they all have independent IP, providing access in the form of network nodes. In the whole architecture, the robot as the core of the system is responsible for completing various tasks, while the environmental intelligent nodes are cooperating with each other to provide various kinds of assistance to help them complete the task. Figure 1 Robot centered distributed intelligent system Due to the distribution and autonomy of distributed robot system, and the dynamic characteristics of the observation environment, this paper uses a-gent to map each intelligent component, and constructs the whole distributed component-based robot system into a MAS system, which is divided into three different levels of agents according to the level of component Intelligence: A1, A2 and IAAs (interaction agents) Among them, A1 represents the mobile robot with the highest level of intelligence. Its main function is to plan and allocate the overall tasks, and to control and coordinate other intelligent components in the environment. A2 represents other intelligent component nodes in the environment, and completes the task allocation process through task modeling, matching and filtering The obtained sub task is decomposed into a series of specific robot executable action sequences. IAAs represents the service object human agent with behavior interaction with robot. Through interaction and negotiation with A1, the problem of avoiding human-computer conflict / blocking and unfriendliness is solved. Based on the fully distributed structure, the advantages of the centralized structure are added, and the composite agent is introduced With the concept of mobile agent, a multi-agent structure of a highly reliable distributed robot system is established. The specific structure is shown in Figure 2 Figure 2 multi agent structure of robot system Different from the fully distributed structure, the centralized structure and the distributed structure are combined. At the same time, the composite agent and the mobile agent technology are introduced. The robot agent and the intelligent component agent are redefined. Their functions are divided into two parts: a'1 and a'2, in which a'1 Its function is overall task planning, which can monitor the whole system as a whole. The specific function is to analyze the task according to the schedule and return status of the service object agent, determine the task plan to be executed by the system, and complete the task allocation and planning on the basis of the automatic task discovery and allocation mechanism In general, the planning function of the whole system is completed by robot a '1 and intelligent component a' 2. At the same time, a '1 has mobility. When the function of robot human node fails, it can be controlled by a' 1 on intelligent component node Instead, this mobile agent technology increases the reliability and stability of the whole planning system. The main function of a'2 is to decompose and execute tasks. Normally, only a'2 on the intelligent component node works, and only when its function fails will it transfer to a'2 on the robot node In this centralized and distributed system architecture, there is only one node responsible for the overall task planning and execution control at any time, so it can ensure the global optimization and controllability of the system In addition, an automatic task discovery and assignment mechanism based on semantic function modeling and matching is proposed. On this basis, the automatic task planning and execution strategy of robot system is realized. The task assignment process is completed by modeling, matching and filtering among agents, which effectively solves the problems caused by the unknown environment state. Meanwhile, IAAs is introduced The agent proposes a series of rules for human-computer harmony, which can effectively realize the planning and execution of human-friendly robot autonomous tasks, improve the adaptability of the system to the service object and meet the needs of human-computer harmony 2. Basic hierarchical structure of task planning and execution system in component environment The task planning and control of the whole distributed component-based robot system is divided into five layers: task planning layer, task allocation layer, IAAs management layer, autonomous planning layer and component management layer, as shown in Figure 3 The task plan layer analyzes the task according to the IAAs schedule and the status returned by the IAAs management layer to determine the task plan to be activated by the system Tasks are composed of multiple decomposed subtasks with different execution priorities. Therefore, they need to be sorted and combined according to the priority order of subtasks, so as to determine the execution order finally. At the same time, monitor and coordinate the tasks of the whole system, suspend or withdraw the unfinished tasks due to the dynamic environment, and then optimize the tasks according to the priority order Plan again in advance and execution time Figure 3 hierarchical structure of distributed component-based robot system  The task allocation layer mainly completes the task allocation process to intelligent components and robots, and meets the established optimization objectives. The semantic web technology is introduced, and an automatic task discovery and allocation mechanism based on semantic function modeling and matching is proposed. On this basis, the automatic task planning and execution strategy of robot system is realized Through modeling, matching and filtering, the task assignment process is completed IAAs management is mainly responsible for the construction and update management of service object agents with behavioral interaction with robots. In order to plan and execute human-friendly autonomous tasks of robots, IAAs management draws lessons from some customary principles in interpersonal interaction, such as not infringing other's personal space, the minimum scope of human-computer interaction, maintaining visibility to others, and abiding by behaviors Driving, predicting and giving courtesy to others, quantifying these principles, building constraints into the robot task planner The autonomous planning layer mainly runs on robot nodes and each intelligent component node, and decomposes the obtained sub tasks into a series of specific robot executable action sequences, including the establishment of autonomous planning model and the development of efficient planning and execution algorithm. This part of function runs autonomously on each intelligent node without cooperation with other nodes The component management layer manages and monitors all kinds of intelligent components of the auxiliary robot in a centralized way, so as to make rational and efficient use of the component resources in the intelligent environment. There are various component resources in the distributed component-based robot system, such as intelligent visual nodes, mobile cars, etc., which must be monitored and managed globally Under the unified management of component management layer, each intelligent component matches its functional model with task demand model, selects the tasks that meet the requirements and sorts them according to the functional similarity, so as to form a list of candidate services that can be directly called. When abnormal events occur, such as node failure, component management layer can send fault signals through the network and send fault signals to the robot or Other components send out resource request information, so as to reallocate tasks and plan execution functions. 3.Task planning method The multi-agent negotiation (task planning and allocation) model is shown in Figure 4. Robot agent is the organizer and manager of task planning in distributed robot system, the only interface with IAAs, which is responsible for obtaining new tasks from IAAs on a regular basis and assigning and executing tasks in time. Other service component agents are in robot Under the control of robot agent, the task assignment process is completed by modeling, matching and filtering, and the sub tasks are decomposed, and the task execution results are fed back to robot a-gent. Finally, the robot agent is fed back to IAAs, which effectively solves the problem of environmental dynamic unknowns By this way of task assignment and execution, the autonomous operation of robot system can be realized to the greatest extent, and the stability and reliability of the system can be enhanced The query request of robot agent to service component agent includes task demand model description. Service component agent first uses its own matching algorithm to judge whether to give feedback, but only completes part of matching work (for example, only pays attention to the matching of entry parameters, that is, first ensures that the entry parameters provided by the robot can support the task operation), and the other part is handed over to the robot to complete（ For example, the matching of export parameters ensures that the task can provide the required information). AFigure 4 task planning and allocation model Conclusion In this paper, the basic framework of robot task planning and execution system under the intelligent service component environment is established Through a series of rules of human-computer harmony, the task allocation has certain adaptability to the service object. This architecture disperses machine intelligence into intelligent components, and effectively solves the bottleneck problem of too much computation in the planning problem. 摘 要 針對(duì)移動(dòng)機(jī)器人的應(yīng)用需求，研究并建立與人共存智能環(huán)境中對(duì)人友好的通用任務(wù)規(guī)劃與執(zhí)行系統(tǒng)(HFPES)的模型與架構(gòu)，包括建立分布式機(jī)器人系統(tǒng)的MAS模型。在完全分布式結(jié)構(gòu)的基礎(chǔ)上，加入集中式結(jié)構(gòu)的優(yōu)點(diǎn)，引入復(fù)合Agent和移動(dòng)Agent概念，建立一種高可靠性的分布式機(jī)器人系統(tǒng)的多Agent結(jié)構(gòu)。同時(shí)針對(duì)系統(tǒng)中機(jī)器人智能體給出其Agent建模的分層混合結(jié)構(gòu)，包括任務(wù)計(jì)劃層、任務(wù)分配層、IAAs管理層、自主規(guī)劃層和構(gòu)件管理層，并采用多Agent協(xié)商的方法進(jìn)行任務(wù)規(guī)劃分配，有效解決了環(huán)境動(dòng)態(tài)未知性帶來(lái)的問(wèn)題。通過(guò)這種任務(wù)分配、執(zhí)行方式可以最大程度上實(shí)現(xiàn)智能構(gòu)件環(huán)境下機(jī)器人系統(tǒng)的自主運(yùn)行，增強(qiáng)系統(tǒng)的 穩(wěn)定性和可靠性。 關(guān)鍵詞： 動(dòng)機(jī)器人 任務(wù)規(guī)劃 執(zhí)行系統(tǒng) 多Agent 引 言 未來(lái)移動(dòng)機(jī)器人真正成為人類便捷友好的助手，首先需要能夠適應(yīng)在與人共存、高度動(dòng)態(tài)的應(yīng)用環(huán)境中與人和諧共處，這對(duì)于機(jī)器人能夠在已知應(yīng)用環(huán)境模型中根據(jù)任務(wù)需求進(jìn)行任務(wù)起著關(guān)鍵的作用。同時(shí)，為了有效降低機(jī)器人成本、提高機(jī)器人執(zhí)行任務(wù)的能力，智能環(huán)境或智能空間與機(jī)器人相結(jié)合已成為必然趨勢(shì)。 機(jī)器人規(guī)劃研究主要在于機(jī)器人處于有噪音的各類環(huán)境模型中，各種感知信息都在一定程度上具有不確定性，這樣機(jī)器人就需要將感知和執(zhí)行進(jìn)行整合來(lái)進(jìn)行直接規(guī)劃Hoffman和Breazeal的研究表明機(jī)器人對(duì)人的行為識(shí)別和推理以及機(jī)器人據(jù)此進(jìn)行自身任務(wù)規(guī)劃及執(zhí)行能夠有效提高人機(jī)合作與交互能力。而智能環(huán)境恰好能夠通過(guò)其復(fù)雜聯(lián)合感知和執(zhí)行能力對(duì)其空間中的服務(wù)對(duì)象人行為進(jìn)行監(jiān)控識(shí)別和推測(cè)，在Montreuil等提出的人機(jī)聯(lián)合任務(wù)規(guī)劃器HATP（human aware task planner），以及Galindo等提出的人機(jī)協(xié)作規(guī)劃方法中，機(jī)器人并沒(méi)有考慮服務(wù)對(duì)象人的行為動(dòng)作習(xí)慣，而只是簡(jiǎn)單預(yù)測(cè)人的行為規(guī)劃提供給機(jī)器人規(guī)劃器Mausam提出的CoMDPs（concurrent markov processes）方法是完全可觀的MDPs( mark- ov decision processes)擴(kuò)展，同時(shí)提出如何計(jì)算得到一系列能夠被同時(shí)安全執(zhí)行且具有一定持續(xù)時(shí)間的規(guī)劃行為，但該方法僅針對(duì)機(jī)器人單一智能體且隨機(jī)的行為動(dòng)作執(zhí)行時(shí)間大大增加了計(jì)算負(fù)擔(dān)，并不適合機(jī)器人在線規(guī)劃任務(wù)。針對(duì)以上應(yīng)用，本文研究在移動(dòng)機(jī)器人傳統(tǒng)的確定型任務(wù)規(guī)劃和執(zhí)行中，考慮服務(wù)對(duì)象人類存在等動(dòng)態(tài)環(huán)境，并充分利用智能服務(wù)構(gòu)件，通過(guò)機(jī)器人任務(wù)規(guī)劃和執(zhí)行與在線再規(guī)劃執(zhí)行，達(dá)到優(yōu)化服務(wù)功能。 1.分布式移動(dòng)機(jī)器人系統(tǒng)架構(gòu) 為體現(xiàn)機(jī)器人的主動(dòng)性，本系統(tǒng)采用一種以機(jī)器人為中心的服務(wù)構(gòu)件化的移動(dòng)機(jī)器人系統(tǒng)基本架構(gòu)，如圖1所示。 該系統(tǒng)由移動(dòng)機(jī)器人和環(huán)境智能節(jié)點(diǎn)兩部分組成，它們同處在一個(gè)環(huán)境當(dāng)中，相互之間通過(guò)網(wǎng)絡(luò)相連，這里的網(wǎng)絡(luò)可以是無(wú)線的，也可以是有線的，既包括局域網(wǎng)，也包括互聯(lián)網(wǎng)。因此，這里的環(huán)境包括本地環(huán)境和互聯(lián)網(wǎng)環(huán)境，環(huán)境智能節(jié)點(diǎn)也包括本地智能節(jié)點(diǎn)和互聯(lián)網(wǎng)智能節(jié)點(diǎn)，環(huán)境智能節(jié)點(diǎn)實(shí)際上就是各個(gè)輔助資源設(shè)備，它們分布在空間中，可以固定地嵌入環(huán)境，也可以具備一定移動(dòng)性，例如移動(dòng)小車、活動(dòng)機(jī)械臂等等，它們都有獨(dú)立的IP，以網(wǎng)絡(luò)節(jié)點(diǎn)的形式提供訪問(wèn)，在整個(gè)架構(gòu)中，機(jī)器人作為系統(tǒng)的核心，負(fù)責(zé)完成各種任務(wù)，而環(huán)境智能節(jié)點(diǎn)則是互相協(xié)作起來(lái)為其提供各種各樣的輔助，協(xié)助其完成任務(wù)。 圖 1 以機(jī)器人為中心的分布式智能系統(tǒng) 由于分布式機(jī)器人系統(tǒng)的分布性和自主性的特點(diǎn)，以及所處觀測(cè)環(huán)境的動(dòng)態(tài)特性，本文采用 A-gent 映射各個(gè)智能構(gòu)件，將整個(gè)分布式構(gòu)件化機(jī)器人系統(tǒng)構(gòu)建成一個(gè) MAS 系統(tǒng)，按照構(gòu)件智能水 平的高低將其劃分成3個(gè)不同層次的Agent：A1，A2和IAAs( interaction agents)，其中 A1 表示智能水平最高的移動(dòng)機(jī)器人，其主要作用是總體任務(wù)規(guī) 劃和分配，同時(shí)能夠?qū)Νh(huán)境中的其他智能構(gòu)件進(jìn)行控制和協(xié)調(diào)． A2 表示環(huán)境中其他的智能構(gòu)件節(jié)點(diǎn)，通過(guò)任務(wù)建模、匹配、篩選完成任務(wù)分配過(guò)程，并將獲得的子任務(wù)分解成一系列具體的機(jī)器人可執(zhí)行的動(dòng)作序列。IAAs 表示和機(jī)器人有行為交互的服務(wù)對(duì)象人智能體，通過(guò)與A1的交互和協(xié)商解決避免人機(jī)沖突/阻塞和不友好的問(wèn)題。本文提出在完全分布式結(jié)構(gòu)的基礎(chǔ)上，加入集中式結(jié)構(gòu)的優(yōu)點(diǎn)，引入復(fù)合Agent和移動(dòng)Agent概念，建立一種高可靠性的分布式機(jī)器人系統(tǒng)的多Agent結(jié)構(gòu)，具體結(jié)構(gòu)如圖 2 所示。 圖 2 機(jī)器人系統(tǒng)多 Agent 結(jié)構(gòu) 與完全分布式結(jié)構(gòu)不同的是，將集中式結(jié)構(gòu)和分布式結(jié)構(gòu)相結(jié)合，同時(shí)引入復(fù)合Agent和移動(dòng)Agent技術(shù)，對(duì)機(jī)器人 Agent 和智能構(gòu)件 Agent 進(jìn)行了重新定義，將其功能分成兩部分：A'1和A'2 ，其中A'1的功能是總體任務(wù)全局規(guī)劃，能夠?qū)φ麄€(gè)系統(tǒng)進(jìn)行整體集中監(jiān)控，具體功能主要是根據(jù)服務(wù)對(duì)象人智能體的日程安排和其返回的狀態(tài)進(jìn)行任務(wù)分析，確定系統(tǒng)將要執(zhí)行的任務(wù)計(jì)劃，并在自動(dòng)任務(wù)發(fā)現(xiàn)分配機(jī)制的基礎(chǔ)上完成任務(wù)分配和規(guī)劃，同時(shí)對(duì)智能節(jié)點(diǎn)進(jìn)行監(jiān)督控制。正常情況下只有機(jī)器人節(jié)點(diǎn)上的A'1起作用，其他智能構(gòu)件上的A'1不發(fā)揮作用。通常整個(gè)系統(tǒng)的規(guī)劃功能通過(guò)機(jī)器人A' 1與智能構(gòu)件A'2 共同完成。同時(shí)A'1具有移動(dòng)性，在機(jī)器人節(jié)點(diǎn)功能失效時(shí)可由智能構(gòu)件節(jié)點(diǎn)上的A' 1代替，這種移動(dòng)智能體技術(shù)增加了整個(gè)規(guī)劃系統(tǒng)的可靠性和穩(wěn)定性。A'2 的主要功能是任務(wù)的分解執(zhí)行，正常情況下只有智能構(gòu)件節(jié)點(diǎn)上的A'2起作用，只有當(dāng)其功能失效時(shí)才會(huì)轉(zhuǎn)移到機(jī)器人節(jié)點(diǎn)的A'2上。在這種集中式和分布式相結(jié)合的系統(tǒng)架構(gòu)下任何時(shí)刻都有且只有一個(gè)節(jié)點(diǎn)負(fù)責(zé)整體任務(wù)規(guī)劃與執(zhí)行控制，因此能夠保證系統(tǒng)的全局最優(yōu)性和可控性。另外，提出基于語(yǔ)義功能建模與匹配的自動(dòng)任務(wù)發(fā)現(xiàn)分配機(jī)制，并在此基礎(chǔ)上實(shí)現(xiàn)了機(jī)器人系統(tǒng)自動(dòng)任務(wù)規(guī)劃與執(zhí)行策略，各 Agent 間通過(guò)建模、匹配、篩選完成任務(wù)分配過(guò)程，有效解決了環(huán)境動(dòng)態(tài)未知性帶來(lái)的問(wèn)題。同時(shí)引入 IAAs 智能體，提出一系列人機(jī)和諧相處的規(guī)則，有效實(shí)現(xiàn)對(duì)人友好的機(jī)器人自主任務(wù)規(guī)劃與執(zhí)行，提高了系統(tǒng)對(duì)服務(wù)對(duì)象人的適應(yīng)性，滿足人機(jī)和諧相處的需求。 2.構(gòu)件化環(huán)境下任務(wù)規(guī)劃與執(zhí)行系統(tǒng)基本層次結(jié)構(gòu) 將整個(gè)分布式構(gòu)件化機(jī)器人系統(tǒng)的任務(wù)規(guī)劃與控制分為任務(wù)計(jì)劃層、任務(wù)分配層、IAAs 管理層、自主規(guī)劃層和構(gòu)件管理層 5 個(gè)層次，如圖 3 所示。 其中任務(wù)計(jì)劃層根據(jù)IAAs的日程安排和IAAs管理層返回的狀態(tài)進(jìn)行任務(wù)分析，確定系統(tǒng)將要激活的任務(wù)計(jì)劃。任務(wù)由多個(gè)分解的子任務(wù)構(gòu)成，子任務(wù)的執(zhí)行優(yōu)先級(jí)各不相同，因此還需要按照各個(gè)子任務(wù)的優(yōu)先級(jí)順序?qū)ζ溥M(jìn)行排序、組合等，從而最終確定其執(zhí)行順序，同時(shí)對(duì)整個(gè)系統(tǒng)任務(wù)進(jìn)行監(jiān)控與協(xié)調(diào)，對(duì)由于動(dòng)態(tài)環(huán)境而沒(méi)有完成的任務(wù)進(jìn)行掛起或收回，再根據(jù)其任務(wù)優(yōu)先級(jí)和執(zhí)行時(shí)間重新進(jìn)行規(guī)劃。 圖 3 分布式構(gòu)件化機(jī)器人系統(tǒng)層次結(jié)構(gòu)圖 任務(wù)分配層主要完成任務(wù)對(duì)智能構(gòu)件和機(jī)器人的分配過(guò)程，同時(shí)滿足既定的優(yōu)化目標(biāo)。引入語(yǔ)義網(wǎng)技術(shù)，提出基于語(yǔ)義功能建模與匹配的自動(dòng)任務(wù)發(fā)現(xiàn)分配機(jī)制，并在此基礎(chǔ)上實(shí)現(xiàn)了機(jī)器人系統(tǒng)自動(dòng)任務(wù)規(guī)劃與執(zhí)行策略，各 Agent 間通過(guò)建模、匹配、篩選完成任務(wù)分配過(guò)程，有效解決了環(huán)境動(dòng)態(tài)未知性帶來(lái)的問(wèn)題。 IAAs 管理層主要負(fù)責(zé)對(duì)和機(jī)器人有行為交互的服務(wù)對(duì)象智能體進(jìn)行構(gòu)建和更新管理，為了進(jìn)行對(duì)人友好的機(jī)器人自主任務(wù)規(guī)劃與執(zhí)行，借鑒人際相處中的若干習(xí)慣原則，如不侵犯他人的個(gè)人空間、人機(jī)交互最小范圍、對(duì)他人保持可視性、遵守行駛靠向、預(yù)測(cè)并禮讓他人等，對(duì)此類原則進(jìn)行量化，構(gòu)建約束條件融入機(jī)器人任務(wù)規(guī)劃器中。 自主規(guī)劃層主要在機(jī)器人節(jié)點(diǎn)和各個(gè)智能構(gòu)件節(jié)點(diǎn)上運(yùn)行，將獲得的子任務(wù)分解成一系列具體的機(jī)器人可執(zhí)行的動(dòng)作序列，包括建立自主規(guī)劃模型和開(kāi)發(fā)高效的規(guī)劃執(zhí)行算法，這部分功能在各個(gè)智能節(jié)點(diǎn)上自主運(yùn)行，而不需與其他節(jié)點(diǎn)協(xié)作執(zhí)行。 構(gòu)件管理層對(duì)輔助機(jī)器人的各種智能構(gòu)件進(jìn)行集中管理和監(jiān)控，從而能夠合理高效地利用智能環(huán)境中的構(gòu)件資源。分布式構(gòu)件化機(jī)器人系統(tǒng)中的構(gòu)件資源多種多樣，如智能視覺(jué)節(jié)點(diǎn)、移動(dòng)小車等，必須對(duì)這些資源進(jìn)行全局監(jiān)控和管理。在構(gòu)件管理層的統(tǒng)一管理下，各個(gè)智能構(gòu)件將其功能模型與任務(wù)需求模型進(jìn)行匹配，篩選出符合要求的任務(wù)并按功能相似度排序，從而形成可直接調(diào)用的候選服務(wù)列表，當(dāng)異常事件發(fā)生時(shí)，如節(jié)點(diǎn)出現(xiàn)故障，構(gòu)件管理層即可通過(guò)網(wǎng)絡(luò)發(fā)出故障信號(hào)并再次向機(jī)器人或其他構(gòu)件發(fā)出資源請(qǐng)求信息，從而重新進(jìn)行任務(wù)分配和規(guī)劃執(zhí)行功能。 3.任務(wù)規(guī)劃方法 多 Agent 協(xié)商(任務(wù)規(guī)劃分配)模型如圖 4 所示。機(jī)器人 Agent 是分布式機(jī)器人系統(tǒng)任務(wù)規(guī)劃的組織者和管理者，是與服務(wù)對(duì)象智能體 IAAs 的唯一接口，負(fù)責(zé)定期從IAAs 獲取新任務(wù)，并將任務(wù)及時(shí)分配和執(zhí)行。其他服務(wù)構(gòu)件 Agent 均在機(jī)器人的控制下通過(guò)建模、匹配、篩選完成任務(wù)分配過(guò)程，并分解執(zhí)行得到的子任務(wù)，并向機(jī)器人Agent 進(jìn)行任務(wù)執(zhí)行結(jié)果反饋操作，從而最終通過(guò)機(jī)器人 Agent 反饋給 IAAs，有效解決了環(huán)境動(dòng)態(tài)未知性帶來(lái)的問(wèn)題。通過(guò)這種任務(wù)分配、執(zhí)行方式可以最大程度上實(shí)現(xiàn)智能構(gòu)件環(huán)境下機(jī)器人系統(tǒng)的自主運(yùn)行，增強(qiáng)系統(tǒng)的穩(wěn)定性和可靠性。機(jī)器人智能體對(duì)服務(wù)構(gòu)件智能體的查詢請(qǐng)求中包含任務(wù)需求模型描述，服務(wù)構(gòu)件智能體先采用自己的匹配算法判斷是否反饋，但只完成一部分匹配工作(例如只關(guān)注入口參數(shù)的匹配，即先保證機(jī)器人提供的入口參數(shù)能夠支持任務(wù)運(yùn)行)，另一部分則交給機(jī)器人完成(例如出口參數(shù)的匹配，保證任務(wù)能夠提供所需信息)。這樣，最終得到的是“雙方都認(rèn)可的服務(wù)”。 圖 4 任務(wù)規(guī)劃分配模型 圖 4 任務(wù)規(guī)劃分配模型 結(jié) 語(yǔ) 本文建立了智能服務(wù)構(gòu)件環(huán)境下機(jī)器人任務(wù)規(guī)劃與執(zhí)行系統(tǒng)基本框架。通過(guò)機(jī)器人Agent智能構(gòu)件Agent與交互人Agent之間不斷的協(xié)作過(guò)程完成任務(wù)的規(guī)劃分配與執(zhí)行。通過(guò)一系列人機(jī)和諧相處的規(guī)則，使任務(wù)分配對(duì)服務(wù)對(duì)象人具有一定適應(yīng)性。該架構(gòu)將機(jī)器智能分散到智能構(gòu)件中，有效解決規(guī)劃問(wèn)題中計(jì)算量過(guò)大的瓶頸問(wèn)題。 16