Simulation planning execution method and device of intelligent agent, equipment and medium

Abstract

The embodiment of the invention provides a simulation planning execution method of an intelligent agent, which can be applied to the technical field of artificial intelligence. The intelligent agent simulation planning execution method comprises the steps of calling a preset skill library and a preset action library to perform a first simulation planning process in a second planning execution process based on scene understanding information and corresponding execution abnormity feedback information in a first planning execution process, and generating a fuzzy decision path; according to task execution feedback information of the fuzzy decision path, calling a preset skill library and a preset action library to perform a second simulation planning process in a third planning execution process, and generating a target decision path; and through the target decision path, executing a target skill and a target action consistent with the corresponding code of the fuzzy decision path in the first simulation planning process in a third planning execution process, and completing simulation planning execution. The embodiment of the invention further provides a simulation planning execution device and equipment of the intelligent agent, a storage medium and a program product.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of artificial intelligence, in particular to the technical field of image processing, and more particularly to an agent simulation planning execution method, device, equipment, medium and product. BACKGROUND

[0002] Artificial intelligence (AI) is an important driving force for a new round of scientific and technological revolution and industrial change, and is a new key technology science that studies, develops and produces a new intelligent machine (i.e. an agent) that can react in a similar way to human intelligence.

[0003] The agent can react in a similar way to human intelligence mainly by executing various action tasks, so that the agent can be more consistent with human real actions in terms of speaking or body movement execution. However, in the planning stage of the agent, it is difficult to express simulation for different planning tasks, but the existing agent usually cannot effectively support fuzzy simulation, such as for different tasks of the same action, the same simulation knowledge cannot be compatible, which often leads to abnormal execution of the agent when executing the corresponding task. SUMMARY

[0004] In view of at least one of the problems of the prior art that the simulation is complex and cannot be modeled, the execution is abnormal due to the agent being unable to support fuzzy simulation, and the like, embodiments of the present application aim to support fuzzy simulation to realize a simulation planning execution method, device, equipment, medium and product of a simulation execution unified agent, so as to provide a skill library and action library knowledge that can be used for simulation and execution (i.e. simulation execution unified), so that the agent does not need to model the scene to provide accurate simulation model, thereby improving the task simulation planning execution efficiency, realizing the effect of more consistent execution reaction with human real actions, and achieving a higher intelligent level.

[0005] One aspect of the embodiments of the present application provides a method for simulating planning and execution of an agent, comprising: based on scenario understanding information and corresponding execution abnormal feedback information in a first planning and execution process, calling a preset skill library and a preset action library to perform a first simulation planning process in a second planning and execution process, and generating a fuzzy decision path; according to task execution feedback information of the fuzzy decision path, calling the preset skill library and the preset action library to perform a second simulation planning process in a third planning and execution process, and generating a target decision path; and through the target decision path, performing target skills and target actions consistent with the corresponding code of the fuzzy decision path in the first simulation planning process in the third planning and execution process, and completing the simulation planning and execution.

[0006] According to an embodiment of the present application, before the first simulation planning process in the second planning and execution process is performed based on the scenario understanding information and the corresponding execution abnormal feedback information in the first planning and execution process, the method further comprises: in the first planning and execution process, generating a first task intention based on the scenario understanding information; and according to the first task intention, calling the preset skill library and the preset action library to perform a third simulation planning process in the first planning and execution process, and generating a simulation abnormal decision path, which is used to generate the execution abnormal feedback information when executed.

[0007] According to an embodiment of the present application, in the first simulation planning process in the second planning and execution process, based on the scenario understanding information and the corresponding execution abnormal feedback information in the first planning and execution process, the method further comprises: in the second planning and execution process, generating a second task intention according to the first task intention corresponding to the scenario understanding information and the execution abnormal feedback information; and according to the second task intention, calling the preset skill library and the preset action library to perform the first simulation planning process in the second planning and execution process, and generating the fuzzy decision path.

[0008] According to an embodiment of the present application, in the first simulation planning process in the second planning and execution process, based on the scenario understanding information and the corresponding execution abnormal feedback information in the first planning and execution process, the method further comprises: in the second planning and execution process, generating a second task intention according to the first task intention corresponding to the scenario understanding information and the execution abnormal feedback information; and according to the second task intention, calling the preset skill library and the preset action library to perform the first simulation planning process in the second planning and execution process, and generating the fuzzy decision path.

[0009] According to an embodiment of the present application, before the target decision path is generated by calling the preset skill library and the preset action library to perform the second simulation planning process in the third planning execution process according to the task execution feedback information of the fuzzy decision path, the method further comprises: generating a third task intention of a fourth planning execution process according to the current execution state and the current scene understanding information during the task execution of the fuzzy decision path; and generating an intermediate decision path matched with the task execution feedback information by calling the preset skill library and the preset action library to perform a fourth simulation planning process in the fourth planning execution process according to the third task intention.

[0010] According to an embodiment of the present application, in the target decision path generated by calling the preset skill library and the preset action library to perform the second simulation planning process in the third planning execution process according to the task execution feedback information of the fuzzy decision path, the method further comprises: generating a fourth task intention according to the task execution feedback information; and generating the target decision path by calling the preset skill library and the preset action library to perform the second simulation planning process in the third planning execution process according to the fourth task intention.

[0011] According to an embodiment of the present application, in the target decision path generated by calling the preset skill library and the preset action library to perform the second simulation planning process in the third planning execution process according to the fourth task intention, the method further comprises: calling the preset skill library to perform a continuous simulation execution according to the fourth task intention; in response to the continuous simulation execution, calling the preset action library to perform an execution simulation judgment of an execution action to generate at least one execution simulation path of the second simulation planning process in the third planning execution process; and performing value evaluation on each execution simulation path in the at least one execution simulation path to generate the target decision path.

[0012] Another aspect of the embodiment of the present application provides an agent simulation planning execution device, which comprises a fuzzy path generation module, a target path generation module and a target path execution module. The fuzzy path generation module is configured to generate a fuzzy decision path by calling a preset skill library and a preset action library to perform a first simulation planning process in a second planning execution process based on scene understanding information and corresponding execution exception feedback information in a first planning execution process; the target path generation module is configured to generate a target decision path by calling the preset skill library and the preset action library to perform a second simulation planning process in a third planning execution process according to task execution feedback information of the fuzzy decision path; and the target path execution module is configured to execute target skills and target actions consistent with the fuzzy decision path in the first simulation planning process in the third planning execution process through the target decision path to complete the simulation planning execution.

[0013] Another aspect of the present invention provides an electronic device including one or more processors and a memory for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors cause the one or more processors to perform the above-described simulation planning execution method for an intelligent agent.

[0014] Another aspect of the present invention provides a computer-readable storage medium having executable instructions stored thereon, which, when executed by a processor, cause the processor to perform the above-described simulation planning execution method for an intelligent agent.

[0015] Another aspect of the present invention provides a computer program product, including a computer program that, when executed by a processor, implements the above-described simulation planning execution method for an intelligent agent.

[0016] The simulation planning and execution method for intelligent agents provided in this invention can at least partially solve the problems existing in the related art, such as the inability to model complex simulations and execution anomalies caused by the inability of intelligent agents to support fuzzy simulations. Therefore, it can achieve at least one of the following technical effects:

[0017] First, by leveraging the knowledge from action and skill libraries to achieve a unified design for simulation execution, the agent can seamlessly switch between simulation and actual operation. Furthermore, by providing a fuzzy programming approach, action modeling or skill execution effects can be achieved based on the scenario and task, thus avoiding the need to develop a unified model to model all possible situations and significantly improving planning and execution efficiency. Moreover, the ability to freely combine the skill and action libraries enhances the flexibility of knowledge representation, ensuring the timeliness, accuracy, and reliability of planning.

[0018] It should be understood that the above general description and the following specific embodiments are merely exemplary and illustrative, and do not limit the scope of the invention. Attached Figure Description

[0019] The above-described features, other objects, and advantages of the present invention will become clearer from the following description of embodiments of the invention with reference to the accompanying drawings, in which:

[0020] Figure 1 The illustration schematically depicts an application scenario of a simulated planning execution method, apparatus, device, medium, and program product for an intelligent agent according to embodiments of the present invention.

[0021] Figure 2 A flowchart illustrating a simulation planning execution method for an intelligent agent according to an embodiment of the present invention is shown schematically.

[0022] Figure 3This diagram illustrates an application scenario of the simulated planning and execution method of an intelligent agent according to an embodiment of the present invention.

[0023] Figure 4 This diagram illustrates another application scenario of the simulated planning execution method of the intelligent agent according to an embodiment of the present invention.

[0024] Figure 5 A schematic diagram illustrating the structure of a simulated planning execution device for an intelligent agent according to an embodiment of the present invention is shown; and

[0025] Figure 6 A block diagram of an electronic device suitable for implementing a simulated planning execution method for an intelligent agent, according to an embodiment of the present invention, is shown schematically.

[0026] The accompanying drawings mentioned above are part of the specification of embodiments of the present invention, illustrating exemplary embodiments of the invention. The drawings, together with the description in the specification, serve to illustrate the principles of the embodiments of the present invention. It should be understood that the above general description with reference to the drawings and the following detailed description are merely exemplary and illustrative, and do not limit the scope of the invention. Detailed Implementation

[0027] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the spirit of the contents disclosed in the present invention will be clearly explained below with reference to the accompanying drawings and detailed description. After understanding the embodiments of the present invention, any person skilled in the art can make changes and modifications based on the technology taught in the present invention without departing from the spirit and scope of the present invention.

[0028] The illustrative embodiments and descriptions of the present invention are used to explain the invention, but are not intended to limit the invention. Furthermore, elements / components using the same or similar reference numerals in the drawings and embodiments are used to represent the same or similar parts.

[0029] The terms "first," "second," etc., used in this invention do not specifically refer to any order or sequence, nor are they intended to limit the invention; they are merely used to distinguish elements or operations described using the same technical terms.

[0030] The directional terms used in this invention, such as up, down, left, right, front, or back, are merely for reference to the accompanying drawings. Therefore, the directional terms used are for illustrative purposes and not for limiting the scope of this invention.

[0031] The terms “comprising,” “including,” “having,” “containing,” etc., used in this invention are all open-ended terms, meaning that they include but are not limited to.

[0032] The term "and / or" as used in this invention includes any or all combinations of the things mentioned.

[0033] In this invention, "multiple" includes "two" and "more than two"; in this invention, "multiple groups" includes "two groups" and "more than two groups".

[0034] The terms "approximately," "about," etc., used in this invention are intended to modify any quantity or error that may vary slightly, but these slight variations or errors do not change the essence of the quantity or error. Generally, the range of slight variations or errors modified by such terms may be 20% in some embodiments, 10% in some embodiments, 5% in some embodiments, or other values. Those skilled in the art should understand that the aforementioned values ​​can be adjusted according to actual needs and are not limited thereto.

[0035] All terms used herein (including technical and scientific terms) have the meanings commonly understood by those skilled in the art, unless otherwise defined. It should be noted that the terms used herein are to be interpreted in a manner consistent with the context of this specification, and not in an idealized or overly rigid way.

[0036] When expressions such as "at least one of A, B, and C" are used, they should generally be interpreted in accordance with the meaning commonly understood by those skilled in the art (e.g., "a system having at least one of A, B, and C" should include, but is not limited to, systems having A alone, having B alone, having C alone, having A and B, having A and C, having B and C, and / or having A, B, and C, etc.). When expressions such as "at least one of A, B, or C" are used, they should generally be interpreted in accordance with the meaning commonly understood by those skilled in the art (e.g., "a system having at least one of A, B, or C" should include, but is not limited to, systems having A alone, having B alone, having C alone, having A and B, having A and C, having B and C, and / or having A, B, and C, etc.). Those skilled in the art should also understand that any conjunction and / or phrase that substantially arbitrarily indicates two or more optional items, whether in the specification, claims, or drawings, should be understood to indicate the possibility of including one of these items, either of these items, or both items. For example, the phrase “A or B” should be understood as including the possibility of “A” or “B”, or “A and B”.

[0037] Intelligent agents can make responses similar to human intelligence primarily through the execution of various action tasks, enabling them to increasingly mimic real human actions, whether in speech or physical movements. However, during the planning phase of an intelligent agent, it is difficult to express the simulations involved in different planning tasks. Existing intelligent agents typically cannot effectively support fuzzy simulations; for example, different tasks involving the same action may not be compatible with the same simulation knowledge. This often leads to execution anomalies when the intelligent agent performs the corresponding tasks.

[0038] Specifically, for tasks like opening a refrigerator to find food or looking in a mirror to check for stains, the contents seen after opening the refrigerator and the appearance seen in the mirror cannot be obtained during the simulation planning process. For example, both tasks require the action of "look at," but finding food in the refrigerator and looking in the mirror correspond to different application scenarios. These two different tasks cannot be compatible with the same simulation knowledge. For instance, the former task requires modeling the food that might be seen, while the latter requires modeling the stains that might be seen. Modeling in such scenarios cannot be achieved using a single "look at" action model, easily leading to simulation planning errors. Therefore, when modeling actions and skills for such tasks, it is necessary to consider supporting effective fuzzy simulation to ensure that the same simulation knowledge can be effectively used during execution.

[0039] In view of at least one of the problems existing in the prior art, such as the inability to model complex situations for precise simulations and the execution anomalies caused by the inability to support fuzzy simulation, the embodiments of the present invention aim to provide a simulation planning and execution method, device, equipment, medium and product for intelligent agents that can support fuzzy simulation to achieve unified simulation and execution. This provides a method that allows skill base and action base knowledge to be used for both simulation and execution (i.e., unified simulation and execution), so that intelligent agents no longer need to provide precise simulation models for specific scenarios, thereby improving the efficiency of task simulation planning and execution, achieving an effect that is more in line with the execution response of real human actions, and thus achieving a higher level of intelligence.

[0040] One aspect of the present invention provides a simulated planning execution method for an intelligent agent, comprising: based on scene understanding information and corresponding execution anomaly feedback information in a first planning execution process, invoking a preset skill library and a preset action library to perform a first simulated planning process in a second planning execution process, generating a fuzzy decision path; based on task execution feedback information of the fuzzy decision path, invoking a preset skill library and a preset action library to perform a second simulated planning process in a third planning execution process, generating a target decision path; and through the target decision path, executing a target skill and a target action consistent with the code corresponding to the fuzzy decision path in the first simulated planning process in the third planning execution process, thereby completing the simulated planning execution.

[0041] Figure 1 The illustration schematically depicts an application scenario of a simulation planning execution method, apparatus, device, medium, and program product for an intelligent agent according to embodiments of the present invention.

[0042] like Figure 1 As shown, application scenario 100 according to this embodiment may include terminal devices 101, 102, and 103, network 104, and server 105. Network 104 is used as a medium to provide a communication link between terminal devices 101, 102, and 103 and server 105. Network 104 may include various connection types, such as wired or wireless communication links or fiber optic cables, etc.

[0043] Users can use terminal devices 101, 102, and 103 to interact with server 105 via network 104 to receive or send messages, etc. Various communication client applications can be installed on terminal devices 101, 102, and 103, such as shopping applications, web browser applications, search applications, instant messaging tools, email clients, social media platform software, etc. (for example only).

[0044] Terminal devices 101, 102, and 103 can be various electronic devices with displays and web browsing capabilities, including but not limited to smartphones, tablets, laptops, and desktop computers.

[0045] Server 105 can be a server that provides various services, such as a backend management server that supports websites browsed by users using terminal devices 101, 102, and 103 (for example only). The backend management server can analyze and process data such as received user requests, and feed back the processing results (such as web pages, information, or data obtained or generated according to user requests) to the terminal devices.

[0046] It should be noted that the intelligent agent simulation planning execution method provided in this embodiment of the invention can generally be executed by server 105. Correspondingly, the intelligent agent simulation planning execution device provided in this embodiment of the invention can generally be located in server 105. The intelligent agent simulation planning execution method provided in this embodiment of the invention can also be executed by a server or server cluster that is different from server 105 and capable of communicating with terminal devices 101, 102, 103 and / or server 105. Correspondingly, the intelligent agent simulation planning execution device provided in this embodiment of the invention can also be located in a server or server cluster that is different from server 105 and capable of communicating with terminal devices 101, 102, 103 and / or server 105.

[0047] It should be understood that Figure 1 The number of terminal devices, networks, and servers shown is merely illustrative. Depending on implementation needs, any number of terminal devices, networks, and servers can be included.

[0048] The following will be based on Figure 1 The described scene, through Figures 2-4 The simulation planning and execution method of the intelligent agent according to the disclosed embodiments is described in detail.

[0049] like Figure 2 As shown, one aspect of an embodiment of the present invention provides a method for simulating planning and execution of an intelligent agent, comprising operations S201 to S203.

[0050] In operation S201, based on the scenario understanding information and corresponding execution exception feedback information in the first planning execution process, the preset skill library and preset action library are called to perform the first simulated planning process in the second planning execution process, and a fuzzy decision path is generated.

[0051] In operation S202, based on the task execution feedback information of the fuzzy decision path, the preset skill library and preset action library are invoked to perform the second simulated planning process in the third planning execution process, generating the target decision path; and

[0052] In operation S203, through the target decision path, the target skills and target actions that are consistent with the fuzzy decision path codes in the first simulation planning process are executed in the third planning execution process to complete the simulation planning execution.

[0053] In this embodiment of the invention, the intelligent agent can be the executing subject of the simulation planning and execution method described above, or it can be an executor controlled by the simulation planning and execution method. Specifically, it can be a humanoid intelligent robot or other AI device, which typically has its own actuators capable of performing specific action tasks. For example, a humanoid robot can use a robotic hand to perform the action task of picking up objects, or even perform action tasks such as packaging products on an industrial production line. These action tasks can serve as the target execution tasks of the intelligent agent. To complete these target execution tasks, the intelligent agent needs to sequentially complete a series of execution actions.

[0054] Typically, once an intelligent agent is powered on, it enters an interactive state with its environment (E). This involves acquiring environmental data through various sensors, such as "seeing" (e.g., visual sensors), "touching" (e.g., tactile sensors), and "hearing" (e.g., microphones). Specifically, the environment (E) can be an interactive environment in which the intelligent agent autonomously perceives and recognizes changes in its surroundings. For example, if it's cloudy outside and the indoor environment darkens, the intelligent agent can sense the dimming of its ambient light and actively perceive the change to understand that "the weather has turned cloudy, so the indoor light has dimmed." Furthermore, the environment (E) can also be a scenario where the intelligent agent interacts with a user based on its own state. For instance, if a user is moving items, the agent can assess the user's physical condition and conversation to determine if the user is unable to move the item or needs to make multiple attempts. The agent can then proactively help the user and send a personalized voice message like, "I'll help you move it; you can rest for a while."

[0055] like Figure 3 As shown, the intelligent agent in this embodiment of the invention can achieve task execution based on a multi-round loop mechanism of "environment perception - simulation planning - task execution". Specifically, in the environment perception stage S1, the intelligent agent acquires environmental data and forms scene understanding information for its environment E, and can generate an execution intention corresponding to the execution target based on this scene understanding information, i.e., Belief Update; further, in the simulation planning stage S2, the intelligent agent can perform simulation planning based on the execution intention to generate the corresponding target execution task, i.e., Plan; then, in the task execution stage S3, the intelligent agent performs the actual execution of the target execution task for the environment E, thereby completing the execution target of the current round of "environment perception - simulation planning - task execution", i.e., Execute. Typically, each task execution target can be achieved step by step through multiple rounds of "environment perception - simulation planning - task execution" in a continuous or intermittent manner (such as abortion or suspension).

[0056] The planning and execution process can be understood as a single round of "environmental perception - simulated planning - task execution" by an intelligent agent. Each planning and execution process corresponds to three stages: environmental perception stage S1, simulated planning stage S2, and task execution stage S3, as detailed below. Figure 3 The first planning execution process S301, the second planning execution process S302, the third planning execution process S303, and the fourth planning execution process S304 shown can each correspond to different "environmental perception-simulation planning-task execution" processes under the same execution objective. Therefore, scene understanding information can be the environmental data and cognitive information obtained by the agent during the planning execution process, which can be considered as the content observed by the agent and the "thinking" content about these observations.

[0057] Execution anomaly feedback information can be the content of anomalies that occur in the task execution phase S3 during the first planning execution process, such as the "empty action" execution anomaly caused by not receiving a useful planning task, along with information on the cause of the anomaly. This can help determine whether the execution anomaly feedback information is due to a situation in the simulation planning phase S2 where the simulation is too complex to be modeled.

[0058] The pre-defined skill base is the main knowledge system of the agent during the aforementioned planning and execution process. This knowledge system carries the agent's procedural knowledge representation, such as knowledge of finding objects and moving objects. For finding objects, the agent will first search for objects in the nearest possible location, and then search in other, unexplored locations further away. For moving objects, the agent will first pick up the object and then move it to the appropriate location to put it down. The procedural rules agreed upon for these action tasks can be understood as the corresponding skills (U). The pre-defined skill base can be implemented based on a reinforcement learning policy model, containing two main logical aspects: policy functions and skill knowledge, giving it a certain planning ability, such as planning to chase another agent (even a user or pet) by running. Therefore, the pre-defined skill base can simultaneously support the flexible planning and execution process of tasks, such as... Figure 3 As shown.

[0059] The preset action library is an action system designed to connect the simulation of an agent's mental world with its execution in the real world, enabling interaction with the real world. Each action instance uniquely corresponds to an action execution in the real world, which can be a long-step or short-step action, such as picking up an object, pressing a switch, moving to a destination, or looking at an object. Each action in the preset action library strictly corresponds to a specific action in environment E (the real world). When executing this action, the agent directly reproduces the action in the environment. The modeling of this action execution is also embedded in this action instance, allowing the model of the action instance to be retrieved for simulation planning during mental world simulation, and the action to be reproduced in environment E according to the action alignment protocol during real world execution. Therefore, the preset actions can connect the agent's simulation planning and task execution, ensuring the final simulation execution is unified, such as... Figure 3 As shown. It should be noted that each action in the preset action library can be indivisible, atomic-level knowledge.

[0060] In this embodiment of the invention, skills in the preset skill library can be freely combined, or combined with corresponding actions in the action library to generate new skills. Each skill in the preset skill library can call each other to form a target skill, or call action instances from the preset action library to form a target action, which can be specifically reflected in the code representation of the target skill and the target action.

[0061] In this embodiment of the invention, simulation (i.e., simulated execution) refers to the modeling of the operation and evolution of the perceived system or process by the intelligent agent. This is generally reflected in the agent's understanding of the environment E, and can be understood as the agent's pre-planned execution of each planned task. It is equivalent to imagining the execution process in the "brain" to obtain the imagined execution result. In this way, the agent can plan a better execution path for the task execution phase S3 to achieve actual execution.

[0062] The simulation planning process can be understood as the process of simulating and generating planning tasks (i.e., Plan) in the simulation planning phase S2 of each round of "environmental perception-simulation planning-task execution", specifically as follows: Figure 3 The diagram illustrates the first simulation planning process S321 in the second planning execution process S302, the second simulation planning process S322 in the third planning execution process S303, the third simulation planning process S323 in the first planning execution process S301, and the fourth simulation planning process S324 in the fourth planning execution process S304. Specifically, the simulation planning process may include a process of planning and evaluating execution paths based on simulation modeling and calling preset skill libraries and preset action libraries, such as... Figure 4Operation S401, as shown, may further include a process of generating decision paths by simulating and evaluating the execution of each path based on the execution path planning, such as... Figure 4 The operation shown is S402.

[0063] Based on the scenario understanding information and execution anomaly feedback information obtained during the first planning execution process, a first simulation planning process is performed for the simulation planning phase of the second planning execution process to plan fuzzy decision paths. The fuzzy decision path can be a planning task that implements fuzzy simulation based on the execution anomaly feedback information obtained during the first planning execution process. Specifically, it can be represented by the execution path of executable actions in the task execution phase of the second planning execution process. The target action and target skill decided by this fuzzy decision path can be expressed in code form. The task execution feedback information can be the task execution result formed by executing at least the fuzzy decision path in the task execution phase S3 of the second planning execution process.

[0064] The target decision path can be a planned task that can achieve the target task execution, planned for the second simulated planning process corresponding to the simulated planning stage S2 in the third planning execution process. Specifically, it can be represented by the execution path of actions that can be executed in the task execution stage of the third planning execution process. The target actions and target skills decided by this target decision path can also be expressed in code form. In this embodiment of the invention, the target actions and target skills of this target decision path are consistent with the codes of the target actions and target skills of the aforementioned fuzzy decision path.

[0065] like Figure 3 As shown, in the task execution stage S3 corresponding to the third planning execution process S303, the execution process of the target action and target skill can be executed for the above target decision path. Specifically, the corresponding execution action can be reproduced in environment E, so that the corresponding skill can be expressed and the corresponding execution result can be generated.

[0066] Therefore, the intelligent agent simulation planning and execution method of this invention can achieve targeted fuzzy simulation planning by combining the simulation logic of different simulation planning processes, which is difficult to achieve in cases where fuzzy simulation is difficult to implement. Based on the flexible planning design of the preset skill library and preset action library, it ensures that the target action and target skill of the target decision path are consistent with the code of the target action and target skill of the aforementioned fuzzy decision path, thereby achieving the effect of unified fuzzy simulation execution (i.e., simulation execution and real execution), which greatly improves the simulation execution efficiency and accuracy for fuzzy tasks and avoids the situation of feedback of execution abnormalities to the user again.

[0067] This allows the same knowledge to be used for both simulation and execution, achieving a unified practice for both simulation and execution. This eliminates the need for situations where execution is impossible or where two different sets of knowledge must be provided, as is often the case with traditional technologies. For example, the skill of placing an object requires picking up object A and placing it at location B. Clearly, simulation and execution share a unified understanding of this skill. However, agent execution occurs in the real world, not within its own mental space. Unified execution requires that the simulation of skill knowledge always be based on the current situation. For instance, if picking up an object other than A requires replanning, it shouldn't mechanically execute an outdated initial plan.

[0068] In summary, the simulation planning and execution method for intelligent agents in this invention allows the knowledge from preset skill libraries and preset action libraries to be used for both simulation and execution. Especially in situations where precise simulation is too complex to model, such as when an intelligent agent is looking in a mirror or searching for something and it's difficult to model what it can see, a fuzzy simulation solution can be provided. Therefore, users no longer need precise simulation models for such difficult-to-model scenarios, but only need to focus on the effect of its actions in the relevant scenarios. Those skilled in the art should understand that the simulation mentioned in this invention mainly refers to the modeling of actions and skill libraries, and includes the modeling of actions with long step lengths.

[0069] First, by leveraging the knowledge from action and skill libraries to achieve a unified design for simulation execution, the agent can seamlessly switch between simulation and actual operation. Furthermore, by providing a fuzzy programming approach, action modeling or skill execution effects can be achieved based on the scenario and task, thus avoiding the need to develop a unified model to model all possible situations and significantly improving planning and execution efficiency. Moreover, the ability to freely combine the skill and action libraries enhances the flexibility of knowledge representation, ensuring the timeliness, accuracy, and reliability of planning.

[0070] To enable those skilled in the art to have a clearer understanding of the above-described intelligent agent simulation planning and execution method of the embodiments of the present invention, further combined with Figures 3-4 The following is a more in-depth explanation of a specific implementation case of "intelligent body looking in a mirror to check if there is dirt on the face".

[0071] like Figure 3 and Figure 4 As shown, according to an embodiment of the present invention, before operation S201, based on the scene understanding information and corresponding execution anomaly feedback information in the first planning execution process, calls the preset skill library and preset action library to perform the first simulated planning process in the second planning execution process and generates the fuzzy decision path, the method further includes:

[0072] During the execution of the first plan, the first task intent is generated based on scenario understanding information;

[0073] Based on the first task intent, the preset skill library and preset action library are invoked to perform the third simulation planning process in the first planning execution process, generating a simulated abnormal decision path. The simulated abnormal decision path is used to generate execution abnormal feedback information when it is executed.

[0074] like Figure 3 As shown, in the environment perception phase S1 of the first planning execution process S301, the agent can perceive the environment E and learn about its interaction status with the environment E, such as whether it is having a conversation with a human or is currently performing an action. Simultaneously, it perceives all environmental data, such as voice messages from humans (e.g., "Do you have something on your face?"). Based on this environmental data, the agent can generate scene understanding information (i.e., belief pg), thereby generating a task intent, or first task intent, in the environment perception phase S1. For example, based on the relationship with the human who issued the voice message (e.g., the agent's user) combined with the voice message "Do you have something on your face?", the agent can generate the intent to "confirm whether there is something on your face and immediately and accurately reply to the user." Therefore, the first task intent corresponds to the task objective in this round of the first planning execution process S301.

[0075] In the simulation planning phase S2 of the first planning execution process S301, a task path for the aforementioned first task intent is planned and evaluated by calling a preset skill library and a preset action library, generating multiple skill- and action-based planning paths, and simulating the execution of each planning path. For example, during the execution of the simulated response, the agent needs to simulate "checking if there is dirt on the face," but because this check is not a deterministic action, it cannot be confirmed, resulting in no response being generated, i.e., a fuzzy task is created. At this time, based on the evaluation of the simulated execution of the above-mentioned planning paths, a final decision path of "no response can be generated because the check cannot be simulated" can be generated. Therefore, this simulated abnormal decision path can be the decision abnormal content that cannot be effectively simulated to generate an executable path in the simulation planning phase S2.

[0076] In the task execution phase S3 of the first planning execution process S301, the agent cannot effectively execute the simulated abnormal decision path and cannot reproduce the relevant skills or actions. At this time, it can only send empty actions to the environment E, that is, no executable path is decided, the execution is abnormal, and the execution abnormal feedback information is generated.

[0077] In summary, for the first planning and execution process described above, the task execution target can be determined based on scenario understanding information, and simulated execution anomalies for fuzzy tasks can be provided. This facilitates the intelligent agent to subsequently execute the simulated planning and execution method of this embodiment of the invention in a targeted manner, thereby improving the efficiency and accuracy of task planning and execution.

[0078] like Figure 3 and Figure 4 As shown, according to an embodiment of the present invention, in operation S201, based on the scene understanding information and corresponding execution anomaly feedback information in the first planning execution process, a preset skill library and a preset action library are invoked to perform the first simulated planning process in the second planning execution process, generating a fuzzy decision path, including:

[0079] During the execution of the second plan, the second task intent is generated based on the first task intent corresponding to the scenario understanding information and the execution exception feedback information.

[0080] Based on the second task intent, the preset skill library and preset action library are invoked to perform the first simulated planning process in the second planning execution process, generating a fuzzy decision path.

[0081] like Figure 3 As shown, in the environmental perception stage S1 of the second planning execution process S302, for the first task intent "confirm whether there is something on your face and immediately and accurately reply to the user" generated in the first planning execution process S301, and the execution anomaly feedback information that no executable path has been determined, a new second task intent "go look in the mirror to confirm whether there is something on your face" can be generated. Therefore, this second task intent is the updated intent for the agent to realize the execution goal of "confirm whether there is something on your face" in response to the above execution anomaly feedback information, based on the existing scene understanding information.

[0082] In the simulation planning phase S2 of the second planning execution process S302, the agent can further call upon the preset skill library and preset action library to plan and evaluate the task path for the second task intent, generating new planning paths for multiple skills and actions, and simulating the execution of each planning path. For example, in the process of simulating looking in the mirror, the agent needs to simulate "looking in the mirror to check if there is dirt on the face," but because the content that can be observed by "looking in the mirror" (there may be dirt on the face, or there may not be dirt) cannot be confirmed, that is, this action is not a deterministic action. Therefore, the agent can plan action formation paths of "finding a mirror," "walking to the mirror," and "looking in the mirror (observing the content reflected in the mirror)." Then, based on the possible results of the "looking in the mirror" simulation, the agent evaluates and selects the best path as the fuzzy decision path.

[0083] Therefore, this fuzzy decision path can be the optimal evaluated action path output by the agent based on its assessment and decision-making regarding the possible effects of fuzzy actions. This fuzzy decision path can have executable target skills and target actions, specifically represented by knowledge system codes from a pre-set skill library and a pre-set action library.

[0084] In the task execution phase S3 of the second planning execution process S302, the agent can effectively execute the aforementioned fuzzy decision paths, reproducing the relevant target actions and skills. This involves executing the fuzzy decision paths of "finding a mirror," "walking to the mirror," and "looking in the mirror (observing the reflection)," while simultaneously updating the execution states of these actions to the preset skill library. At this point, in the real environment E, the agent can at least execute the processes of "finding a mirror" and "walking to the mirror."

[0085] It is evident that, for the second planning execution process described above, the task intent and decision path can be updated based on the execution anomalies of the first planning execution process, thereby ensuring that the best-evaluated fuzzy decision path can be generated, and ensuring that the agent can at least first complete the determinable actions, thereby improving the efficiency of simulated planning execution.

[0086] like Figure 3 and Figure 4 As shown, according to an embodiment of the present invention, in the first simulation planning process of generating a fuzzy decision path during the second planning execution process of calling a preset skill library and a preset action library according to a second task intent, the process includes:

[0087] Based on the second task intent, a preset skill library is invoked to generate simulated execution skills;

[0088] By simulating the execution of skills and calling a preset action library to perform fuzzy simulation judgment of the execution actions, at least one fuzzy simulation path of the first simulation planning process in the second planning execution process is generated;

[0089] For each fuzzy simulation path in at least one fuzzy simulation path, a value assessment is performed to generate a fuzzy decision path.

[0090] like Figure 3 As shown, in the first simulation planning process S321, based on the second task intention of "looking in the mirror to check if there is anything on my face," a preset skill library is invoked to plan the corresponding skill knowledge to generate a simulated execution skill. This simulated execution skill is the skill knowledge constituted by combining and filtering skill knowledge from the preset skill library according to the second task intention, such as "simulating looking in the mirror to check if there is anything dirty on my face."

[0091] Fuzzy simulation judgment can be understood as the process of judging the expected execution result of action instances that cannot be effectively simulated due to the uncertainty of the simulation execution result. For example... Figure 4 As shown, in the simulation execution S401, the corresponding action instances from the preset action library are invoked through the simulation execution skill to perform fuzzy simulation judgment. Specifically, based on the aforementioned simulation execution skill of "simulating looking in the mirror to check if there is dirt on the face," action instances such as "finding a mirror," "walking to the mirror," and "looking in the mirror (observing the reflection)" from the preset action library can be invoked. While the action instances of "finding a mirror" and "walking to the mirror" can be simulated, the action instance of "looking in the mirror" (e.g., look at) is difficult to simulate the new information that can be obtained by looking in the mirror (e.g., whether there is dirt on the face). Therefore, at least two fuzzy simulation paths can be generated: (i) if looking in the mirror reveals dirt on the face, the expected effect of the simulated response action is "responding that there is dirt"; (ii) if looking in the mirror reveals no dirt on the face, the expected effect of the simulated response action is "responding that there is no dirt." Therefore, the fuzzy simulation path can be a set of actions that the agent re-plans for action instances that are difficult to simulate (but have the potential for fuzzy simulation).

[0092] For the aforementioned fuzzy simulation paths (I) and (II), a comprehensive value assessment can be conducted by combining the agent's current execution state with the scene understanding information corresponding to the latest environmental perception data. For example, when the agent is in a kitchen and is performing the action of cooking, the agent can assess, based on the scene understanding information obtained from environmental cognition, that the probability of "finding dirt on the face" in fuzzy simulation path (I) is 70%, and the probability of "finding no dirt on the face" in fuzzy simulation path (II) is 30%. If the agent is in a living room and is not currently in a task execution state where there is a possibility of dirt contamination, the agent can assess, based on the scene understanding information obtained from environmental cognition, that the probability of "finding dirt on the face" in fuzzy simulation path (I) is 30%, and the probability of "finding no dirt on the face" in fuzzy simulation path (II) is 70%.

[0093] Furthermore, regarding the value assessment of fuzzy simulation paths, the agent can also combine the historical interaction content of the person who sent the voice information "Do you have something on your face?" to understand the interaction relationship between the person who sent the voice information and the agent. For example, it can confirm whether the person who sent the voice information often jokes with the agent. If so, it can also combine the agent's other scene understanding information to assess that the probability of "finding something on your face" in fuzzy simulation path (I) is 30%, and the probability of "finding nothing on your face" in fuzzy simulation path (II) is 70%.

[0094] like Figure 4As shown, based on the above value assessment of fuzzy simulation paths, at least one fuzzy simulation path can be screened according to the magnitude of the assessment result, thereby selecting the fuzzy simulation path with the largest or smallest assessment result as the fuzzy decision path. For example, if the probability of fuzzy simulation path (I) "finding dirt on the face" is 30%, and the probability of fuzzy simulation path (II) "finding no dirt on the face" is 70%, then fuzzy simulation path (II) "finding no dirt on the face" can be selected as the fuzzy decision path.

[0095] Therefore, for fuzzy actions such as looking in a mirror, which are usually difficult to simulate, the final action planning and path output can be achieved, thereby avoiding task execution anomalies caused by task planning anomalies, thus improving the simulation planning efficiency of the agent and ensuring the efficiency and accuracy of task execution.

[0096] like Figure 3 and Figure 4 As shown, according to an embodiment of the present invention, before operation S202 calls the preset skill library and preset action library to perform the second simulation planning process in the third planning execution process based on the task execution feedback information of the fuzzy decision path, and generates the target decision path, the following is further included:

[0097] During the execution of a fuzzy decision-making path, the third task intent of the fourth planning execution process is generated based on the current execution status and the information understood in the current scenario.

[0098] Based on the third task intent, the preset skill library and preset action library are invoked to carry out the fourth simulation planning process in the fourth planning execution process, generating an intermediate decision path that matches the task execution feedback information.

[0099] like Figure 3 As shown, in the task execution phase S3 of the second planning execution process S302, actions can be executed on the fuzzy decision path, namely, "finding the mirror", "walking to the mirror", and "looking in the mirror (observing the reflection)". In the real environment E, the agent can at least execute the processes of "finding the mirror" and "walking to the mirror".

[0100] like Figure 3As shown, in the fourth planning execution process S304, the agent can simply execute the "find the mirror" and "walk to the mirror" processes of the aforementioned fuzzy decision path. At this time, no new intention is generated in its corresponding environmental perception stage S1. Therefore, in the fourth simulation planning process S324 corresponding to its simulation planning stage S2, no skill library simulation will be performed, and no new decision path will be generated. In its corresponding task execution stage S3, the agent will not execute a new decision path, but simply execute the fuzzy decision path. That is, the fuzzy decision path is always in the execution state. For example, the agent is always in the process of "walking to the mirror". However, other "unexpected" situations may occur during the execution of this fuzzy decision path. For example, there may be an obstacle on the "walking to the mirror" path (such as a chair placed arbitrarily by the user or a mop that may fall over), or the user may hear the latest voice instruction message from the user, "Please help me move my things first".

[0101] An intelligent agent can detect its environment and its own state in real time to confirm the progress of the current target task. Specifically, it can assess its spatial environment and its own state in each iteration. The current execution state can be at least one of the following: the agent's own action at the current moment, or its environmental perception data. The state information of its own action can include the name of the action being performed and the corresponding execution information (such as the torque, angle, estimated execution time, and spatial position of key joints). The environmental perception data can include information about people and objects in the agent's current space, such as the name, size, position (distance), and even shape and color of obstacles in the path, or dynamic changes in people in the surrounding environment, such as their speed and facial expressions. The current scene understanding information is the cognitive understanding information generated by the agent in response to its current execution state.

[0102] For unexpected situations that occur during the execution of the fuzzy decision path of "looking in the mirror (observing the content reflected in the mirror)," the agent can introduce the fourth planning execution process S304. In the corresponding environment perception stage S1, the agent's current execution state is confirmed and current scene understanding information is generated, such as "finding a chair placed randomly on the walking route" or "hearing a voice instruction from the user to move things." At this time, the agent can generate a new task intention based on the current scene understanding information, namely the third task intention, such as "before looking in the mirror (observing the content reflected in the mirror), the chair needs to be put back in place" or "before looking in the mirror (observing the content reflected in the mirror), the user needs to be helped to move things."

[0103] Therefore, in the simulation planning stage S2 of the fourth planning execution process S304, the agent can also perform task simulation planning again based on the aforementioned third task intent and generate a new decision path as an intermediate insertion task in the execution process of the fuzzy decision path, i.e., an intermediate decision path. This intermediate decision path can include the execution actions of the original fuzzy decision path, such as "straighten the chair - lift the chair - find the original position of the chair - walk to the original position - place the chair in the correct posture in the original position - find the mirror - walk to the mirror - look in the mirror (observe the reflection)". In the corresponding task execution stage S3, by executing this intermediate insertion task, the agent can support operations such as inserting new tasks, suspending or interrupting the original task according to the actual situation during the execution of the fuzzy decision path, and can resume the execution of the original task after the new task is completed.

[0104] In other words, the action instances of the aforementioned fuzzy decision-making path can actually be understood as being in an ongoing execution state, only with the insertion of new, higher-priority action instances. This allows the agent to complete one or more new tasks during the execution of one task based on the latest environmental perception and understanding, while also continuing the execution of the original fuzzy decision-making path. Specifically, a pre-set skill library allows the execution of action instances of the fuzzy decision-making path to be interrupted anywhere. The agent can perceive the latest environmental information and update its belief state, enabling the insertion and execution of new tasks, and can continue the simulated execution of unexecuted action instances of the original task. Therefore, it allows a corresponding task to transition to a state where the skill is being executed while performing an action, or to transition to a suspended state after completing an action. This supports performing other actions while performing an action, or suspending the task and resuming its execution upon receiving a specific signal.

[0105] Specifically, the intelligent agent can possess more intelligent and human-like capabilities for task suspension or task management. Therefore, the simulation planning and execution method for the intelligent agent in this embodiment of the invention can also support flexible task planning during the fuzzy simulation planning and execution process, enabling the intelligent agent to interrupt, suspend, resume, and destroy tasks. For example, by suspending tasks, the intelligent agent can handle multiple tasks concurrently; for instance, the intelligent agent can walk and chat with the user simultaneously. Thus, this can further improve the task planning and execution efficiency of the intelligent agent.

[0106] like Figure 3 and Figure 4 As shown, according to an embodiment of the present invention, in operation S202, based on the task execution feedback information of the fuzzy decision path, the second simulation planning process in the third planning execution process is invoked to generate the target decision path, including:

[0107] A fourth task intent is generated based on task execution feedback information;

[0108] Based on the fourth task intent, the preset skill library and preset action library are invoked to perform the second simulation planning process in the third planning execution process, generating the target decision path.

[0109] like Figure 3 As shown, in the environmental perception stage S1 of the third planning execution process S303, for the execution of the fuzzy decision path of "finding a mirror - walking to the mirror - looking in the mirror" in the second planning execution process S302 (which may also include the continuation of the fuzzy decision path after the fourth planning execution process S304), corresponding task execution feedback information can be generated. In the corresponding environmental perception stage S1, the agent can generate a new fourth task intent based on the task execution feedback information and the latest execution state information obtained by the agent: "After completing finding the mirror and walking to the mirror, continue to check the image content in the mirror to confirm whether there is anything on the face." Therefore, this fourth task intent is the latest intent of the agent to achieve the execution goal of "confirming whether there is anything on the face" based on the latest scene understanding information and the above task execution feedback information. At this time, the agent can obtain the reflected image content in the mirror as environmental perception data, which can include the agent's facial image at this time.

[0110] In the simulation planning phase S2 of the third planning execution process S303, the agent can further call upon the preset skill library and preset action library to plan and evaluate the task path for the fourth task intent. That is, through the second simulation planning process S322, multiple latest planning paths for skills and actions are generated, and each planning path is simulated and executed. For example, in the simulation execution process of checking the content of the image in the mirror, the agent needs to compare its historical facial image with the latest facial image in the mirror obtained in the environmental perception phase S1 to evaluate the simulation result of "whether there is dirt on the face". If dirt is found on the face, the simulation generates the action of "'There is dirt on the face'". Therefore, the agent can plan the action of "responding 'There is dirt on the face'" to form the latest path as the target decision path.

[0111] Therefore, the target decision path can be the optimal evaluation action path output by the agent based on the evaluation and decision-making of possible execution effects confirmed by the latest environmental perception data. This target decision path can have executable target skills and target actions, specifically represented by knowledge system codes from a preset skill library and a preset action library.

[0112] In the task execution phase S3 of the third planning execution process S303, the agent can effectively execute the aforementioned target decision path, reproducing the relevant target actions and skills, executing the target decision path of "looking in the mirror to confirm - replying that there is dirt on the face," and simultaneously updating the execution status of "finding the mirror" and the target decision path to the preset skill library. At this time, in the real environment E, the agent can perform the process of "replying that there is dirt on the face" to the user.

[0113] As can be seen, for the third planning execution process S303 mentioned above, the task intent and decision path can be updated based on the task execution feedback information of the second planning execution process S302 and the fourth planning execution process S304, thereby ensuring that the optimal target decision path can be generated and that the agent can complete the corresponding deterministic action task, thereby further improving the efficiency of simulation planning execution.

[0114] like Figure 3 and Figure 4 As shown, according to an embodiment of the present invention, in the second simulation planning process of generating a target decision path during the third planning execution process, which involves calling a preset skill library and a preset action library according to a fourth task intent, the process includes:

[0115] According to the fourth task intent, the preset skill library is invoked to simulate the successive execution of skills.

[0116] In response to the continuation of the simulation execution, the preset action library is invoked to perform the execution simulation judgment of the action to be executed, and at least one execution simulation path of the second simulation planning process in the third planning execution process is generated;

[0117] For each execution simulation path in at least one execution simulation path, a value assessment is performed to generate a target decision path.

[0118] like Figure 3 As shown, in the second simulation planning process S322, based on the fourth task intention of "after finding the mirror and walking to the mirror, continue to check the image in the mirror to confirm whether there is anything on the face," the preset skill library can be invoked to plan the corresponding skill knowledge for subsequent simulation execution. This subsequent simulation execution is relative to the fourth simulation planning process S324, and is a continuation of the simulation execution skills generated in the first simulation planning process S321.

[0119] Continuing simulation execution can be understood as the process of judging the expected execution result of an action instance for a skill that has not yet been completed in a simulation. For example... Figure 3As shown, in the second simulation planning process S322, the action instances from the preset action library are invoked through simulated execution skills for execution simulation judgment. Specifically, based on the action instances of the completed fuzzy decision paths, action path planning is performed again for the execution objective of "confirming whether there is something on the face," generating at least one execution simulation path. This execution simulation path can be a set of actions formed by the agent's replanning of fuzzy decision paths for partially completed action paths.

[0120] For each simulated execution path, a comprehensive value assessment can be performed by combining the agent's latest execution state and the latest environmental perception data corresponding to the scene understanding. For example, if the agent finds dirt on its face by looking in the mirror, the probability of the simulated execution path of "confirming and replying that there is dirt on the face" is evaluated as 70%. However, if the agent finds no dirt on its face by looking in the mirror, the probability of the simulated execution path of "confirming and replying that there is dirt on the face" is evaluated as 30%. But if the agent finds nothing on its face by looking in the mirror, the probability of the simulated execution path of "confirming and replying that there is dirt on the face" will decrease significantly.

[0121] Based on the value assessment of each execution simulation path described above, at least one execution simulation path can be filtered according to the magnitude of the assessment result, and the execution simulation path with the largest or smallest assessment result can be selected as the target decision path. Specifically, when it is confirmed that the expected execution results of the target decision path and the fuzzy decision path are consistent (e.g., responding that there is dirt on the face), the codes for the target skill and target action of the target decision path are consistent with the corresponding codes of the fuzzy decision path, thereby achieving a unified simulation execution effect when the target decision path is executed.

[0122] It's important to note that the main logic of the preset skill library comprises two aspects: policy functions (such as `sample_args`, which selects which tool to use to wipe what), and skill knowledge (such as `u_function`, which continuously searches for stains on objects and performs wiping actions). The preset skill library can call action instances from the preset action library, corresponding to the actions of wiping the mouth and wiping stains, respectively. In fact, the preset skill libraries can directly call each other or call actions, allowing for rich expression similar to code. It's worth noting that the `yield from` statement in the skill knowledge code can be used in Python to call a coroutine to obtain results. In this scenario expressed by the preset skill library, the `yield from` statement, during simulation, can directly call the model of the corresponding action instance to obtain the belief state change estimated by that action model; furthermore, when executed, the `yield from` statement can directly execute the action in environment E, returning the belief state obtained after perception.

[0123] In summary, the intelligent agent simulation planning and execution method described in the embodiments of the present invention can achieve at least one of the following technical effects:

[0124] First, by achieving a unified design for simulation execution through action and skill libraries, the agent can seamlessly switch between simulation and execution operations. Furthermore, the ability to freely combine skills, similar to code, greatly increases the flexibility of knowledge representation. In addition, the fuzzy programming design allows for modeling action or skill execution effects based on the scenario and task, eliminating the need to develop a single, unified model for all possible scenarios, significantly reducing development costs. Finally, the interruptible and resumable nature of the preset skill library in this embodiment makes alignment and combination exceptionally flexible. Simultaneously, each time an action is executed, the latest world state and belief state are acquired, ensuring the timeliness, accuracy, and reliability of the planning.

[0125] Based on the above-described simulation planning and execution method for intelligent agents, this invention also provides a simulation planning and execution device for intelligent agents. The following will be combined with... Figure 5 The device is described in detail.

[0126] Figure 5 A schematic diagram of a simulated planning execution device for an intelligent agent according to an embodiment of the present invention is shown.

[0127] like Figure 5 As shown, the intelligent agent simulation planning execution device 500 of this embodiment includes a fuzzy path generation module 510, a target path generation module 520 and a target path execution module 530.

[0128] The fuzzy path generation module 510 is used to generate a fuzzy decision path by calling a preset skill library and a preset action library to perform a first simulated planning process in the second planning execution process, based on the scene understanding information and corresponding execution exception feedback information obtained during the first planning execution process. In one embodiment, the fuzzy path generation module 510 can be used to perform the operation S201 described above, which will not be repeated here.

[0129] The target path generation module 520 is used to generate a target decision path by calling a preset skill library and a preset action library to perform a second simulated planning process in the third planning execution process based on the task execution feedback information of the fuzzy decision path. In one embodiment, the target path generation module 520 can be used to perform the operation S202 described above, which will not be repeated here.

[0130] The target path execution module 530 is used to execute the target skill and target action, which are consistent with the code corresponding to the fuzzy decision path in the first simulation planning process, during the third planning execution process, through the target decision path, thereby completing the simulation planning execution. In one embodiment, the target path execution module 530 can be used to execute the operation S203 described above, which will not be repeated here.

[0131] According to embodiments of the present invention, any plurality of modules in the fuzzy path generation module 510, the target path generation module 520, and the target path execution module 530 may be combined into one module, or any one of these modules may be split into multiple modules. Alternatively, at least a portion of the functionality of one or more of these modules may be combined with at least a portion of the functionality of other modules and implemented in one module. According to embodiments of the present invention, at least one of the fuzzy path generation module 510, the target path generation module 520, and the target path execution module 530 may be at least partially implemented as hardware circuitry, such as a field-programmable gate array (FPGA), a programmable logic array (PLA), a system-on-a-chip, a system-on-a-substrate, a system-on-package, an application-specific integrated circuit (ASIC), or any other reasonable means of integrating or packaging circuitry, or implemented in software, hardware, or firmware, or in any suitable combination of any of these three implementation methods. Alternatively, at least one of the fuzzy path generation module 510, the target path generation module 520, and the target path execution module 530 may be at least partially implemented as a computer program module, which, when run, can perform corresponding functions.

[0132] Figure 6 A block diagram of an electronic device suitable for implementing a simulated planning execution method for an intelligent agent, according to an embodiment of the present invention, is shown schematically.

[0133] The electronic device provided in the embodiments of the present invention includes one or more processors and a memory, wherein the memory is used to store one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors execute the above-described simulation planning execution method of the intelligent agent.

[0134] like Figure 6 As shown, an electronic device 600 according to an embodiment of the present invention includes a processor 601, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 602 or a program loaded from a storage portion 608 into a random access memory (RAM) 603. The processor 601 may include, for example, a general-purpose microprocessor (e.g., a CPU), an instruction set processor and / or an associated chipset and / or a special-purpose microprocessor (e.g., an application-specific integrated circuit (ASIC)), etc. The processor 601 may also include onboard memory for caching purposes. The processor 601 may include a single processing unit or multiple processing units for performing different actions of the method flow according to an embodiment of the present invention.

[0135] RAM 603 stores various programs and data required for the operation of electronic device 600. Processor 601, ROM 602, and RAM 603 are interconnected via bus 604. Processor 601 executes various operations of the method flow according to embodiments of the present invention by executing programs in ROM 602 and / or RAM 603. It should be noted that the programs may also be stored in one or more memories other than ROM 602 and RAM 603. Processor 601 may also execute various operations of the method flow according to embodiments of the present invention by executing programs stored in said one or more memories.

[0136] According to an embodiment of the present invention, the electronic device 600 may further include an input / output (I / O) interface 605, which is also connected to a bus 604. The electronic device 600 may also include one or more of the following components connected to the I / O interface 605: an input section 606 including a keyboard, mouse, etc.; an output section 607 including a cathode ray tube (CRT), liquid crystal display (LCD), etc., and a speaker, etc.; a storage section 608 including a hard disk, etc.; and a communication section 609 including a network interface card such as a LAN card, modem, etc. The communication section 609 performs communication processing via a network such as the Internet. A drive 610 is also connected to the I / O interface 605 as needed. A removable medium 611, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed on the drive 610 as needed so that computer programs read from it can be installed into the storage section 608 as needed.

[0137] The present invention also provides a computer-readable storage medium having executable instructions stored thereon, which, when executed by a processor, cause the processor to execute the above-described simulation planning execution method for an intelligent agent.

[0138] The computer-readable storage medium may be included in the device / apparatus / system described in the above embodiments; or it may exist independently and not assembled into the device / apparatus / system. The computer-readable storage medium carries one or more programs, which, when executed, implement the method according to the embodiments of the present invention.

[0139] According to embodiments of the present invention, a computer-readable storage medium may be a non-volatile computer-readable storage medium, such as including, but not limited to: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In the present invention, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. For example, according to embodiments of the present invention, a computer-readable storage medium may include ROM 602 and / or RAM 603 and / or one or more memories other than ROM 602 and RAM 603 described above.

[0140] Embodiments of the present invention also include a computer program product comprising a computer program that, when executed by a processor, implements the above-described simulation planning execution method for an intelligent agent.

[0141] The computer program includes program code for performing the methods shown in the flowchart. When the computer program product is run on a computer system, the program code is used to enable the computer system to implement the methods provided in the embodiments of the present invention.

[0142] When the computer program is executed by the processor 601, it performs the functions defined in the system / apparatus of this invention. According to embodiments of the invention, the systems, apparatuses, modules, units, etc., described above can be implemented by computer program modules.

[0143] In one embodiment, the computer program may rely on a tangible storage medium such as an optical storage device or a magnetic storage device. In another embodiment, the computer program may also be transmitted and distributed in the form of signals over a network medium, and downloaded and installed via the communication section 609, and / or installed from the removable medium 611. The program code contained in the computer program can be transmitted using any suitable network medium, including but not limited to: wireless, wired, etc., or any suitable combination thereof.

[0144] In such an embodiment, the computer program can be downloaded and installed from a network via the communication section 609, and / or installed from the removable medium 611. When the computer program is executed by the processor 601, it performs the functions defined in the system of this embodiment of the invention. According to embodiments of the invention, the systems, devices, apparatuses, modules, units, etc., described above can be implemented by computer program modules.

[0145] According to embodiments of the present invention, program code for executing the computer programs provided in the embodiments of the present invention can be written in any combination of one or more programming languages. Specifically, these computational programs can be implemented using high-level procedural and / or object-oriented programming languages, and / or assembly / machine languages. Programming languages ​​include, but are not limited to, languages ​​such as Java, C++, Python, "C", or similar programming languages. The program code can be executed entirely on the user's computing device, partially on the user's device, partially on a remote computing device, or entirely on a remote computing device or server. In cases involving remote computing devices, the remote computing device can be connected to the user's computing device via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computing device (e.g., via the Internet using an Internet service provider).

[0146] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present invention. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram or flowchart, and combinations of blocks in a block diagram or flowchart, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0147] Furthermore, all actions involving the acquisition of information, signals, or data in this invention are carried out in compliance with the relevant data protection laws, regulations, and policies of the country where the invention is located, and with the authorization granted by the owner of the corresponding device.

[0148] Those skilled in the art will understand that the features described in the various embodiments and / or claims of the present invention can be combined or combined in various ways, even if such combinations or combinations are not explicitly described in the present invention. In particular, the features described in the various embodiments and / or claims of the present invention can be combined or combined in various ways without departing from the spirit and teachings of the present invention. All such combinations and / or combinations fall within the scope of the present invention.

[0149] The embodiments of the present invention have been described above. However, these embodiments are merely illustrative and not intended to limit the scope of the invention. Although various embodiments have been described above, this does not mean that the measures in the various embodiments cannot be used advantageously in combination. The scope of the invention is defined by the appended claims and their equivalents. Various substitutions and modifications can be made by those skilled in the art without departing from the scope of the invention, and all such substitutions and modifications should fall within the scope of the invention.

Claims

1. A simulation planning and execution method for an intelligent agent, applied to fuzzy simulation execution unification, characterized in that, include: Based on the scenario understanding information and corresponding execution anomaly feedback information during the first planning execution process, the preset skill library and preset action library are invoked to perform the first simulated planning process during the second planning execution process, generating a fuzzy decision path; Based on the task execution feedback information of the fuzzy decision path, the preset skill library and preset action library are invoked to perform the second simulated planning process in the third planning execution process, and the target decision path is generated. as well as Through the target decision path, the target skills and target actions that are consistent with the fuzzy decision path code in the first simulation planning process are executed during the third planning execution process, thereby completing the simulation planning execution.

2. The method according to claim 1, characterized in that, Before generating the fuzzy decision path by invoking the preset skill library and preset action library to perform the first simulated planning process in the second planning execution process based on the scene understanding information and corresponding execution anomaly feedback information in the first planning execution process, the process further includes: During the execution of the first plan, a first task intent is generated based on the scenario understanding information; Based on the first task intent, a third simulated planning process is performed in the first planning execution process by calling a preset skill library and a preset action library to generate a simulated abnormal decision path. The simulated abnormal decision path is used to generate the execution abnormal feedback information when it is executed.

3. The method according to claim 1, characterized in that, The process of generating a fuzzy decision path by invoking a preset skill library and a preset action library to perform a first simulated planning process in the second planning execution process, based on scene understanding information and corresponding execution anomaly feedback information obtained during the first planning execution process, includes: During the execution of the second plan, a second task intent is generated based on the first task intent corresponding to the scenario understanding information and the execution exception feedback information; Based on the second task intent, the preset skill library and preset action library are invoked to perform the first simulated planning process in the second planning execution process, generating a fuzzy decision path.

4. The method according to claim 3, characterized in that, The first simulated planning process, which generates a fuzzy decision path, in the second planning execution process based on the second task intent and by calling a preset skill library and a preset action library, includes: Based on the second task intent, a preset skill library is invoked to generate a simulated execution skill; The simulated execution skill calls a preset action library to perform fuzzy simulation judgment of the execution action, generating at least one fuzzy simulation path of the first simulated planning process in the second planning execution process; For each fuzzy simulation path in at least one fuzzy simulation path, a value assessment is performed to generate the fuzzy decision path.

5. The method according to claim 1, characterized in that, Before generating the target decision path, the process further includes: (The text then goes on to describe the second simulation planning process within the third planning execution, which involves calling a preset skill library and a preset action library based on the task execution feedback information from the fuzzy decision path.) During the task execution process of the fuzzy decision path, the third task intent of the fourth planning execution process is generated based on the current execution status and the current scenario understanding information. Based on the third task intent, a fourth simulation planning process is performed in the fourth planning execution process by calling a preset skill library and a preset action library to generate an intermediate decision path that matches the task execution feedback information.

6. The method according to claim 4, characterized in that, The second simulation planning process in the third planning execution process, which generates the target decision path by calling a preset skill library and a preset action library based on the task execution feedback information of the fuzzy decision path, includes: A fourth task intent is generated based on the task execution feedback information; Based on the fourth task intent, the preset skill library and preset action library are invoked to perform the second simulated planning process in the third planning execution process, generating the target decision path.

7. The method according to claim 6, characterized in that, The second simulation planning process, which generates the target decision path, in the third planning execution process based on the fourth task intent and by calling the preset skill library and preset action library, includes: According to the fourth task intent, the preset skill library is invoked to perform the continuous simulation execution of the simulated execution skills; In response to the continuous simulation execution, a preset action library is invoked to perform execution simulation judgment of the execution action, and at least one execution simulation path of the second simulation planning process in the third planning execution process is generated; For each execution simulation path in at least one execution simulation path, a value assessment is performed to generate the target decision path.

8. A simulation planning execution device for an intelligent agent, applied to fuzzy simulation execution unification, characterized in that, include: The fuzzy path generation module is used to generate a fuzzy decision path by calling a preset skill library and a preset action library to perform the first simulated planning process in the second planning execution process based on the scene understanding information and corresponding execution exception feedback information in the first planning execution process. The target path generation module is used to generate a target decision path by calling a preset skill library and a preset action library to perform a second simulated planning process in the third planning execution process based on the task execution feedback information of the fuzzy decision path; as well as The target path execution module is used to execute the target skills and target actions that are consistent with the code corresponding to the fuzzy decision path in the first simulation planning process during the third planning execution process, through the target decision path, to complete the simulation planning execution.

9. An electronic device, comprising: One or more processors; Memory, used to store one or more programs. When the one or more programs are executed by the one or more processors, the one or more processors perform the method according to any one of claims 1 to 9.

10. A computer-readable storage medium having executable instructions stored thereon, which, when executed by a processor, cause the processor to perform the method of any one of claims 1 to 9.

Images