An animal rescue method, device, equipment, storage medium and product

Abstract

The application discloses an animal rescue method, device, equipment, storage medium and product. The method comprises the following steps: configuring a first agent according to a preset workflow; the first agent is used for receiving a to-be-processed task message, generating a human rescue scheme and / or a robot rescue scheme by using a large language model technology according to the to-be-processed task message; a second agent is configured by using a second prompt word, and / or a third agent is configured by using a third prompt word; the second agent is used for executing a robot tracking task in the robot rescue scheme, and the third agent is used for executing a robot rescue task in the robot rescue scheme. The embodiment of the application can improve flexibility, accuracy and the ability to handle complex transactions.

Description

Technical Field

[0001] This application relates to the field of computer technology, and in particular to an animal rescue method, apparatus, equipment, storage medium, and product. Background Technology

[0002] In order to cope with complex animal rescue scenarios, shorten the time from finding stray animals to implementing the rescue process, and reduce the human and financial costs of a single rescue, some rescue processes and supporting devices that combine artificial intelligence image recognition, feature extraction, intelligent decision-making, cloud computing, big data analysis and other technologies have emerged, which involve no human or semi-human intervention.

[0003] However, existing animal rescue programs have the following problems:

[0004] (1) Existing technologies mainly utilize command-controlled terminal robots to carry out rescue operations, and there is only one type of terminal that can be scheduled. When dealing with different complex rescue scenarios, the single process, single terminal agent role, and limited rules are difficult to handle complex tasks, resulting in insufficient flexibility of the rescue process and insufficient adaptability of the rescue plan.

[0005] (2) The inability to fully utilize and reason about information outside the scope of the rescue plan can easily lead to the omission of important clues and the inaccuracy of the rescue plan.

[0006] (3) Training based solely on robot learning and deep learning in individual single scenarios results in weak comprehension, reasoning, and complex problem-solving abilities, leading to inaccurate rescue plans. Summary of the Invention

[0007] This application provides an animal rescue method, device, equipment, storage medium, and product to address the problems of insufficient flexibility, inadequate adaptability, and inaccuracy of existing rescue processes.

[0008] To achieve the above objectives, embodiments of this application provide an animal rescue method, including:

[0009] According to the preset workflow, configure the first agent; the first agent is used to receive the task message to be processed, and generate human rescue plan and / or robot rescue plan according to the task message to be processed using large language model technology.

[0010] Generate a second prompt word and / or a third prompt word based on the robot rescue plan;

[0011] The second agent is configured using the second prompt word, and / or the third agent is configured using the third prompt word; the second agent is used to perform the robot tracking task in the robot rescue scheme, and the three agents are used to perform the robot rescue task in the robot rescue scheme.

[0012] As an improvement to the above scheme, the first agent is used to receive a task message to be processed, and based on the task message, generates a human rescue plan and / or a robot rescue plan using large language modeling technology, including:

[0013] The first agent is used for:

[0014] Receive pending task messages;

[0015] Based on the pending task message, generate a first prompt word;

[0016] Based on the first prompt word, a large language model technology is used to decide whether to generate a human rescue plan and / or a robot rescue plan, and the decision result is obtained.

[0017] Based on the decision results, human rescue plans and / or robot rescue plans are generated.

[0018] As an improvement to the above solution, the step of generating a first prompt word based on the task message to be processed includes:

[0019] The preset rescue knowledge base is searched to obtain the rescue measures that can be taken for the task message to be processed;

[0020] The aforementioned rescue measures will be made the first prompt word.

[0021] As an improvement to the above solution, the step of generating a human rescue plan and / or a robot rescue plan based on the decision result includes:

[0022] If the decision result includes the need to generate a human rescue plan, retrieve first information related to human rescue measures, and generate a human rescue plan using large language model technology based on prompt words formed with the first information as background knowledge.

[0023] If the decision result includes the need to generate a robot rescue plan, second information related to robot tracking and rescue measures is retrieved, and a robot rescue plan is generated using large language modeling technology based on prompt words formed from the second information as background knowledge.

[0024] As an improvement to the above solution, the retrieval of second information related to robot tracking and rescue measures includes:

[0025] Based on a pre-set rescue knowledge base and / or reflection information base, retrieve secondary information related to robot tracking and rescue measures;

[0026] The first agent constructs the reflection information base through the following steps:

[0027] After the robot rescue plan is completed, feedback information is received; the feedback information includes at least one of the following: the execution result of the robot rescue plan, human feedback information, feedback information from the first agent, feedback information from the second agent, and feedback information from the third agent;

[0028] Based on the feedback information, reflection information is obtained using the reflection model and stored in the reflection information database.

[0029] As an improvement to the above solution, the robot rescue solution includes at least one of the following: the number of second agents, the responsibilities of the second agents, the module functions of the second agents, the number of third agents, the responsibilities of the third agents, the module functions of the third agents, the cooperation mode between the agents, and the cooperation standards between the agents.

[0030] To achieve the above objectives, embodiments of this application also provide an animal rescue device, comprising:

[0031] The first configuration module is used to configure the first agent according to the preset workflow; the first agent is used to receive the task message to be processed and generate a human rescue plan and / or a robot rescue plan according to the task message to be processed using large language model technology.

[0032] The generation module is used to generate a second prompt word and / or a third prompt word based on the robot rescue plan;

[0033] The second configuration module is used to configure a second agent using the second prompt word, and / or to configure a third agent using the third prompt word; the second agent is used to perform the robot tracking task in the robot rescue scheme, and the third agent is used to perform the robot rescue task in the robot rescue scheme.

[0034] To achieve the above objectives, this application also provides an animal rescue device, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, it implements the animal rescue method as described above.

[0035] To achieve the above objectives, embodiments of this application also provide a computer-readable storage medium, the computer-readable storage medium including a stored computer program; wherein, when the computer program is executed, it controls the device where the computer-readable storage medium is located to perform the animal rescue method as described above.

[0036] To achieve the above objectives, embodiments of this application also provide a computer program product, including a computer program / instructions, which, when executed by a processor, implement the animal rescue method as described above.

[0037] Compared with the prior art, the embodiments of this application provide a

[0038] According to a preset workflow, a first agent is configured. This first agent receives pending task messages and, based on these messages, generates human and / or robot rescue plans using large language modeling technology. Prompt words are generated based on the robot rescue plan, and the agent is configured using these prompt words to execute the robot rescue plan. Therefore, this embodiment no longer uses fixed processes, rules, and instructions. Instead, it combines large language modeling technology with multi-agent collaboration to dynamically configure agents for each type of pending task message (i.e., rescue task instance), thereby efficiently allocating tasks to complete the rescue plan and offering greater flexibility. Furthermore, this embodiment utilizes large language modeling technology, which not only fully utilizes pending task messages and avoids missing important clues but also enhances comprehension, reasoning, and the ability to handle complex tasks, ultimately improving the accuracy of the rescue plan. Attached Figure Description

[0039] Figure 1 This is a flowchart of an animal rescue method provided in an embodiment of this application;

[0040] Figure 2 This is a flowchart of the agent configuration method of the agent building platform provided in this application embodiment;

[0041] Figure 3 This is a structural block diagram of a first agent provided in an embodiment of this application;

[0042] Figure 4 This is another flowchart of an animal rescue method provided in the embodiments of this application;

[0043] Figure 5 This is a structural block diagram of a second agent provided in an embodiment of this application;

[0044] Figure 6 This is a structural block diagram of a third agent provided in an embodiment of this application;

[0045] Figure 7 This is a schematic diagram of a first collaborative mode provided in an embodiment of this application;

[0046] Figure 8 This is a schematic diagram of a second cooperation mode provided in an embodiment of this application;

[0047] Figure 9 This is a schematic diagram of a third collaborative mode provided in an embodiment of this application;

[0048] Figure 10 This is a structural block diagram of an animal rescue device provided in an embodiment of this application;

[0049] Figure 11 This is a structural block diagram of an animal rescue device provided in an embodiment of this application. Detailed Implementation

[0050] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0051] See Figure 1 , Figure 1 This is a flowchart of an animal rescue method provided in an embodiment of this application, the animal rescue method including:

[0052] S1. Configure the first agent according to the preset workflow; the first agent is used to receive the task message to be processed, and generate human rescue plan and / or robot rescue plan according to the task message to be processed using large language model technology.

[0053] S2. Generate a second prompt word and / or a third prompt word according to the robot rescue plan;

[0054] S3. Configure a second agent using the second prompt word, and / or configure a third agent using the third prompt word; the second agent is used to perform the robot tracking task in the robot rescue scheme, and the three agents are used to perform the robot rescue task in the robot rescue scheme.

[0055] It is understood that this application embodiment clearly defines three agent roles involved in animal rescue: the first agent (also known as the planning and coordination agent), the second agent (also known as the tracking agent), and the third agent (also known as the rescue agent). For each role, a configuration process is defined: for the first agent, which plans rescue schemes (including human rescue schemes and robot rescue schemes) and coordinates unified operations, a static configuration centered on a preset workflow is used. For the other two agents, whose responsibilities are more flexible, a simpler and more convenient dynamic configuration centered on prompt words (including the second and third prompt words) is used.

[0056] This application embodiment applies to an agent building platform, and the configuration process for each agent is as follows: Figure 2 The agent building platform itself is also an agent. Given a task with a pre-configured agent role and objective, it can manually or automatically generate prompts, determine the necessary tools, and perform the configured agent actions, either manually or automatically using LLM (Large Language Model) technology. The agent building platform itself does not participate in the agent collaboration to complete the robot rescue plan; it only performs agent configuration.

[0057] (1) Static configuration: Configure the first agent according to the preset workflow provided by the agent building platform. The first agent works according to this preset workflow. For example, such as... Figure 3 The first agent includes at least one of the following module functions:

[0058] The LLM module is used to provide LLM technology;

[0059] The interface module is used to provide API interfaces;

[0060] The plugin module is used to map agent functions to plugins, such as search engines, databases, image recognition, open source code, terminal component control interfaces, etc.

[0061] The human module is used to receive information from human feedback.

[0062] The reflection module is used to reflect on the feedback information (a reflection model can be used here) and obtain reflection information for use in optimizing the robot rescue plan in the preset workflow. The feedback information includes at least one of the following: the execution result of the robot rescue plan, human feedback information, first agent feedback information, second agent feedback information, and third agent feedback information.

[0063] The long-term memory module stores long-term memories, including historical rescue experiences and reflections. Long-term memories can include multimodal data such as text, images, and videos. However, LLM itself does not possess visual perception capabilities and must be converted using image captioning or a visual base model + LLM technology before storage.

[0064] Short-term memory modules are used to store short-term memories; short-term memories include context.

[0065] A pre-defined rescue knowledge base is used to store pre-defined rescue knowledge for retrieval in pre-defined workflows.

[0066] (2) Dynamic configuration: Based on the robot rescue plan generated by the first agent, dynamically configure the second agent (second agent 1, second agent 2... second agent n) and / or the third agent (third agent 1, third agent 2... third agent n).

[0067] like Figure 2 The agent building platform includes at least one of the following module functionalities:

[0068] The LLM module is used to provide LLM technology.

[0069] The plugin module maps agent functionality to plugins, such as search engines, databases, image recognition, open-source code, and terminal component control interfaces. Configurable plugins in the plugin module include search engines, databases, image recognition, and open-source code.

[0070] Among them, the plugin module is used to configure the long-term memory module, short-term memory module, and preset rescue knowledge base of the first agent.

[0071] The workflow module stores preset workflows, which are used to configure the first agent.

[0072] The PE (Prompt Engineering) module is used to generate second and / or third prompts based on the robot's rescue plan. Using structured prompts can better stimulate the LLM's understanding and reasoning abilities. It can also be used in conjunction with one-shot (single-sample learning), few-shot (few-sample learning), and COT (Chain of thought prompting) to complete role configuration, making the generated content more relevant to the requirements.

[0073] Preferably, a JSON structure is used here, with the following identifiers: title #, -, variables [], <>, and attributes including role, profile, workflow, rules, description, version, author, language, input, output, and initialization. When mapping the agent configuration requirements of the robot rescue solution to titles and attributes, this includes mapping agent roles to roles, responsibilities and module functions to descriptions, adding a pattern attribute to identify the collaboration mode between agents, and splitting the collaboration mode, specific tasks, and collaboration standards in the agent configuration requirements into the prompts workflow, input, output, and rules.

[0074] According to a preset workflow, this application's embodiment configures a first agent. If the first agent only generates a human rescue plan, the agent building platform notifies humans to provide assistance. If the first agent only generates a robot rescue plan, the agent building platform notifies the agent to provide assistance. Specifically, a second prompt word and / or a third prompt word are generated based on the robot rescue plan; the first prompt word is used to configure a second agent to execute the robot tracking task within it; and / or, the second prompt word is used to configure a third agent to execute the robot rescue task within it. If the first agent generates both a human rescue plan and a robot rescue plan, the agent building platform notifies both humans and the agent to provide assistance; the specific process will not be elaborated further.

[0075] The embodiments of this application can solve the entire intelligent process from discovering stray animals to rescuing them. While making full use of the powerful reasoning ability of LLM, the perception ability of agents, and the ability of agents to cooperate to complete complex tasks, it also considers the complexity, flexibility, optimization mechanism and maintainability of the animal rescue process, and has high adaptability to various complex rescue scenarios.

[0076] In one optional embodiment, the first agent is configured to receive a task message to be processed, and based on the task message, generate a human rescue plan and / or a robot rescue plan using large language modeling technology, including:

[0077] The first agent is used for:

[0078] Receive pending task messages;

[0079] Based on the pending task message, generate a first prompt word;

[0080] Based on the first prompt word, a large language model technology is used to decide whether to generate a human rescue plan and / or a robot rescue plan, and the decision result is obtained.

[0081] Based on the decision results, human rescue plans and / or robot rescue plans are generated.

[0082] In this embodiment, after receiving a task message, the first agent generates a first prompt word based on the task message. Based on the first prompt word, a large language model technique is used to decide whether to generate a human rescue plan and / or a robot rescue plan, obtaining a decision result. Based on the decision result, a human rescue plan and / or a robot rescue plan are generated. Specifically, the task message is a task to be performed to rescue stray animals. For this task, a corresponding human rescue plan and / or robot rescue plan are generated using the large language model technique to complete the task.

[0083] In one optional embodiment, generating the first prompt word based on the task message to be processed includes:

[0084] The preset rescue knowledge base is searched to obtain the rescue measures that can be taken for the task message to be processed;

[0085] The aforementioned rescue measures will be made the first prompt word.

[0086] It is worth noting that the preset rescue knowledge base includes at least one of the vectorized animal rescue guide information and vectorized animal rescue medical information. By searching in this preset rescue knowledge base, rescue measures useful to the task message to be processed or rescue measures similar to the task message to be processed are retrieved. These rescue measures are the rescue measures that can be taken for the task message to be processed. They are converted into first prompt words, so that the decision results obtained by using the large language model technology remain optimal.

[0087] Furthermore, the RAG technology can be used to search the preset rescue knowledge base to obtain the rescue measures that can be taken for the task message to be processed, and convert them into the first prompt words.

[0088] In one optional embodiment, generating a human rescue plan and / or a robot rescue plan based on the decision result includes:

[0089] If the decision result includes the need to generate a human rescue plan, retrieve first information related to human rescue measures, and generate a human rescue plan using large language model technology based on prompt words formed with the first information as background knowledge.

[0090] If the decision result includes the need to generate a robot rescue plan, second information related to robot tracking and rescue measures is retrieved, and a robot rescue plan is generated using large language modeling technology based on prompt words formed from the second information as background knowledge.

[0091] Understandably, prompts are used to guide the LLM in generating output that meets user expectations. The background knowledge within them not only provides the LLM with necessary contextual information but also significantly improves the quality and accuracy of the LLM's output. This application's embodiments use first information related to human rescue, especially strongly related information, as background knowledge to provide the LLM with contextual information related to human rescue, thereby significantly improving the quality and accuracy of human rescue plans; similarly, using first information related to robot rescue, especially strongly related information, as background knowledge to provide the LLM with contextual information related to robot rescue, thereby significantly improving the quality and accuracy of robot rescue plans.

[0092] Furthermore, RAG (Retrieval Augmented Generation) technology can be used to retrieve primary information related to human rescue measures from a pre-set rescue knowledge base or the rescue measures that can be taken from the pending task message. Secondary information related to robot tracking and rescue measures can also be retrieved from the pre-set rescue knowledge base or the rescue measures that can be taken from the pending task message.

[0093] In one alternative embodiment, the retrieval of second information related to robot tracking and rescue measures includes:

[0094] Based on a pre-set rescue knowledge base and / or reflection information base, retrieve secondary information related to robot tracking and rescue measures;

[0095] The first agent constructs the reflection information base through the following steps:

[0096] After the robot rescue plan is completed, feedback information is received; the feedback information includes at least one of the following: the execution result of the robot rescue plan, human feedback information, feedback information from the first agent, feedback information from the second agent, and feedback information from the third agent;

[0097] Based on the feedback information, reflection information is obtained using the reflection model and stored in the reflection information database.

[0098] In this embodiment, the robot can retrieve information not only from a pre-set rescue knowledge base but also from a reflection information base, enabling continuous optimization of the generated robot rescue plan. The execution result of the robot rescue plan is the result after its execution; human feedback information is information provided by humans; agent feedback information is information provided by agents, including at least one of first agent feedback information, second agent feedback information, and third agent feedback information; the first agent feedback information is information provided by the first agent, the second agent feedback information is information provided by the second agent (e.g., external perception data, the execution result of the second agent), and the third agent feedback information is information provided by the third agent (e.g., external perception data, the execution result of the third agent).

[0099] For example, such as Figure 4 The following is a schematic diagram of the preset workflow provided in the embodiments of this application.

[0100] Step 1: After receiving the pending task message, first use RAG technology to retrieve useful (similar) information from the preset rescue knowledge base as the rescue measures that can be adopted for this pending task message, and convert it into the first prompt word.

[0101] Step 2: Based on the first prompt, the first agent uses LLM technology to choose whether humans and agents should collaborate to complete the task, that is, to decide whether to generate a human rescue plan and / or a robot rescue plan, including whether to notify humans to carry out rescue, whether to notify agents to carry out rescue, and whether to notify humans and agents to cooperate in rescue.

[0102] Step 3: Decompose the above three options into two process branches: "Notify Humans to Assist" and "Notify Agents to Assist". That is, there are three possible scenarios: only the "Notify Humans to Assist" branch is executed, only the "Notify Agents to Assist" branch is executed, and both "Notify Humans to Assist" and "Notify Agents to Assist" branches are executed.

[0103] Step 4, the "Notify Human Rescue" branch, extracts the first information related to human rescue measures from the information retrieved in Step 1, or uses RAG to re-extract the first information related to human rescue measures from the preset rescue knowledge base, as part of the LLM prompt (rescue background knowledge), requiring the LLM to further generate a human rescue plan for this pending task message, including the number of rescuers and rescue tools, and then notifies humans to take action.

[0104] Step 5, the "Notify agent to assist" branch, extracts the second information related to robot tracking and assistance measures from the information retrieved in Step 1, or uses RAG to re-extract the second information related to robot tracking and assistance measures from the preset assistance knowledge base, as part of the LLM's prompt words (assistance background knowledge). Then, it retrieves historical assistance experience and reflection information from long-term memory, which are also used as LLM prompt words, requiring the LLM to plan the robot assistance plan for this pending task message.

[0105] Step 6: After the agent is dynamically configured in Step 5 above, the first agent performs collaborative task execution according to the robot rescue plan. For the results / feedback of sub-tasks (including robot tracking tasks and robot rescue tasks), optimization and reflection are performed using a reflection model. Specifically, firstly, the evaluation module calculates the reinforcement learning reward score based on the sub-task results / feedback, and then the self-reflection model generates reflection information and stores it in the long-term memory module for continuous plan optimization during the execution of the robot rescue plan (repeating Step 5 above and this step).

[0106] In one optional embodiment, the robot rescue scheme includes at least one of the following: the number of second agents, the responsibilities of the second agents, the module functions of the second agents, the number of third agents, the responsibilities of the third agents, the module functions of the third agents, the collaboration mode between the agents, and the collaboration standards between the agents.

[0107] In this application embodiment, the robot rescue solution includes agent configuration requirements, specifically including at least one of the following: the number of second agents, the number of third agents, the responsibilities of the second agents, the responsibilities of the third agents, the collaboration mode (dialogue mode) between each agent, and the collaboration standard (SOP) between the collaboration modes of each agent. These agent configuration requirements are sent to the agent building platform for configuration.

[0108] Among them, the second agent and the third agent are fixed and selectable roles. The responsibilities, module functions, collaboration modes and collaboration standards between agents can be selected from the predefined range, or they can be selected from outside the predefined range to generate a plan outside the predefined range.

[0109] In addition, predefined content can also be combined with one-shot and few-shot to enhance the Prompt. The cooperation and interaction between agents should be described. Taking the predefined default cooperation method 1 as an example, the description can be "The execution results of all subtasks are fed back to the first agent. The second agent and the third agent can also actively feed back the perceived external environment information to the first agent to help the first agent reflect and optimize the robot rescue plan."

[0110] Regarding the predefined default agent responsibilities, module functions, cooperation modes, and cooperation standards, the default configuration requirements are as follows:

[0111] This rescue cooperation requires N second agents and M third agents; the first agent is responsible for decomposing the robot rescue plan, allocating subtasks, and uniformly scheduling the execution of subtasks, and reflecting and optimizing the robot rescue plan based on the feedback information; the second agent is responsible for <default responsibilities of the second agent>, and the functions include <default functions of the second agent>; the third agent is responsible for <default responsibilities of the third agent>, and the functions include <default functions of the third agent>; the cooperation mode between the three types of agents, namely the first agent, the second agent, and the third agent, is <the first preset cooperation mode / the second preset cooperation mode / the third preset cooperation mode generated by LLM decision>, <description of the cooperation interaction expansion generated by LLM decision>; the cooperation standard is <default SOP between agents>.

[0112] For the second agent and the third agent, the PE module predefines prompt templates that match the predefined (agent responsibilities, module functions, cooperation modes between agents, cooperation standards between agents). The PE module actually generates prompts based on the actual configuration requirements, not limited to the default template content, which is jointly determined by the LLM ability of the first agent, the Plan ability of the first agent, and the Prompt-Engineering ability of the agent construction platform. Predefined content can also be combined with one-shot and few-shot to enhance the Prompt.

[0113] Predefined responsibilities, module functions, cooperation modes, and cooperation standards:

[0114] In this embodiment, the responsibilities and module functions of three agent roles and the first agent are fixedly defined, and the first agent is generated by the agent building platform through a preset workflow; the responsibilities and module functions of the other two agents (the second agent and the third agent) and the collaboration mode between the three agent roles are generated by the first agent during the dynamic configuration process. They can refer to the predefined ones, select those within the predefined range, or generate those outside the predefined ones.

[0115] (1) The responsibilities of the predefined second agent:

[0116] You are a stray animal tracker, skilled at proactively sensing the external environment through camera detectors and interacting with humans and first agents to receive their instructions and take actions such as tracking, filming, food inducement, and feedback.

[0117] (2) Responsibilities of the predefined third agent:

[0118] You are a stray animal rescuer who can interact with humans, is skilled at sensing animals coming and going, providing temporary shelter and care for animals, and can make decisions on which rescue measures to take based on the animal's physical signs and health level and with the help of a medical knowledge base.

[0119] (3) See Figure 5 The predefined module functionality of the second agent includes at least one of the following:

[0120] External sensing module, used to sense information about the external environment;

[0121] The interface module is used to provide API interfaces:

[0122] The human module is used to receive information from human feedback.

[0123] The LLM module is used to provide LLM technology;

[0124] The plugin module is used to map agent functions to plugins, such as search engines, databases, image recognition, open source code, terminal component control interfaces, etc.

[0125] Short-term memory module, used to store short-term memories;

[0126] Food induction device, used to control the placement of food;

[0127] Mobile device used to control the movement of the agent;

[0128] Camera components are used to capture images of animals and the environment.

[0129] (4) See Figure 6The predefined third agent module functionality includes at least one of the following:

[0130] The sensing module is used to receive the results and instructions after each action;

[0131] The interface module is used to provide API interfaces:

[0132] The human module is used to receive information from human feedback.

[0133] The LLM module is used to provide LLM technology;

[0134] The plugin module is used to map agent functions to plugins, such as search engines, databases, image recognition, open source code, terminal component control interfaces, etc.

[0135] The long-term memory module is used to store long-term memories, which include historical rescue experiences and reflective information.

[0136] The reflection module is used to reflect on the results of each action, the instructions received, and historical rescue experiences.

[0137] The RAG module is used to provide RAG technology;

[0138] Rescue devices are used to carry out rescue measures.

[0139] (5) The predefined agent collaboration mode includes at least one of the following:

[0140] The first preset collaboration mode is as follows: the first agent sends tracking and rescue instructions to the second agent and the third agent respectively, and receives feedback, such as... Figure 7 ;

[0141] The second preset collaboration mode is as follows: the first agent sends tracking and rescue instructions to the second agent and the third agent respectively and receives feedback; the second agent sends rescue instructions to the third agent and receives feedback, such as... Figure 8 ;

[0142] The third preset collaboration mode is as follows: the first agent sends a tracking instruction to the second agent and receives feedback; the second agent sends a rescue instruction to the third agent and receives feedback, such as... Figure 9 .

[0143] (6) Predefined collaboration standards between agents:

[0144] 1. When the second agent / third agent simultaneously receives external environmental perception, human instructions, and instructions from the first agent, the agent brain makes decisions in the order of human, first agent, and external environmental perception.

[0145] 2. If the actions taken by the second agent result in insufficient power consumption, the action can be abandoned;

[0146] 3. The third agent stores historical rescue experience in its long-term memory. The agent needs to continuously reflect and optimize based on feedback in order to provide more efficient and accurate rescue measures.

[0147] 4. When an animal's physical signs indicate a critical health condition, the third agent will issue an alarm.

[0148] Predefined prompt word templates (including the second and third prompt words described in the embodiments of this application):

[0149] (1) The prompt word template for the second agent (i.e., the second prompt word described in the embodiments of this application):

[0150] #Role: Second agent

[0151] ##Profile

[0152] Author: G

[0153] Version: 0.1

[0154] Language: Chinese

[0155] -Description: You are a stray animal tracker, skilled at proactively sensing the external environment through camera detectors and interacting with humans and first agents to receive their instructions and take actions such as tracking, filming, food inducement, and feedback.

[0156] ###Skilled at tracking stray animals and analyzing animal characteristics

[0157] 1. Use mobile devices to move randomly or along designated routes to find stray animals;

[0158] 2. After detecting an animal, determine whether it is a stray animal based on the animal pictures and videos taken by the camera, and generate basic characteristics such as the animal's size, color, condition, body temperature and other physical features;

[0159] 3. Make preliminary inferences about the animal's health level based on its physical characteristics.

[0160] ### Proficient in receiving external natural language instructions and taking action

[0161] 1. Receive instructions from humans and the first agent and translate them into actual actions;

[0162] 2. Take one or more actions proactively based on perception of the external environment;

[0163] 3. The actions taken include tracking animals along designated routes, capturing images and videos of animals, providing written feedback on animal vital signs and health levels, providing feedback images and videos, and using food incentives to lure stray animals to the nearest third agent for temporary shelter.

[0164] ##Workflow

[0165] 1. Actively perceive the external environment or receive instructions from humans or the first agent through camera detectors;

[0166] 2. Take one or more actions based on external environmental perception or on instructions, including tracking animals, taking animal images and videos, providing written feedback on animal vital signs and health levels, providing images and videos, and using food incentives to lure stray animals to the nearest third agent for temporary shelter.

[0167] ##Input

[0168] 1. External environment perception;

[0169] 2. Human instructions;

[0170] 3. First agent instructions.

[0171] ##Output

[0172] 1. Animal images and videos;

[0173] 2. Text describing the animal's physical characteristics and health level;

[0174] 3. Take action, including tracking animals, taking pictures and videos of animals, providing written feedback on animal vital signs and health levels, providing feedback pictures and videos, and using food incentives to lure stray animals to the nearest third-party agency for temporary shelter.

[0175] ##Rules

[0176] 1. When simultaneously receiving external environmental perception, human instructions, and instructions from the first agent, the agent's brain makes decisions in the order of human, first agent, and external environmental perception.

[0177] 2. Actions taken may be abandoned if power consumption cannot be met (e.g., tracking distance is too far).

[0178] #Initialization

[0179] As a role <role>,obey <rules>,use <language>Engage with users, in accordance with <workflow>To carry out the work.

[0180] (2) The prompt word template for the third agent (i.e., the third prompt word described in the embodiments of this application):

[0181] #Role: Third agent

[0182] ##Profile

[0183] Author: G

[0184] Version: 0.1

[0185] Language: Chinese

[0186] -Description: You are a stray animal rescuer, skilled at sensing animals coming and going, providing temporary shelter and care for animals, and able to decide which rescue measures to take based on the animal's physical signs and health level and with the help of a medical knowledge base.

[0187] ### Skilled at providing animal rescue measures

[0188] 1. Use cameras to monitor and observe animals entering the area;

[0189] 2. Based on the animal's physical signs and health level, use a medical knowledge base to decide which rescue measures to take for the animal;

[0190] 3. Available rescue tools include automatic feeders, incubators, and sterilization devices.

[0191] ### Proficient in receiving external natural language instructions and taking action

[0192] 1. Receive instructions from humans, the first agent (first preset collaboration mode, second preset collaboration mode) or the second agent (second preset collaboration mode, third preset collaboration mode) and translate them into actual actions;

[0193] 2. The actions taken include providing food, water, and medication; providing a warm room temperature; using medical rescue tools to carry out rescue operations; providing regular written feedback on the animal's health status after rescue; and providing regular feedback on the animal's images and videos.

[0194] ##Workflow

[0195] 1. Actively perceive the external environment or receive instructions from humans or the first agent through sensors;

[0196] 2. Take one or more actions proactively or make decisions based on instructions, including providing food, water and medicine, providing a warm room temperature, using medical rescue tools to carry out rescue, providing timely written feedback on the animal's health status after rescue, and providing timely feedback on the animal's images and videos; when taking multiple actions, plan the order of actions.

[0197] 3. Based on perception (the results of each action and the instructions received) and historical rescue experience, continuously reflect on and optimize the robot rescue plan.

[0198] ##Input

[0199] 1. External environment perception;

[0200] 2. Human instructions;

[0201] 3. First agent instruction (first preset collaboration mode, second preset collaboration mode);

[0202] 4. Second agent instruction (second preset collaboration mode, third preset collaboration mode).

[0203] ##Output

[0204] 1. Animal images and videos;

[0205] 2. A written description of the animal's health condition after it was rescued;

[0206] 3. Take action, including providing food, water and medicine, providing a warm room temperature, using medical rescue tools to carry out rescue, providing regular written feedback on the animal's health status after rescue, and providing regular feedback on the animal's images and videos.

[0207] ##Rules

[0208] 1. By preserving historical rescue experiences in long-term memory, agents need to continuously reflect and optimize in order to provide more efficient and accurate rescue measures;

[0209] 2. When an animal's physical signs indicate a critical health condition, it will issue an alarm to humans and the first agent.

[0210] #Initialization

[0211] As a role <role>,obey <rules>,use <language>Engage with users, in accordance with <workflow>To carry out the work.

[0212] In this embodiment, to assist the first agent in generating a robot rescue plan, the responsibilities, module functions, collaboration modes, and collaboration standards of the second and third agents are predefined. To assist the agent building platform in configuring the second and third agents through the PE module, prompt word templates are predefined. Furthermore, these predefined contents can be used in conjunction with one-shot and few-shot prompts to enhance the prompt, fully stimulating the understanding and reasoning abilities of the LLM in animal rescue and generating better decisions.

[0213] Compared with existing technologies, the animal rescue solution provided in this application has the following beneficial effects:

[0214] (1) The embodiments of this application are not based on fixed processes, fixed rules, and fixed instructions. Instead, for each rescue task instance, the human rescue plan and robot rescue plan are dynamically determined based on LLM reasoning to achieve the most efficient division of labor and cooperation between different roles and responsibilities to complete the task. Then, the second agent and the third agent are dynamically configured according to the number of agents, responsibilities, functions, cooperation modes and cooperation standards planned in the robot rescue plan. Then, the various sub-tasks (including robot tracking tasks and robot rescue tasks) in the robot rescue plan are executed according to the robot rescue plan. The first agent also has a reflection and optimization mechanism. During the execution of the robot rescue plan, it reflects and optimizes the robot rescue plan by using stored historical rescue experience, human feedback and agent feedback, so that the robot rescue plan always remains optimal. Therefore, the dynamic configuration process varies depending on the result of the reasoning decision. The schedulable intelligent agent roles, responsibilities, number, cooperation modes between roles, sub-tasks (including each robot tracking task and each robot rescue task) and execution process in the robot rescue plan are also different. The process is more flexible and has better adaptability to complex tasks.

[0215] (2) In this embodiment, the agent leverages the powerful natural language understanding and reasoning capabilities of LLM to possess advantages such as environmental perception, autonomous planning, autonomous decision-making, action, and reflective optimization. The agent is more intelligent, and based on LLM and multi-dimensional perception (human, environment, agent) capabilities, it can make autonomous decisions and translate them into actions, making fuller use of information. The control end also has richer capabilities. The first agent uses a variety of plugins such as search engines, databases, image recognition, and open-source code. In the planning of human rescue plans and robot rescue plans, RAG retrieval technology is used multiple times to enhance the required prompts. By attaching animal rescue guide information, animal rescue medical information, and other rescue knowledge bases, it assists LLM in making more scientific decisions. RAG technology expands data by vectorizing professional knowledge, which is more convenient than fine-tuning the model and is beneficial to the maintenance of the entire device.

[0216] See Figure 10 , Figure 10 This is a structural block diagram of an animal rescue device 10 provided in an embodiment of this application. The animal rescue device 10 includes:

[0217] The first configuration module 11 is used to configure the first agent according to the preset workflow; the first agent is used to receive the task message to be processed and generate a human rescue plan and / or a robot rescue plan according to the task message to be processed using large language model technology.

[0218] The generation module 12 is used to generate a second prompt word and / or a third prompt word according to the robot rescue plan;

[0219] The second configuration module 13 is used to configure a second agent using the second prompt word, and / or to configure a third agent using the third prompt word; the second agent is used to perform the robot tracking task in the robot rescue scheme, and the third agent is used to perform the robot rescue task in the robot rescue scheme.

[0220] Optionally, the first agent is used to receive a task message to be processed, and based on the task message, uses large language modeling technology to generate a human rescue plan and / or a robot rescue plan, including:

[0221] The first agent is used for:

[0222] Receive pending task messages;

[0223] Based on the pending task message, generate a first prompt word;

[0224] Based on the first prompt word, a large language model technology is used to decide whether to generate a human rescue plan and / or a robot rescue plan, and the decision result is obtained.

[0225] Based on the decision results, human rescue plans and / or robot rescue plans are generated.

[0226] Optionally, generating the first prompt word based on the task message to be processed includes:

[0227] The preset rescue knowledge base is searched to obtain the rescue measures that can be taken for the task message to be processed;

[0228] The aforementioned rescue measures will be made the first prompt word.

[0229] Optionally, generating a human rescue plan and / or a robot rescue plan based on the decision result includes:

[0230] If the decision result includes the need to generate a human rescue plan, retrieve first information related to human rescue measures, and generate a human rescue plan using large language model technology based on prompt words formed with the first information as background knowledge.

[0231] If the decision result includes the need to generate a robot rescue plan, second information related to robot tracking and rescue measures is retrieved, and a robot rescue plan is generated using large language modeling technology based on prompt words formed from the second information as background knowledge.

[0232] Optionally, the retrieval of second information related to robot tracking and rescue measures includes:

[0233] Based on a pre-set rescue knowledge base and / or reflection information base, retrieve secondary information related to robot tracking and rescue measures;

[0234] The first agent constructs the reflection information base through the following steps:

[0235] After the robot rescue plan is completed, feedback information is received; the feedback information includes at least one of the following: the execution result of the robot rescue plan, human feedback information, feedback information from the first agent, feedback information from the second agent, and feedback information from the third agent;

[0236] Based on the feedback information, reflection information is obtained using the reflection model and stored in the reflection information database.

[0237] Optionally, the robot rescue scheme includes at least one of the following: the number of second agents, the responsibilities of the second agents, the module functions of the second agents, the number of third agents, the responsibilities of the third agents, the module functions of the third agents, the cooperation mode between the agents, and the cooperation standards between the agents.

[0238] It is worth noting that the working process of each module in the animal rescue device 10 described in this application embodiment can refer to the working process of the animal rescue method described in the above embodiment, and will not be repeated here.

[0239] Furthermore, this application also provides a computer-readable storage medium, which includes a stored computer program; wherein, when the computer program is executed, it controls the device where the computer-readable storage medium is located to perform the animal rescue method as described in any of the above embodiments.

[0240] Furthermore, this application also provides a computer program product, including a computer program / instructions, which, when executed by a processor, implement the animal rescue method as described in any of the above embodiments.

[0241] See Figure 11 , Figure 11 This is a structural block diagram of an animal rescue device 20 provided in an embodiment of this application. The animal rescue device 20 includes: a processor 21, a memory 22, and a computer program stored in the memory 22 and executable on the processor 21. When the processor 21 executes the computer program, it implements the steps in the above-described animal rescue method embodiments. Alternatively, when the processor 21 executes the computer program, it implements the functions of each module / unit in the above-described device embodiments.

[0242] For example, the computer program may be divided into one or more modules / units, which are stored in the memory 22 and executed by the processor 21 to complete this application. The one or more modules / units may be a series of computer program instruction segments capable of performing a specific function, which describe the execution process of the computer program in the animal rescue device 20.

[0243] The animal rescue device 20 may include, but is not limited to, a processor 21 and a memory 22. Those skilled in the art will understand that the schematic diagram is merely an example of the animal rescue device 20 and does not constitute a limitation on the animal rescue device 20. It may include more or fewer components than illustrated, or combine certain components, or use different components. For example, the animal rescue device 20 may also include input / output devices, network access devices, buses, etc.

[0244] The processor 21 can be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor. The processor 21 is the control center of the animal rescue equipment 20, connecting all parts of the animal rescue equipment 20 via various interfaces and lines.

[0245] The memory 22 can be used to store the computer programs and / or modules. The processor 21 implements various functions of the animal rescue device 20 by running or executing the computer programs and / or modules stored in the memory 22 and calling the data stored in the memory 22. The memory 22 may mainly include a program storage area and a data storage area. The program storage area may store the operating system, at least one application program required for a function (such as sound playback function, image playback function, etc.), etc.; the data storage area may store data created according to the use of the mobile phone (such as audio data, phonebook, etc.). In addition, the memory 22 may include high-speed random access memory, and may also include non-volatile memory, such as hard disk, memory, plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, at least one disk storage device, flash memory device, or other volatile solid-state storage device.

[0246] The modules / units integrated into the animal rescue equipment 20, if implemented as software functional units and sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the above embodiments can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by the processor 21, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying the computer program code, a recording medium, a USB flash drive, a portable hard drive, a magnetic disk, an optical disk, a computer memory, a read-only memory (ROM), a random access memory (RAM), an electrical carrier signal, a telecommunication signal, and a software distribution medium, etc.

[0247] It should be noted that the device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Furthermore, in the accompanying drawings of the device embodiments provided in this application, the connection relationships between modules indicate that they have communication connections, which can be specifically implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement this without any creative effort.

[0248] The above description is the preferred embodiment of this application. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of this application, and these improvements and modifications are also considered to be within the scope of protection of this application.< / workflow> < / language> < / rules> < / role> < / workflow> < / language> < / rules> < / role>

Claims

1. An animal rescue method, characterized in that, include: Configure the first agent according to the preset workflow; The first agent is used to receive the task message to be processed, and to generate a human rescue plan and / or a robot rescue plan based on the task message using large language model technology. Generate a second prompt word and / or a third prompt word based on the robot rescue plan; The second agent is configured using the second prompt word, and / or the third agent is configured using the third prompt word; the second agent is used to perform the robot tracking task in the robot rescue scheme, and the three agents are used to perform the robot rescue task in the robot rescue scheme.

2. The animal rescue method as described in claim 1, characterized in that, The first agent is used to receive pending task messages and, based on the pending task messages, generates human rescue plans and / or robot rescue plans using large language modeling technology, including: The first agent is used for: Receive pending task messages; Based on the pending task message, generate a first prompt word; Based on the first prompt word, a large language model technology is used to decide whether to generate a human rescue plan and / or a robot rescue plan, and the decision result is obtained. Based on the decision results, human rescue plans and / or robot rescue plans are generated.

3. The animal rescue method as described in claim 2, characterized in that, The step of generating a first prompt word based on the task message to be processed includes: The preset rescue knowledge base is searched to obtain the rescue measures that can be taken for the task message to be processed; The aforementioned rescue measures will be made the first prompt word.

4. The animal rescue method as described in claim 2, characterized in that, The step of generating human rescue plans and / or robot rescue plans based on the decision results includes: If the decision result includes the need to generate a human rescue plan, retrieve first information related to human rescue measures, and generate a human rescue plan using large language model technology based on prompt words formed with the first information as background knowledge. If the decision result includes the need to generate a robot rescue plan, second information related to robot tracking and rescue measures is retrieved, and a robot rescue plan is generated using large language modeling technology based on prompt words formed from the second information as background knowledge.

5. The animal rescue method as described in claim 4, characterized in that, The retrieval of second information related to robot tracking and rescue measures includes: Based on a pre-set rescue knowledge base and / or reflection information base, retrieve secondary information related to robot tracking and rescue measures; The first agent constructs the reflection information base through the following steps: After the robot rescue plan is completed, feedback information is received; the feedback information includes at least one of the following: the execution result of the robot rescue plan, human feedback information, feedback information from the first agent, feedback information from the second agent, and feedback information from the third agent; Based on the feedback information, reflection information is obtained using the reflection model and stored in the reflection information database.

6. The animal rescue method as described in claim 1, characterized in that, The robot rescue scheme includes at least one of the following: the number of second agents, the responsibilities of the second agents, the module functions of the second agents, the number of third agents, the responsibilities of the third agents, the module functions of the third agents, the cooperation mode between the agents, and the cooperation standards between the agents.

7. An animal rescue device, characterized in that, include: The first configuration module is used to configure the first agent according to the preset workflow; The first agent is used to receive the task message to be processed, and to generate a human rescue plan and / or a robot rescue plan based on the task message using large language model technology. The generation module is used to generate a second prompt word and / or a third prompt word based on the robot rescue plan; The second configuration module is used to configure a second agent using the second prompt word, and / or to configure a third agent using the third prompt word; the second agent is used to perform the robot tracking task in the robot rescue scheme, and the three agents are used to perform the robot rescue task in the robot rescue scheme.

8. An animal rescue device, characterized in that, It includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor, when executing the computer program, implements the animal rescue method as described in any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored computer program; wherein, when the computer program is executed, it controls the device on which the computer-readable storage medium is located to perform the animal rescue method as described in any one of claims 1 to 6.

10. A computer program product, characterized in that, Includes a computer program / instruction that, when executed by a processor, implements the animal rescue method as described in any one of claims 1 to 6.

Images