A vehicle-based information interaction method, device, equipment and medium

Abstract

The application relates to a vehicle-based information interaction method, device, equipment and medium. First, by acquiring the current scene description information of a target vehicle, more accurate and comprehensive target scene information can be acquired, and the problem of limited scene coverage range is solved. Secondly, by querying known description information matched with the target scene description information, data similar to or corresponding to the current scene can be found. Based on the known description information, more targeted question information can be generated. The problem of high scene construction cost is solved, and the process of constructing a large number of scenes is not needed. Then, through interaction with an intelligent agent, real-time feedback information can be acquired, and the behavior of the target vehicle is controlled according to the feedback information. The intelligent level of the interaction is improved. Finally, by analyzing the target interaction behavior, information such as the target behavior mode and intention can be further understood. The analysis result is stored, so that a decision basis can be provided for subsequent interaction.

Description

Technical Field

[0001] This invention relates to the field of vehicle control, and more specifically to a vehicle-based information interaction method, device, equipment, and medium. Background Technology

[0002] In the field of automotive human-vehicle interaction, a large number of scenario requirements and dialogue requirements have been discovered. However, these are all manually collected and then refined. Existing interaction scenarios are relatively general, and the dialogue processes are simple. Furthermore, the interaction defined by existing vehicles is a limited-state interaction. This interaction method requires pre-defining all possible business scenarios and constructing corresponding intents, dialogue processes, or execution actions.

[0003] Current predictive vehicle interaction processes suffer from the following problems: ① Limited coverage: If the user's needs are outside the predefined skill or scenario range, the system will be unable to identify and respond accordingly. ② High cost of scenario construction: Constructing intents, dialogue flows, or actions requires significant human involvement, including product design and development, resulting in high labor costs. ③ Low intelligence: Lacking self-learning and adaptive capabilities, it cannot autonomously adjust and optimize based on environmental and user behavior feedback. Summary of the Invention

[0004] In view of this, embodiments of the present invention provide a vehicle-based information interaction method, device, equipment, and medium to solve the problems of limited coverage, high scenario construction cost, and low intelligence in the interaction process of existing vehicles in different scenarios.

[0005] In a first aspect, embodiments of the present invention provide a vehicle-based information interaction method, characterized in that the method includes: Obtain the target scene description information of the target vehicle in the current time period; Query known target description information that matches the target scene description information, and generate question information based on the known target description information, wherein the storage time of the known target description information is greater than the storage time of the target scene description information; The intelligent agent is asked a question according to the question information to obtain the feedback information of the intelligent agent, and the target vehicle is controlled to perform the corresponding target interaction behavior based on the feedback information. The intelligent agent is constructed based on a generative language model. The analysis results are obtained based on the target interaction behavior and then stored.

[0006] In an optional embodiment of this application, obtaining the target scene description information of the target vehicle in the current time period includes: The environmental information of the target vehicle in the current time period and the action information of each intelligent agent of the target vehicle in the current time period are extracted from the first storage unit, wherein the first storage unit is used to store the scene description information of the target vehicle in each time period. The target scene description information is generated based on the environmental information and the action information.

[0007] The method provided in this application extracts environmental information and agent action information of the target vehicle from the short-term memory unit in the current time period. Based on this information, target scene description information is generated, which can reflect the changes in the vehicle's environment and actions in a timely manner. This facilitates subsequent comprehensive analysis, allowing for a more comprehensive description of the target vehicle's scene and providing more accurate analysis and judgment.

[0008] In an optional embodiment of this application, the querying of known descriptive information that matches the target scene descriptive information includes: Multiple candidate known description information are obtained based on the second storage unit; Calculate the correlation score between the target scene description information and each of the candidate known description information; The candidate known description information with the highest correlation score is determined as the target known description information.

[0009] The method provided in this application quantifies the degree of matching between each candidate description and the target scene by calculating the correlation score between the target scene description information and each candidate known description information. This allows for the accurate identification of the known description information with the highest correlation score.

[0010] In an optional embodiment of this application, calculating the association score between the target scene description information and each of the candidate known description information includes: Obtain the attenuation level and importance corresponding to the candidate known description information; Calculate the similarity between the target scene description information and each candidate known description information; The association score is calculated based on the degree of attenuation, the degree of importance, and the similarity.

[0011] In an optional embodiment of this application, obtaining the attenuation level and importance corresponding to the candidate known description information includes: Obtain the timestamp corresponding to the candidate known description information, and use the timestamp and the current time period to calculate the attenuation degree; Obtain the storage frequency and influence parameter corresponding to the candidate known description information, and calculate the importance based on the storage frequency and influence parameter, wherein the influence parameter is the influence value of the candidate known description information relative to other known description information.

[0012] In an optional embodiment of this application, generating query information based on the known description information of the target includes: Obtain the target questioning strategy corresponding to the known description information, wherein the target questioning strategy includes question format and guiding information; Semantic recognition is performed on the known description information of the target to obtain a first semantic recognition result; Initial information is generated using the first semantic recognition result and the question format, and the guiding information is filled into the initial information to obtain the question information.

[0013] The method provided in this application generates initial information using the first semantic recognition result and the question format, enabling rapid and automatic generation of question information. Filling the initial information with guidance information helps guide the agent to provide more specific and accurate information, improving the accuracy and effectiveness of the inquiry. It also provides guidance information to help the agent provide more specific and accurate information.

[0014] In one optional embodiment of this application, the step of performing actions on the intelligent agent according to the query information is... Perform a question-asking operation to obtain feedback information from the intelligent agent, including: Send the question information to the intelligent agent; Obtain the response information and / or control instructions generated by the intelligent agent based on the question information; The feedback information is generated based on the response information and / or control instructions.

[0015] The method provided in this application, by quickly sending query information and obtaining feedback information from the intelligent agent, can achieve the need to obtain required information or perform operations in real time and provide accurate feedback information. Through the interaction between the intelligent agent and the vehicle, the efficiency and convenience of operation are improved, reducing the user's time consumption and enhancing the user experience.

[0016] In an optional embodiment of this application, the step of analyzing the target interaction behavior to obtain analysis results and storing the analysis results includes: Obtain target scene observations associated with the target's interactive behavior from the target vehicle's memory stream; Semantic recognition is performed on the observation results of the target scene to obtain a second semantic recognition result; Based on the second semantic recognition result, multiple target question information is generated, and questioning operation is performed on the intelligent agent according to the question information to obtain target feedback information; The target feedback information is used as the analysis result and stored in the second storage unit.

[0017] The method provided in this application firstly performs semantic recognition on the observation results of the target scene to obtain a second semantic recognition result. This enables the parsing and understanding of data in the target scene, thereby obtaining more specific and accurate semantic information. This helps the system better understand the meaning and purpose of the target interaction behavior. Secondly, based on the second semantic recognition result, multiple target question messages are generated, allowing for in-depth exploration and understanding of key details in the target interaction behavior based on the obtained semantic information. This further improves the understanding and analysis capabilities of the target interaction behavior. Then, according to the question messages, a questioning operation is performed on the intelligent agent to obtain target feedback information and interact with the target, thereby obtaining more detailed information about the target interaction behavior. Finally, the advantage of storing the target feedback information as the analysis result in the second storage unit is that the obtained information about the target interaction behavior can be organized and saved to the long-term memory unit, facilitating subsequent analysis and application. This provides a basis and reference for further research on target interaction behavior.

[0018] In an optional embodiment of this application, after analyzing the target interaction behavior to obtain analysis results and storing the analysis results, the method further includes: Obtain historical interaction behaviors related to the target interaction behavior from the memory stream, and determine the historical time period corresponding to the historical interaction behavior; The interactive behavior corresponding to the first time period is obtained from the memory stream, wherein the first time period is the next historical time period of the historical time period; Based on the interaction behavior in the first time period, planning information corresponding to the second time period is generated, wherein the second time period is the next time period after the current time period.

[0019] Secondly, embodiments of the present invention provide a vehicle-based information interaction device, the device comprising: The acquisition module is used to acquire target scene description information of the target vehicle in the current time period; The query module is used to query known description information of the target that matches the target scene description information, and generate question information based on the known description information of the target. The processing module is used to perform a questioning operation on the intelligent agent according to the questioning information, obtain the feedback information of the intelligent agent, and control the target vehicle to perform corresponding target interactive behavior based on the feedback information, wherein the intelligent agent is constructed based on a generative language model; The analysis module is used to analyze the target interaction behavior to obtain analysis results and store the analysis results.

[0020] Thirdly, embodiments of the present invention provide a computer device, including: a memory and a processor, the memory and the processor being communicatively connected to each other, the memory storing computer instructions, and the processor executing the computer instructions to perform the method described in the first aspect or any corresponding embodiment thereof.

[0021] Fourthly, embodiments of the present invention provide a computer-readable storage medium storing computer instructions that cause a computer to perform the method described in the first aspect or any of its corresponding embodiments.

[0022] The beneficial effects of the embodiments of this application are as follows: ① The method provided in this application firstly obtains the current scene description information of the target vehicle to understand its specific situation and environment. This allows for more accurate and comprehensive target scene information, overcoming the problem of limited scene coverage. Secondly, by querying known description information that matches the target scene description information, data similar to or corresponding to the current scene can be found. Based on the known description information, more targeted questions can be generated. This solves the problem of high scene construction costs, eliminating the need for a large-scale scene construction process. Then, through interaction with the intelligent agent, real-time feedback information can be obtained, and the target vehicle's behavior can be controlled based on this feedback. This improves the intelligence level of the interaction and solves the problem of low intelligence during the interaction process. Finally, by analyzing the target's interactive behavior, further understanding of the target's behavior patterns, intentions, and other information can be obtained. Storing the analysis results can facilitate decision-making for subsequent interactions.

[0023] ② The method provided in this application extracts environmental information of the target vehicle and action information of the agent from the short-term memory unit in the current time period. Based on this information, target scene description information is generated, which can reflect the changes in the vehicle's environment and actions in a timely manner. This facilitates subsequent comprehensive analysis, provides a more comprehensive description of the target vehicle's scene, and thus provides more accurate analysis and judgment.

[0024] ③ The method provided in this application utilizes the first semantic recognition result and the question format to generate initial information, enabling rapid and automatic generation of question information. Filling the initial information with guidance information helps guide the agent to provide more specific and accurate information, improving the accuracy and effectiveness of the inquiry. It also provides guidance information to help the agent provide more specific and accurate information.

[0025] ④ The method provided in this application firstly performs semantic recognition on the observation results of the target scene to obtain a second semantic recognition result, which enables the parsing and understanding of the data in the target scene, thereby obtaining more specific and accurate semantic information. This helps the system better understand the meaning and purpose of the target interaction behavior. Secondly, based on the second semantic recognition result, multiple target question information is generated, which allows for in-depth exploration and understanding of key details in the target interaction behavior based on the obtained semantic information. This further improves the understanding and analysis capabilities of the target interaction behavior. Then, according to the question information, a questioning operation is performed on the intelligent agent to obtain target feedback information and interact with the target, thereby obtaining more detailed information about the target interaction behavior. Finally, the advantage of storing the target feedback information as the analysis result in the second storage unit is that the obtained information about the target interaction behavior can be organized and saved to the long-term memory unit, facilitating subsequent analysis and application. This can provide a basis and reference for further research on target interaction behavior. Attached Figure Description

[0026] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0027] Figure 1 This is a flowchart illustrating a vehicle-based information interaction method according to some embodiments of the present invention; Figure 2 This is a schematic diagram of the information processing process of an intelligent agent according to some embodiments of the present invention; Figure 3 This is a schematic diagram of air conditioning parameters according to some embodiments of the present invention; Figure 4 This is a schematic diagram of descriptive information obtained by converting air conditioning parameters according to some embodiments of the present invention; Figure 5 This is a schematic diagram of the intelligent agent interaction process according to some embodiments of the present invention; Figure 6 This is a structural block diagram of a vehicle-based information interaction device according to an embodiment of the present invention; Figure 7 This is a schematic diagram of the hardware structure of a computer device according to an embodiment of the present invention. Detailed Implementation

[0028] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0029] According to embodiments of the present invention, a vehicle-based information interaction method, apparatus, device, and medium are provided. It should be noted that the steps shown in the flowcharts in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowcharts, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0030] Before executing the information interaction method, this embodiment of the application needs to construct a sandbox environment, which contains elements such as past road information, environmental information, and various vehicles in the real scene. During the initialization process, all environmental and road information is aligned with the real world, and a smart agent is constructed using generative language models (LLM). The final smart agent contains a description of social relationships and character attributes, and this description information is permanently stored.

[0031] Generative language models (LLMs), based on deep learning and natural language processing techniques, are capable of generating coherent and logical sequences of language. When constructing intelligent agents, LLM models can be used to generate language instructions, descriptions, or decisions for interaction with the environment. For example, LLM models can generate instructions to control the agent's behavior, such as acceleration, braking, and steering; they can also generate statements describing environmental states, such as road conditions and traffic situations; or they can generate decision statements, such as choosing the best path or avoiding collisions. By combining real-world road and environmental information with the LLM-generated agent in a sandbox environment, various simulation tests can be conducted to improve the agent's performance and adaptability. This simulation and testing environment helps the agent better understand and cope with real-world challenges, improving its ability to be applied in real-world scenarios.

[0032] Within the sandbox, intelligent agents are configured, capable of generating complex interactive behaviors based on the current environment and needs. These behaviors are dynamic, constantly changing with time, memory, and environmental alterations. An intelligent agent is an entity capable of perceiving its environment, making decisions, and performing actions. Intelligent agents can be computer programs, robots, or other autonomous systems, possessing a degree of intelligence and autonomy. In the field of artificial intelligence, intelligent agents typically interact with their environment, acquiring information and performing corresponding operations.

[0033] The sandbox environment construction process in this embodiment is as follows: First, the environment is constructed, mainly including the in-vehicle and external environments. The in-vehicle environment includes in-vehicle signals, temperature, humidity, volume, media playback content, vehicle control unit status, and vehicle speed. The external environment includes weather, traffic congestion, brightness, and descriptions of the external conditions generated through graphics and text. Second, basic information is constructed, which may include user information and corresponding social relationships. User information may include personality, age, interests, user profile, etc. By constructing the sandbox environment, the intelligent agent can dynamically interact with the vehicle based on real-time environmental information.

[0034] This embodiment provides a vehicle-based information interaction method. Figure 1 This is a flowchart of a vehicle-based information interaction method according to an embodiment of the present invention, such as... Figure 1 As shown, the process includes the following steps: Step S11: Obtain the target scene description information of the target vehicle in the current time period.

[0035] In this embodiment of the application, obtaining the target scene description information of the target vehicle in the current time period includes the following steps A1-A2: Step A1: Extract the environmental information of the target vehicle in the current time period and the action information of each intelligent agent of the target vehicle in the current time period from the first storage unit. The first storage unit is used to store the scene description information of the target vehicle in each time period.

[0036] In this embodiment, the environmental information of the target vehicle at the current time period includes first environmental description information corresponding to the in-vehicle environment and second environmental description information corresponding to the out-of-vehicle environment. The first environmental description information includes in-vehicle signal values, temperature values, humidity values, volume values, media playback content, etc. The second environmental description information includes weather information, road condition information, etc.

[0037] The action information of each agent in the current time period can include: Action Description: Other intelligent agents may perform various actions, such as turning a device on or off, moving to a specific location, or performing a task. Action descriptions can include information such as the type of action, its target, parameters, and the result of its execution.

[0038] Language description: Intelligent agents may interact with the system through natural language to express intentions, needs, or provide information. These language descriptions can include questions, requests, instructions, answers, comments, etc.

[0039] State description: Agents may provide state descriptions of themselves or their environment.

[0040] Intent Description: Agents may express their intents or goals. For example, an agent may indicate that its intent is to provide the user with the information they need, or an agent may indicate that its intent is to perform a specific task.

[0041] Emotional or affective descriptions: Intelligent agents may communicate their emotional states or feelings through verbal descriptions. For example, an intelligent agent may indicate that it feels happy, frustrated, or confused.

[0042] Step A2: Generate target scene description information based on environmental information and action information.

[0043] In this embodiment of the application, after obtaining environmental information and action information, the two are combined to generate target scene description information corresponding to the current time period.

[0044] In the embodiments of this application, such as Figure 2 As shown, before retrieving the target scene description information from the first storage unit, it is necessary to collect the current environment, including data perceived by in-vehicle sensors such as vehicle speed, fuel level, and temperature; the status of the vehicle's infotainment system, such as music playback status and navigation information; and external environmental information, such as weather and road conditions. Simultaneously, the current actions of the intelligent agent are also collected. Then, the current environment and current actions are converted into a natural language description. Multimodal information refers to data from different perception modalities, such as images, speech, and text. This multimodal information is processed and analyzed, and then converted into a natural language description that humans can understand. For example, a scene in an image is described as a sentence, and instructions in speech are converted into text. In this way, the system can more easily understand and process this information and interact with the user in the form of natural language.

[0045] As an example, Figure 3 This is a schematic diagram of the collected air conditioning parameters. Figure 3The air conditioning parameters include: "AC_Environmental Temp" is the outside temperature, "AC Internal Temp" is the inside temperature, "RLS_DewPoint" is the moisture content, "RLS "Windshield Temperature" is the windshield temperature, "RLSSolarLeft" and "RLS SolarRight" are the left and right solar irradiance, respectively, and "RLS_relative humidity" is the relative humidity.

[0046] Based on the description information obtained from the above air conditioner parameter conversion to text and the recommended air conditioner parameter settings, such as... Figure 4 As shown, the specific content is as follows: I am currently in a car. The original outside temperature is 16.0 degrees Celsius, the inside temperature is 21.0 degrees Celsius, the left-side illuminance is 300.0 lux, the right-side illuminance is 300.0 lux, the windshield temperature is 18.0 degrees Celsius, the engine operating status is unknown, the inside dew point temperature is 22.0 degrees Celsius, the inside relative humidity is 49.0 RH, the inside carbon dioxide concentration is 0.0 ppm, the inside PM2.5 concentration is 0.0 L ug / m3, the car speed is 0.0 km / h, the current month is April, the current date is the 2nd, the current hour is 1 hour, the outside rainfall intensity is 0.0 level, and the outside PM2.5 concentration is 0.0 ug / m3. Please recommend a suitable car air conditioning setting for me.

[0047] Based on your description, the recommended car air conditioning settings are as follows: air conditioning is off, driver fan speed is 0.0, driver temperature is 27.0 degrees Celsius, air compressor is off, front airflow mode is unknown, air circulation mode is external circulation, and defrost is off.

[0048] I am currently in a car. The original outside temperature is 16.0 degrees Celsius, the inside temperature is 21.0 degrees Celsius, and the outside temperature is 26.0 degrees Celsius. Please recommend a suitable car air conditioning setting for me.

[0049] Based on your description, the current recommended settings for your car's air conditioning are as follows: air conditioning is off, driver's fan speed is set to level 0, driver's temperature is set to 25.5 degrees Celsius, air compressor is off, front airflow mode is unknown, air conditioning internal / external loop mode is external loop, and defrost is off.

[0050] It should be noted that the first storage unit can be understood as a short-term memory unit, mainly used to store the scene description information of the vehicle in each time period. By using the first storage unit to store the scene description information of the vehicle in each time period, it is convenient to conduct subsequent analysis of the scene description information and execute corresponding decisions.

[0051] The method provided in this application extracts environmental information and agent action information of the target vehicle from the short-term memory unit in the current time period. Based on this information, target scene description information is generated, which can reflect the changes in the vehicle's environment and actions in a timely manner. This facilitates subsequent comprehensive analysis, allowing for a more comprehensive description of the target vehicle's scene and providing more accurate analysis and judgment.

[0052] Step S12: Query the known target description information that matches the target scene description information, and generate question information based on the known target description information. The storage time of the known target description information is longer than the storage time of the target scene description information.

[0053] In this embodiment of the application, querying known descriptive information that matches the target scene description information includes the following steps B1-B3: Step B1: Obtain multiple candidate known description information based on the second storage unit.

[0054] In this embodiment, the second storage unit can be understood as a long-term memory unit, mainly used to store known descriptive information such as experiences, knowledge, and historical information from a relatively long period of time. Therefore, obtaining multiple candidate known descriptive information from the second storage unit can be achieved by using the time dimension to obtain relevant candidate known descriptive information from the second storage unit. For example, if the current time period is in the morning, then the time node corresponding to the candidate known descriptive information is also in the morning.

[0055] Step B2: Calculate the correlation score between the target scene description information and each candidate known description information; In this embodiment of the application, calculating the association score between the target scene description information and each candidate known description information includes the following steps: Step B201: Obtain the attenuation level and importance of the candidate known description information.

[0056] In this embodiment of the application, obtaining the attenuation level and importance corresponding to the candidate known description information includes: (1) Obtain the timestamp corresponding to the candidate known description information, and use the timestamp and the current time period to calculate the attenuation degree.

[0057] Specifically, the formula for calculating the degree of attenuation is as follows: The decay rate is calculated as 1 x exp(-t / d), where t is the time difference between the current time and the retrieved relevant information, and d is the decay factor or time scale parameter, representing the rate of decay over time.

[0058] (2) Obtain the storage frequency and influence parameters corresponding to the candidate known description information, and calculate the importance based on the storage frequency and influence parameters. The influence parameter is the influence value of the candidate known description information relative to other known description information.

[0059] Specifically, importance is used to distinguish between common types of known descriptive information and important types of known descriptive information. Importance indices are obtained through direct querying of the agent or calculated by designing an indicator. The formula for calculating importance is as follows: Importance = mf(m) × umd(m), where mf (Memory Frequency) is the frequency of occurrence of a specific piece of information m in the memory stream. It can be calculated by recording the number of recalls or corresponding counts for a known piece of information. The influence parameter umd (Uniqueness Memory Degree) is the influence value of a specific piece of information m relative to known descriptive information.

[0060] Step B202: Calculate the similarity between the target scene description information and each candidate known description information.

[0061] In this embodiment of the application, the target scene description information and each candidate known description information are converted into sentence vectors, and then the similarity between every two sentence vectors is calculated: cosine_similarity(v1, v2) = dot_product(v1, v2) / (norm(v1) × norm(v2)) Here, `dot_product` is the dot product operation of the sentence vectors. `norm` is the norm of the vectors, and `v1` and `v2` are the sentence vectors respectively.

[0062] Step B203: Calculate the association score based on the degree of attenuation, importance, and similarity.

[0063] In this embodiment of the application, the calculation process of the correlation score is as follows: final score = degree of attenuation × importance × relevance.

[0064] Step B3: The candidate known description information with the highest correlation score is determined as the target known description information.

[0065] In this embodiment of the application, the correlation score between the target scene description information and each candidate known description information is obtained, and the candidate known description information with the highest correlation score is determined as the target known description information.

[0066] The method provided in this application quantifies the degree of matching between each candidate description and the target scene by calculating the correlation score between the target scene description information and each candidate known description information. This accurately identifies the candidate known description information with the highest correlation score, providing precise target known description information.

[0067] In this embodiment of the application, generating query information based on known target description information includes the following steps C1-C3: Step C1: Obtain the target questioning strategy corresponding to the known descriptive information, wherein the target questioning strategy includes the question format and guiding information.

[0068] In this embodiment, the question format refers to the structure and form of the question, which determines the way the question is asked. It can be an open-ended question, a multiple-choice question, a fill-in-the-blank question, etc. Guiding information refers to additional information provided during the questioning process to guide the user in answering the question or providing relevant information.

[0069] Step C2: Perform semantic recognition on the known description information of the target to obtain the first semantic recognition result.

[0070] In this embodiment, the semantic recognition process is a process in which a generative language model uses natural language processing techniques to transform text information into corresponding semantic representations. In this step, the known descriptive information of the target is analyzed and parsed to obtain a first semantic recognition result. This result can be text analysis, classification, or other forms of semantic representation.

[0071] As an example, the known descriptive information of the target could be: During the evening rush hour, 25-year-old Xiao Zhang is listening to singer A's song Y on a highway in a rainy environment. The first semantic recognition result is: In-car environment: 22 degrees Celsius. Time period: Evening rush hour. Weather: Rainy weather. Environment: Highway. Media played: Singer A's song Y. Person information: Xiao Zhang. Age: 25.

[0072] Step C3: Generate initial information using the first semantic recognition result and the question format, and fill the initial information with the guiding information to obtain the question information.

[0073] In this embodiment, initial information is generated by combining the result of the first semantic recognition with the question format using a generative language model. A fragment of a sentence can be generated using the generative language model and then combined with a fixed part of the question format. Based on the key information that needs to be filled in the question format, the required guiding information is filled into the initial information. This guiding information can be obtained from a database, knowledge graph, or other resources. The initial information filled with guiding information is then combined to form the final question information.

[0074] For example, suppose the question format is: "How is the weather in {city} today?". The first semantic recognition result is a weather-related question, requiring city information. A generative language model is used to generate sentence fragments, such as "How is the weather today?". The city information is then filled into the initial information to obtain the complete sentence: "How is the weather in {city} today?". The final question information is the sentence filled with the guiding information.

[0075] The method provided in this application generates initial information using the first semantic recognition result and the question format, enabling rapid and automatic generation of question information. Filling the initial information with guidance information helps guide the agent to provide more specific and accurate information, improving the accuracy and effectiveness of the inquiry. It also provides guidance information to help the agent provide more specific and accurate information.

[0076] Step S13: Perform a questioning operation on the intelligent agent according to the questioning information, obtain the feedback information of the intelligent agent, and control the target vehicle to perform the corresponding target interaction behavior based on the feedback information. The intelligent agent is constructed based on a generative language model.

[0077] In this embodiment of the application, performing a questioning operation on the intelligent agent according to the questioning information to obtain the feedback information of the intelligent agent includes: sending the questioning information to the intelligent agent; obtaining the response information and / or control instructions generated by the intelligent agent based on the questioning information; and generating feedback information based on the response information and / or control instructions.

[0078] In this embodiment, the query information is sent to the intelligent agent via an appropriate communication method to request the acquisition of necessary information or the execution of corresponding operations. Obtaining the intelligent agent's response information and / or control commands: After receiving the query information, the intelligent agent processes it accordingly and generates response information and / or control commands as feedback information. The feedback information can be a text-based response or control commands used to control the target vehicle to perform corresponding target interactive behaviors. Specifically, such as... Figure 2 As shown, the intelligent agent can obtain corresponding instructions or responses through the plug-in system.

[0079] In this embodiment, the interaction between the intelligent agent and the vehicle is primarily achieved through the Tool module. The Tool module provides an interface connecting the intelligent agent and the vehicle, enabling the intelligent agent to send control commands to the vehicle and obtain vehicle status and environmental information. The Tool module acts as a bridge, connecting the intelligent agent and the vehicle, allowing them to communicate and interact effectively.

[0080] Specifically, the Tool module adapts to existing interaction interfaces, primarily to enable the intelligent agent to control the vehicle more realistically and provide real-time feedback on vehicle status and environmental information. The specific implementation process is as follows: Design an adapter: Since the voice control interface and vehicle control interface may differ from the Tool module, an adapter needs to be designed for intermediate conversion and processing to ensure correct data transmission. The adapter's function is to format the data output by the intelligent agent according to the requirements of the voice control interface and vehicle control interface.

[0081] Connect the vehicle's voice control system and the adapter: Connect the adapter to the vehicle's voice control system so that it can receive voice commands and pass them to the Tool module.

[0082] Parsing and processing voice and control commands: The adapter parses and processes received voice commands to understand the user's intent and translate it into specific control commands for the vehicle. Simultaneously, the adapter also passes these control commands to the Tool module for further processing.

[0083] Providing vehicle status feedback or voice responses, and updating environmental information: After receiving and processing control commands, the Tool module will take corresponding actions based on the actual situation of the vehicle and provide real-time feedback on the vehicle's status to the adapter. The adapter will then transmit this status information to the voice control system for voice responses to the user. Simultaneously, the Tool module will continuously update the vehicle's environmental information to provide more accurate data support for the intelligent agent.

[0084] As an example, a question is sent to the agent: "A: I made noodles for my child this morning, what should I make for breakfast tomorrow?" The agent receives and processes the question: After receiving the question, the agent uses speech recognition or text parsing technology to convert the question into an understandable semantic representation, understands the need, and replies: "Your child is currently in junior high school and has a lot of academic pressure. You could try making some nutritious breakfasts, such as eight-treasure porridge." This reply is the feedback information.

[0085] As another example, send a question to the agent: Send a question to the agent, such as "Please open the car window" or "Play a popular song." The agent receives and processes the question: After receiving the question, the agent uses speech recognition or text parsing technology to convert the question into an understandable semantic representation, recognizing the user's need to open the car window or play music. The agent generates corresponding control commands based on the user's needs. For the question of opening the car window, the agent will generate a control command such as: "Control the window motor downwards to open the window." For the question of playing music, the agent will generate a control command such as: "Turn on the car's music system and play a popular song." In this case, the control command is feedback information.

[0086] The method provided in this application, by quickly sending query information and obtaining feedback information from the intelligent agent, can achieve the need to obtain required information or perform operations in real time and provide accurate feedback information. Through the interaction between the intelligent agent and the vehicle, the efficiency and convenience of operation are improved, reducing the user's time consumption and enhancing the user experience.

[0087] Step S14: Analyze the target interaction behavior to obtain the analysis results and store the analysis results.

[0088] The method provided in this application first obtains the current scene description information of the target vehicle to understand its specific situation and environment. This allows for more accurate and comprehensive target scene information, overcoming the problem of limited scene coverage. Second, by querying known description information that matches the target scene description information, data similar to or corresponding to the current scene can be found. Based on the known description information, more targeted questions can be generated. This solves the problem of high scene construction costs, eliminating the need for a large-scale scene construction process. Then, through interaction with the intelligent agent, real-time feedback information can be obtained, and the target vehicle's behavior can be controlled based on this feedback. This improves the level of intelligence in the interaction and solves the problem of low intelligence during the interaction process. Finally, by analyzing the target's interactive behavior, further understanding of the target's behavioral patterns, intentions, and other information can be obtained. Storing the analysis results can facilitate decision-making for subsequent interactions.

[0089] like Figure 2 As shown, the process of obtaining analysis results based on target interactive behavior is as follows: obtaining interactive behavior through scene description information in the first storage unit, analyzing the interactive behavior, and storing the results in the second storage unit.

[0090] Specifically, the analysis results are obtained and stored based on the target interaction behavior, including the following steps D1-D4: Step D1: Obtain the target scene observation results associated with the target interaction behavior from the memory stream of the target vehicle.

[0091] In this embodiment of the application, the memory stream of the target vehicle includes scene observation results that develop over time. Therefore, the target scene observation results associated with the target interaction behavior can be obtained from the memory stream. For example, the most recent N scene observation results can be obtained.

[0092] Specifically, the scene observation results include: location information: observation results of the target vehicle at different times or locations, including the target vehicle's position, trajectory, parking location, etc. This information can be used to analyze the target vehicle's usage, driving path, etc.

[0093] Sensor data: Data observed by various sensors on the target vehicle (such as cameras, radar, GPS, etc.). Examples include video streams recorded by cameras and obstacle information detected by radar. These observations can be used to analyze the target vehicle's surrounding environment, detect obstacles, and identify targets.

[0094] Vehicle Status: Status information of the target vehicle, including vehicle speed, engine speed, fuel consumption, and battery level. This information can be used to analyze the target vehicle's performance and driving conditions.

[0095] User behavior: Observations of the behavior of users or drivers of the target vehicle, including driving habits, usage patterns, and operational behaviors. These observations can be used to analyze user needs, preferences, and driving behaviors.

[0096] External environment: Information about the environment surrounding the target vehicle, such as weather, road conditions, and traffic. This information can help understand the target vehicle's behavior and decisions in different scenarios.

[0097] Therefore, obtaining target scene observations associated with target interaction behaviors from the target vehicle's memory stream can include location information, sensor data, vehicle status, user behavior, and external environment information. These observations can help understand the background and context in which the target interaction behaviors occur, thereby providing more accurate and personalized services and support.

[0098] Step D2: Perform semantic recognition on the observation results of the target scene to obtain the second semantic recognition result.

[0099] In this embodiment, the observation results of the target scene are preprocessed, including data cleaning and format conversion, to ensure data accuracy and consistency. Useful feature information is extracted from the observation results. Specifically, different techniques such as text feature extraction and sound feature extraction can be used, and appropriate feature extraction methods are selected based on different data types. The extracted feature information is converted into an understandable semantic representation. A trained generative language model is then applied to recognize the converted semantic representation, resulting in a second semantic recognition result for the observation results of the target scene.

[0100] Step D3: Generate multiple target question information based on the second semantic recognition result, and perform questioning operations on the agent according to the question information to obtain target feedback information.

[0101] In this embodiment, the second semantic recognition result is analyzed and processed to determine the type and content of the target question information that can be generated based on the recognized semantic category or concept. Then, based on the analyzed target question information type and content, multiple specific target question information are generated. These question information can be inquiries about different aspects or details, used to further obtain relevant information or perform specific operations. The generated multiple target question information are used to ask questions to the intelligent agent to obtain the required information or perform corresponding operations. After receiving the target question information, the intelligent agent will perform corresponding processing and generate response information and / or control instructions as target feedback information. This feedback information may include text-based response content, status information, control instructions, etc.

[0102] Step D4: The target feedback information is used as the analysis result and stored in the second storage unit.

[0103] In this embodiment, firstly, semantic recognition is performed on the observation results of the target scene to obtain a second semantic recognition result. This allows for the parsing and understanding of the data in the target scene, resulting in more specific and accurate semantic information. This helps the system better understand the meaning and purpose of the target interaction behavior. Secondly, multiple target question messages are generated based on the second semantic recognition result. Based on the obtained semantic information, key details in the target interaction behavior can be explored and understood in depth. This further improves the understanding and analysis capabilities of the target interaction behavior. Then, the agent performs a questioning operation according to the question messages to obtain target feedback information and interact with the target, thereby obtaining more detailed information about the target interaction behavior. Finally, storing the target feedback information as the analysis result in the second storage unit has the advantage of organizing and saving the obtained information about the target interaction behavior, facilitating subsequent analysis and application. This provides a basis and reference for further research on target interaction behavior.

[0104] In this embodiment, the analysis of interactive behavior provides the agent with recent experience, and the analysis method is determined by identifying the questions that can be asked. Specifically, daily records are used to query a large language model in the agent's memory stream, allowing the LLM model to generate candidate questions. These candidate questions are then posed to the agent to elicit statements, which are then stored as facts in memory.

[0105] As an example, if you encounter traffic congestion on your way to work today, based on your memories of the day, we can generate candidate questions like, "I drove to work at 8 am today and encountered rush hour traffic, getting stuck in traffic for half an hour. What would you do next time?" The analysis result would be, "Next time I would go to work earlier or avoid congested areas," thus completing a full interaction behavior analysis. Alternatively, if the passenger describes that "passenger Zhang San turned on music and played a song by [name]," the analysis would conclude, "The passenger may like [name]'s songs," and this conclusion would be added to the storage unit.

[0106] In this embodiment of the application, after analyzing and storing the analysis results based on the target interaction behavior, the method further includes the following steps E1-E3: Step E1: Obtain historical interaction behaviors related to the target interaction behavior from the memory stream, and determine the historical time period corresponding to the historical interaction behavior.

[0107] In this embodiment, historical interaction behaviors related to the target interaction behavior are extracted from the memory stream. These historical interaction behaviors may include the target vehicle's previous actions, reactions, and interactions with other entities. Simultaneously, the system also needs to determine the historical time periods corresponding to these historical interaction behaviors for subsequent time period analysis and planning.

[0108] Step E2: Obtain the interaction behavior corresponding to the first time period from the memory stream, where the first time period is the next historical time period after the historical time period.

[0109] In this embodiment, interaction behavior data corresponding to the first time period is extracted from the memory stream. The first time period refers to the next time period after the historical time period, and can also be understood as the interaction behavior of the target vehicle in the most recent period. By acquiring this interaction behavior data, the recent dynamics and behavior of the target vehicle can be understood.

[0110] Step E3: Generate planning information for the second time period based on the interaction behavior in the first time period, where the second time period is the next time period after the current time period.

[0111] In this embodiment, planning information for the next time period is generated based on the interaction behavior in the first time period. This next time period is referred to as the second time period, and it is the time period following the current time period. Based on the interaction behavior in the first time period, the behavior pattern and movement of the target vehicle in the next time period are inferred. By generating planning information, the future behavior of the target vehicle can be predicted and planned, providing a basis for subsequent decision-making and control.

[0112] It's important to note that the purpose of planning using target interaction behavior is to ensure the long-term consistency of the agent. It describes the agent's future action sequence and helps maintain consistency in the agent's behavior over time. A plan includes location, event, start time, and duration. Planning begins from top to bottom and recursively generates more details.

[0113] First, a broad plan needs to be developed, outlining the day's agenda. To create an initial plan, the language model can be prompted using summary descriptions of the agents (e.g., name, features, and summaries of recent experiences) and summaries from the previous day. For example, if Zhang San is an office worker, he might plan to drive to and from get off work on a weekday; this is a routine plan.

[0114] Over time, the agent will refine and adjust its plans based on current circumstances and needs. For example, if Zhang San needs to arrive at the company early for a meeting on a certain day, the agent will adjust his start time and duration accordingly to ensure he arrives on time. In this way, the agent can flexibly formulate and adjust its plans based on current circumstances and needs, ensuring their consistency and effectiveness.

[0115] As an example, when a user initiates a voice command, "I made noodles for my child today, what should I make tomorrow?", the agent terminal can obtain current user information, child information, and environmental information based on the memory stream. Then, it retrieves relevant interaction information from the memory stream and finally generates relevant planning information using historical interaction behaviors. The planning information is as follows: March 11, 2023, 7:50-8:20: Description: Old Zhang started making breakfast for Little Zhang; today he made eight-treasure porridge.

[0116] Figure 5 A schematic diagram illustrating the specific implementation process of vehicle-based information interaction provided in the embodiments of this application, such as... Figure 5As shown, the system first collects descriptions of the in-vehicle environment, the out-of-vehicle environment, and actions. These descriptions are then organized to obtain scene description information, which is stored in short-term memory. Next, the system retrieves relevant known description information from long-term memory based on the organized scene description information. Based on this known description information, a question is generated and posed to the agent. Feedback is received, and the vehicle is controlled to perform corresponding interactive behaviors based on the feedback. The interactive behaviors are analyzed, and the analysis results are stored in long-term memory. Furthermore, the system can plan future actions of the vehicle based on the analysis results and interactive behaviors, and store the planned actions.

[0117] This embodiment also provides a vehicle-based information interaction device for implementing the above embodiments and preferred embodiments; details already described will not be repeated. As used below, the term "module" can refer to a combination of software and / or hardware that implements a predetermined function. Although the device described in the following embodiments is preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.

[0118] This embodiment provides a vehicle-based information interaction device, such as... Figure 6 As shown, it includes: Module 41 is used to obtain target scene description information of the target vehicle in the current time period; The query module 42 is used to query known description information of the target that matches the target scene description information, and to generate question information based on the known description information of the target. The processing module 43 is used to perform a questioning operation on the intelligent agent according to the questioning information, obtain the feedback information of the intelligent agent, and control the target vehicle to perform the corresponding target interactive behavior based on the feedback information, wherein the intelligent agent is constructed based on a generative language model; Analysis module 44 is used to analyze the target interaction behavior, obtain analysis results, and store the analysis results.

[0119] In this embodiment of the application, the acquisition module 41 is used to extract the environmental information of the target vehicle in the current time period and the action information of each intelligent agent of the target vehicle in the current time period from the first storage unit. The first storage unit is used to store the scene description information of the target vehicle in each time period; and to generate target scene description information based on the environmental information and action information.

[0120] In this embodiment of the application, the query module 42 is used to obtain multiple candidate known description information based on the second storage unit; calculate the correlation score between the target scene description information and each candidate known description information; and determine the candidate known description information with the highest correlation score as the target known description information.

[0121] In this embodiment of the application, the query module 42 is used to obtain the attenuation degree and importance of the candidate known description information; calculate the similarity between the target scene description information and each candidate known description information; and calculate the association score based on the attenuation degree, importance and similarity.

[0122] In this embodiment of the application, the query module 42 is used to obtain the timestamp corresponding to the candidate known description information, and calculate the attenuation degree using the timestamp and the current time period; obtain the storage frequency and influence parameter corresponding to the candidate known description information, and calculate the importance based on the storage frequency and influence parameter, wherein the influence parameter is the influence value of the candidate known description information relative to other known description information.

[0123] In this embodiment of the application, the query module 42 is used to obtain the target questioning strategy corresponding to the known descriptive information, wherein the target questioning strategy includes the questioning format and the guiding information; perform semantic recognition on the target known descriptive information to obtain the first semantic recognition result; generate initial information using the first semantic recognition result and the questioning format, and fill the initial information with the guiding information to obtain the questioning information.

[0124] In this embodiment of the application, the processing module 43 is used to send a question to the intelligent agent; obtain the response information and / or control instructions generated by the intelligent agent based on the question; and generate feedback information based on the response information and / or control instructions.

[0125] In this embodiment of the application, the analysis module 44 is used to obtain the target scene observation results associated with the target interaction behavior from the memory stream of the target vehicle; perform semantic recognition on the target scene observation results to obtain a second semantic recognition result; generate multiple target question information based on the second semantic recognition result, and perform questioning operation on the intelligent agent according to the question information to obtain target feedback information; and store the target feedback information as the analysis result in the second storage unit.

[0126] In this embodiment of the application, the device further includes: a planning module, configured to obtain historical interaction behaviors related to the target interaction behavior from the memory stream and determine the historical time period corresponding to the historical interaction behavior; obtain the interaction behavior corresponding to the first time period from the memory stream, wherein the first time period is the next historical time period of the historical time period; generate planning information corresponding to the second time period based on the interaction behavior of the first time period, wherein the second time period is the next time period of the current time period.

[0127] Please see Figure 7 , Figure 7 This is a schematic diagram of the structure of a computer device provided in an optional embodiment of the present invention, such as... Figure 7As shown, the computer device includes one or more processors 10, memory 20, and interfaces for connecting the components, including high-speed interfaces and low-speed interfaces. The components communicate with each other via different buses and can be mounted on a common motherboard or otherwise installed as needed. The processors can process instructions executed within the computer device, including instructions stored in or on memory to display graphical information of a GUI on external input / output devices (such as display devices coupled to the interfaces). In some alternative implementations, multiple processors and / or multiple buses can be used with multiple memories and multiple memory modules, if desired. Similarly, multiple computer devices can be connected, each providing some of the necessary operations (e.g., as a server array, a group of blade servers, or a multiprocessor system).

[0128] Processor 10 may be a central processing unit, a network processor, or a combination thereof. Processor 10 may further include a hardware chip. The hardware chip may be an application-specific integrated circuit (ASIC), a programmable logic device (PLD), or a combination thereof. The programmable logic device may be a complex programmable logic device (CAMP), a field-programmable gate array (FPGA), a general-purpose array logic (GDA), or any combination thereof.

[0129] The memory 20 stores instructions executable by at least one processor 10 to cause the at least one processor 10 to perform the method shown in the above embodiments.

[0130] The memory 20 may include a program storage area and a data storage area. The program storage area may store the operating system and applications required for at least one function; the data storage area may store data created based on the use of the computer device as shown by a landing page for an app. Furthermore, the memory 20 may include high-speed random access memory and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some alternative embodiments, the memory 20 may optionally include memory remotely located relative to the processor 10, which can be connected to the computer device via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0131] The memory 20 may include volatile memory, such as random access memory; the memory may also include non-volatile memory, such as flash memory, hard disk or solid-state drive; the memory 20 may also include a combination of the above types of memory.

[0132] The computer device also includes a communication interface 30 for communicating with other devices or communication networks.

[0133] This invention also provides a computer-readable storage medium. The methods described above according to embodiments of the invention can be implemented in hardware or firmware, or implemented as computer code that can be recorded on a storage medium, or implemented as computer code downloaded via a network and originally stored on a remote storage medium or a non-transitory machine-readable storage medium and then stored on a local storage medium. Thus, the methods described herein can be processed by software stored on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. The storage medium can be a magnetic disk, optical disk, read-only memory, random access memory, flash memory, hard disk, or solid-state drive, etc.; further, the storage medium can also include combinations of the above types of memory. It is understood that computers, processors, microprocessor controllers, or programmable hardware include storage components capable of storing or receiving software or computer code, which, when accessed and executed by the computer, processor, or hardware, implements the methods shown in the above embodiments.

[0134] Although embodiments of the invention have been described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of the invention, and such modifications and variations all fall within the scope defined by the appended claims.

Claims

1. A vehicle-based information interaction method, characterized in that, The method includes: Obtain the target scene description information of the target vehicle in the current time period; Query known description information of the target that matches the target scene description information, and generate question information based on the known description information of the target; The intelligent agent is asked a question according to the question information to obtain the feedback information of the intelligent agent, and the target vehicle is controlled to perform the corresponding target interaction behavior based on the feedback information. The intelligent agent is constructed based on a generative language model. The analysis results are obtained based on the target interaction behavior and then stored. The step of generating question information based on the known target description information includes: obtaining a target questioning strategy corresponding to the known description information, wherein the target questioning strategy includes a questioning format and guiding information; performing semantic recognition on the known target description information to obtain a first semantic recognition result; generating initial information using the first semantic recognition result and the questioning format, and filling the initial information with the guiding information to obtain the question information.

2. The method according to claim 1, characterized in that, The acquisition of the target scene description information of the target vehicle in the current time period includes: The environmental information of the target vehicle in the current time period and the action information of each intelligent agent of the target vehicle in the current time period are extracted from the first storage unit, wherein the first storage unit is used to store the scene description information of the target vehicle in each time period. The target scene description information is generated based on the environmental information and the action information.

3. The method according to claim 1, characterized in that, The query for known descriptive information that matches the target scene description information includes: Multiple candidate known description information are obtained based on the second storage unit; Calculate the correlation score between the target scene description information and each of the candidate known description information; The candidate known description information with the highest correlation score is determined as the target known description information.

4. The method according to claim 3, characterized in that, The calculation of the association score between the target scene description information and each candidate known description information includes: Obtain the attenuation level and importance corresponding to the candidate known description information; Calculate the similarity between the target scene description information and each candidate known description information; The association score is calculated based on the degree of attenuation, the degree of importance, and the similarity.

5. The method according to claim 4, characterized in that, The step of obtaining the attenuation level and importance corresponding to the candidate known description information includes: Obtain the timestamp corresponding to the candidate known description information, and use the timestamp and the current time period to calculate the attenuation degree; Obtain the storage frequency and influence parameter corresponding to the candidate known description information, and calculate the importance based on the storage frequency and influence parameter, wherein the influence parameter is the influence value of the candidate known description information relative to other known description information.

6. The method according to claim 1, characterized in that, The step of performing a questioning operation on the agent according to the questioning information to obtain the agent's feedback information includes: Send the question information to the intelligent agent; Obtain the response information and / or control instructions generated by the intelligent agent based on the question information; The feedback information is generated based on the response information and / or control instructions.

7. The method according to claim 1, characterized in that, The process of analyzing and storing the results based on the target interaction behavior includes: Obtain target scene observations associated with the target's interactive behavior from the target vehicle's memory stream; Semantic recognition is performed on the observation results of the target scene to obtain a second semantic recognition result; Based on the second semantic recognition result, multiple target question information is generated, and questioning operation is performed on the intelligent agent according to the question information to obtain target feedback information; The target feedback information is used as the analysis result and stored in the second storage unit.

8. The method according to claim 7, characterized in that, After analyzing the target interaction behavior to obtain analysis results and storing the analysis results, the method further includes: Obtain historical interaction behaviors related to the target interaction behavior from the memory stream, and determine the historical time period corresponding to the historical interaction behavior; The interactive behavior corresponding to the first time period is obtained from the memory stream, wherein the first time period is the next historical time period of the historical time period; Based on the interaction behavior in the first time period, planning information corresponding to the second time period is generated, wherein the second time period is the next time period after the current time period.

9. A vehicle-based information interaction device, characterized in that, The device includes: The acquisition module is used to acquire target scene description information of the target vehicle in the current time period; The query module is used to query known description information of the target that matches the target scene description information, and generate question information based on the known description information of the target; The processing module is used to perform a questioning operation on the intelligent agent according to the questioning information, obtain the feedback information of the intelligent agent, and control the target vehicle to perform corresponding target interactive behavior based on the feedback information, wherein the intelligent agent is constructed based on a generative language model; An analysis module is used to analyze the target interaction behavior to obtain analysis results and store the analysis results; The query module is used to obtain the target questioning strategy corresponding to the known description information, wherein the target questioning strategy includes a questioning format and guiding information; perform semantic recognition on the target known description information to obtain a first semantic recognition result; generate initial information using the first semantic recognition result and the questioning format, and fill the initial information with the guiding information to obtain the questioning information.

10. A computer device, characterized in that, include: A memory and a processor, the memory and the processor being communicatively connected to each other, the memory storing computer instructions, the processor executing the computer instructions to perform the method of any one of claims 1 to 8.

11. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions for causing a computer to perform the method of any one of claims 1 to 8.

Images