Application of AI agent generation task system to scenic spot script killing system

Abstract

The invention belongs to the technical field of game design, and discloses a method for applying an AI agent generation task system to a scenic spot script killer system, which comprises the following steps: generating an evaluation mode according to the type of the script killer, evaluating the related ability of a player under the type, generating a player portrait, sensing the current environment state, and generating an environment portrait; an initial task is allocated for each player, multi-agent simulation is performed by using an MARL model, narrative and scheduling targets are formulated through a strategic layer, the strategic layer is decomposed into specific tasks, personalized tasks are generated in combination with a reward function and a plot security guard mechanism, and difficulty coefficients of the players are dynamically adjusted; and the execution module receives the output of the dynamic decision module, forms a final task through assembly and rendering, and issues the final task to a player terminal, players receive the task and interact, automatically verify the task completion state, and feed back the task completion state to the multi-mode environment sensing module, so that circulation is formed, and dynamic generation of the scenic spot script killing task is realized.

Description

Technical Field

[0001] This invention relates to the field of game design technology, and more specifically, to an AI intelligent agent generation task system applied to a scenic area script-based murder mystery game system. Background Technology

[0002] Currently, large-scale real-life murder mystery games are becoming increasingly popular in China, attracting more and more young people. Murder mystery games involve setting up a real-life scene based on a pre-set script, where players take on roles and advance the plot by completing assigned tasks until the script is fully completed. This provides players with an immersive experience while also serving a social purpose. As a result, large-scale murder mystery games involving multiple players are gaining popularity.

[0003] However, existing scenic area murder mystery games mostly use paper task cards or fixed APP instructions, which results in a limited number of tasks and an inability to match suitable tasks according to individual abilities. The limited number of tasks leads to severe homogenization of tasks when there are too many players, with multiple tourists following the same task route, which can easily cause congestion and reduce the attractiveness of the script. Limiting the number of players also weakens the social attributes and economic benefits, making it difficult to match with large scenic areas. Secondly, the inability to assign suitable tasks to players is due to the different abilities of players. If easy tasks are assigned to more capable players and difficult tasks are assigned to less capable players, it will reduce the sense of participation and experience, giving players a bad experience. In addition, when assigning personalized tasks to players, the number of players (the flow of visitors to the scenic area) and the current weather changes will have a significant impact on the generation and allocation of tasks. Therefore, when generating tasks based on AI intelligent agents in scenic area murder mystery games, it is necessary to fully consider the number of players, their abilities, and the weather conditions in order to generate highly playable murder mystery game tasks. Summary of the Invention

[0004] To overcome the aforementioned shortcomings of existing technologies and achieve the above objectives, this invention provides the following technical solution: an AI intelligent agent generation task system applied to a scenic area script-based murder mystery game system, comprising:

[0005] Multimodal environment perception module: It acquires the type of murder mystery game and generates an evaluation method based on the type. It assesses the player's relevant abilities under this type, generates a player profile, and perceives the current environmental state to generate an environmental profile.

[0006] Dynamic Decision Module: Assigns initial tasks to each player based on player profiles and environment profiles, uses the MARL model for multi-agent simulation, sets narrative and scheduling goals at the strategic level, decomposes them into specific tasks at the tactical level, generates personalized tasks by combining reward functions and story security guard mechanisms, and dynamically adjusts the player's difficulty coefficient.

[0007] Task generation module: Receives the output of the dynamic decision module, assembles and renders it to form the final task, and sends it to the player's terminal. The player receives the task and interacts with it. The task completion status is automatically verified and fed back to the multimodal environment perception module, forming a loop.

[0008] Preferably, the method for generating a player profile by evaluating the player's relevant abilities under a given type based on the evaluation method includes:

[0009] The evaluation method includes mapping evaluation dimensions and test adjustment direction; mapping evaluation dimensions include the player evaluation dimensions for each type and the corresponding evaluation rules for each dimension; test adjustment direction includes the adjustment dimensions for the basic mini-games for each type and the adjustment rules for each dimension.

[0010] Design a basic mini-game, adjust it based on testing, generate an updated mini-game, and push it to the mobile device.

[0011] During the player's gameplay, behavioral data is acquired, and then processed according to the corresponding mapping evaluation dimensions to obtain an evaluation report;

[0012] A unique code is generated for each player, and the evaluation report is linked with the code to form a player profile.

[0013] Preferably, the method for sensing the current environmental state and generating an environmental profile includes:

[0014] Obtain the visitor flow, distribution density, and weather conditions of the scenic area, and standardize them into numerical vectors;

[0015] Using GIS tools, density is mapped onto a scenic area map, with color gradients representing risks, thus creating a heat map;

[0016] Combine the density vector and weather vector into a multidimensional vector; add metadata to the multidimensional vector, including timestamps and risk scores.

[0017] Heatmaps and multidimensional vectors are saved to the data center and output via API interface.

[0018] Preferably, the method for multi-agent simulation using the MARL model includes:

[0019] Define the state space, action space, and reward function. Select a combination of task elements from the action space that maximizes the total reward to generate a personalized task and assign it to the most suitable player.

[0020] The state space uses a multi-layered structure, including a strategic layer and a tactical layer. The strategic layer is responsible for setting narrative and scheduling goals from a global perspective, while the tactical layer is responsible for breaking down high-level instructions into specific tasks.

[0021] The action space is defined as selecting actions from the task component library;

[0022] The total reward of the reward function is defined as the weighted sum of experience reward, efficiency reward, story reward, and adaptive reward. Among them, experience reward is a negative matching score, which is used to quantify the fit between the task and the player's ability; efficiency reward is a negative congestion penalty, which is used to quantify the regional load within the scenic area; story reward is a positive logical consistency score, which is used to quantify the coherence of the task sequence; and adaptive reward is a positive environmental adjustment bonus, which is used to quantify the response to variables.

[0023] Based on player profiles, an initial difficulty coefficient is maintained for each player. When the tactical layer agent generates a task, a pre-designed plot security guard mechanism is used to perform rule verification on the task output by the tactical layer. If the verification passes, the task is sent to the task generation module; otherwise, it is regenerated.

[0024] The system continuously monitors the player's task completion status. If the player succeeds, the positive counter increments by 1; if the player fails, the negative counter increments by 1. The system adjusts the individual's difficulty coefficient based on the counter value and selects components based on the difficulty coefficient in the next task.

[0025] Preferably, the strategic layer's responsibility is to formulate narrative and scheduling objectives from a global perspective, specifically including:

[0026] The strategic layer is used to formulate the narrative goals and scheduling goals for the current stage; the narrative goals refer to the macro direction of the plot's progression, and the scheduling goals refer to the macro strategy for resource allocation;

[0027] The inputs are defined as overall story progress, heatmap, weather conditions, and remaining game time. The output is defined as high-level instructions, which are high-level guiding commands, and the frequency of custom decisions is allowed.

[0028] Preferably, the tactical layer's responsibility is to break down high-level instructions into specific tasks, including:

[0029] Receive high-level instructions from the strategic layer and break them down into specific, executable task elements. Each tactical agent is responsible for a player group or an area.

[0030] The inputs are defined as strategic instructions and micro-states. Strategic instructions are high-level guidance, and micro-states include each player's task progress and precise location, as well as the updated player profile. The output is defined as a personalized task assembled for a single player from the task component library.

[0031] Customizable decision-making frequency, higher than the decision-making frequency at the strategic level.

[0032] Preferably, the task component library includes:

[0033] Task elements refer to atomic-level task features, and the task element library refers to a pre-designed database of atomic-level task features.

[0034] Preferably, the rules for the positive and negative counters are as follows: if the player succeeds once, the positive counter increments by 1; if the player succeeds consecutively, the positive counter increments by 1 continuously until the player fails, at which point the positive counter stops counting and its value is reset to zero, and the negative counter increments by 1; if the player fails consecutively, the negative counter increments by 1 continuously until the player succeeds, at which point the negative counter stops counting and its value is reset to zero.

[0035] Preferably, the plot security guard mechanism is embedded after the tactical layer output as the final verification checkpoint, including logical consistency, plot lock and spatial rationality detection;

[0036] Logical consistency refers to the absence of contradictions between task rewards or clues and the player's existing information; plot lock refers to the unlocking status of a task depending on a plot node; spatial rationality check refers to the feasibility of a task in terms of physical space and time.

[0037] If the rule validation passes, the if-then rule chain is executed sequentially.

[0038] Preferably, the method for forming the final task through assembly and rendering includes: using dynamic narrative generation technology to fill task elements into a template and generate multimodal output.

[0039] The technical effects and advantages of the high-reliability memory chip data read / write verification method of the present invention are as follows:

[0040] 1. It solves the core pain points of existing scenic area murder mystery game tasks, such as congestion, poor experience, and limited scale caused by homogenization and static nature.

[0041] 2. The multimodal environment perception module dynamically generates accurate player ability profiles and environmental status profiles, providing a data foundation for personalized task allocation. Its core innovation lies in the dynamic decision-making module, which employs a hierarchical MARL model (strategic and tactical layers) to achieve synergy between macro-level storyline scheduling and micro-level personalized tasks. This allows for both strategic intelligence to optimize player flow and avoid congestion, and tactical intelligence to match each player with tasks of dynamic difficulty levels, ensuring a smooth flow experience. A storyline safety guardian mechanism guarantees the logical consistency and narrative coherence of generated tasks. Finally, the task generation module vividly renders task elements, enhancing immersion.

[0042] 3. It forms a closed loop of perception-decision-execution-feedback, which can realize the infinite dynamic generation and precise allocation of tasks in complex and ever-changing scenic environment (such as visitor flow and weather), thereby significantly improving the game's playability, social scale and economic benefits. Attached Figure Description

[0043] Figure 1 This is a schematic diagram of the AI ​​intelligent agent generation task system applied to a scenic area script murder mystery system in this invention;

[0044] Figure 2 This is a structural diagram of an electronic device for applying the AI ​​intelligent agent generation task system of this invention to a scenic area script murder mystery system. Detailed Implementation

[0045] 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, and 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.

[0046] Example 1

[0047] Please see Figure 1 This embodiment provides an AI-powered task generation system applied to a scenic area murder mystery game system. Its core innovation lies in abandoning the traditional "pre-generation-distribution" model and instead adopting a real-time dynamic cyclical model of "perception-evaluation-generation-optimization." The system treats the scenic area as a dynamically changing digital twin and players as intelligent agents with different attributes. By continuously sensing multi-party data and employing game theory and reinforcement learning algorithms, it dynamically generates and assigns personalized tasks to each player, thereby achieving optimal overall experience.

[0048] After each task is generated, the system monitors its performance (such as completion rate and player satisfaction) and optimizes the algorithm parameters in the next loop. The entire system is similar to an adaptive ecosystem, capable of handling the complexity of large scenic areas (such as thousands of players during peak periods).

[0049] Use tools such as UML or MindManager to define the interfaces between the various modules of the system. Integrate with a cloud computing platform (such as Alibaba Cloud or AWS) to support digital twin modeling, ensuring that the system can be scaled to scenic areas of different sizes.

[0050] Multimodal environment perception module: It acquires the type of murder mystery game and generates an evaluation method based on the type. It assesses the player's relevant abilities under this type, generates a player profile, and perceives the current environmental state to generate an environmental profile.

[0051] The acquisition of script-based murder mystery game genres is the starting point for this module. Genres typically include reasoning (emphasizing logic and clue analysis), adventure (emphasizing exploration and physical interaction), horror (focusing on psychological resilience and quick decision-making), and social (focusing on teamwork and role-playing), among others.

[0052] The acquisition methods include selecting from the script-based murder mystery menu provided in the APP or mini-program interface when players register or sign up to participate. Each script-based murder mystery is marked with a predefined category tag. For example, when a player signs up for "The Murder Mystery in the Ancient Town", the system automatically defines its associated type as reasoning type, and then transmits it to the system's data center through the API interface. The data center is the system's central cloud database (such as Alibaba Cloud RDS or AWS DynamoDB) used to store and process type data, ensuring data persistence and scalability, and supporting concurrent access by multiple users.

[0053] Based on the type of player, the evaluation method assesses the player's relevant abilities within that type, and the methods for generating player profiles include:

[0054] The evaluation method includes mapping evaluation dimensions and test adjustment direction; mapping evaluation dimensions include the player evaluation dimensions for each type and the corresponding evaluation rules for each dimension; test adjustment direction includes the adjustment dimensions for the basic mini-games for each type and the adjustment rules for each dimension.

[0055] Design a basic mini-game, adjust it based on testing, generate an updated mini-game, and push it to the mobile device.

[0056] During the player's gameplay, behavioral data is acquired, and then processed according to the corresponding mapping evaluation dimensions to obtain an evaluation report;

[0057] A unique code is generated for each player, and the evaluation report is linked with the code to form a player profile;

[0058] Evaluation dimensions refer to the evaluation directions of a player's abilities, such as reasoning ability, observation ability, etc. Evaluation rules refer to the quantitative rules for evaluation dimensions using behavioral data. Generally, they can be set as a function formula. For example, behavioral data includes task completion time, error rate, frequency of asking for help, movement efficiency, and number of times items are used, etc. Then, the evaluation rule for reasoning ability can be set as: Reasoning ability = Task completion time × 0.3 + Error rate × 0.5 + Number of times items are used × 0.2. In the calculation process, pure numerical calculations are used without involving dimensions, and all are normalized. 0.3, 0.5, and 0.2 represent the weights of behavioral data, and the sum of the weights should be 1.

[0059] Adjustable dimensions refer to the directions that can be adjusted in the basic mini-game. For example, if the basic mini-game is a treasure hunt game, the adjustable directions include the number of obstacles, the direction of movement, and the interval between the generation of dynamic obstacles. These can be used as adjustment dimensions. The adjustment rules can be the adjustment range of each dimension. For example, for a mystery novel, the interval between the generation of dynamic obstacles in the basic mini-game can be shortened by 40%.

[0060] The evaluation report and the code are associated by storing them as database key-value pairs. For example, a NoSQL database (such as MongoDB) can be used to store the unique code as the primary key (e.g., player ID: "UUID-12345") and the evaluation report as the value (e.g., {reasoning ability: 75, observation ability: 60}). Timestamps and version numbers are added to support dynamic updates and ensure that the profile can be quickly queried and version controlled.

[0061] Methods for perceiving the current environmental state and generating an environmental profile include:

[0062] Obtain the visitor flow, distribution density, and weather conditions of the scenic area, and standardize them into a numerical vector (using normalization, such as Min-Max scaling to [0,1]);

[0063] Visitor traffic and density in scenic areas can be statistically analyzed in real time using scenic area cameras (which can detect crowds using YOLO) and positioning devices (such as iBeacon or UWB) (the data collection frequency can be set to refresh every minute).

[0064] Weather status can be obtained by accessing APIs (such as China Weather Network or OpenWeatherMap) to obtain temperature, rainfall, wind speed, etc., and combined with local IoT sensors (such as rain gauges and thermometers) to enhance accuracy and obtain real-time weather conditions.

[0065] Use GIS tools (such as ArcGIS or Python's Folium library) to map density onto the scenic area map, with color gradients representing risks (e.g., red for density > 50), to create a heat map;

[0066] Combine the density vector and weather vector into a multidimensional vector (e.g., density vector: [region A: 0.8, region B: 0.2], weather vector: [rainfall: 0.6, temperature: 0.5]); add metadata to the multidimensional vector, including timestamps and risk scores (which can be calculated using predefined rules, e.g., risk score = density × 0.5 + rainfall × 0.3 + wind speed × 0.2, where 0.3, 0.5, and 0.2 represent weights, and the sum of the weights should be 1);

[0067] Heatmaps and multidimensional vectors are saved to the data center and output via API to ensure real-time availability and visualization support (e.g., displaying heatmaps on management system dashboards for administrator monitoring).

[0068] Overall example:

[0069] Suppose a player registers for the "Ancient Town Murder Mystery" (Type: Deduction). First, the system retrieves the type through the app menu and transmits it to the central cloud database via API. Next, it generates an evaluation method based on the type:

[0070] The mapping evaluation dimensions are {reasoning ability: weight 0.4, observation ability: 0.3, social ability: 0.3}, and the evaluation rule is: reasoning ability = (normalized task completion time × 0.4) + (normalized error rate × 0.4) + (normalized help frequency × 0.2);

[0071] The testing has been adjusted, with changes to the basic mini-game "Find the Treasure Chest": the interval between dynamic obstacles has been reduced by 30%, and the number of clues has been increased by 20%. Updates to the basic mini-game (such as adding logic puzzles) have been generated and pushed to the player's app.

[0072] After a player completes a mini-game, behavioral data (e.g., duration: 2 minutes, error rate: 10%) is collected, processed, and an evaluation report (e.g., reasoning ability: 80) is generated. A unique code "PID-56789" is generated for each player, and a player profile is formed by associating these codes with MongoDB key-value pairs.

[0073] Simultaneously, it senses the environmental status: acquires passenger flow (camera detection density: 30 people / 100㎡) and weather (API shows light rain), and generates an environmental profile (heat map shows congested areas in red, vector: [density: 30, rainfall: 2mm / h]).

[0074] Dynamic decision-making is the "brain" of the system, based on a game theory-based Multi-Agent Reinforcement Learning (MARL) model. The system is viewed as a chess player, introducing a multi-agent collaborative decision-making framework of "hierarchical goals" and "course learning," with the scenic area as the chessboard and the players as the pieces. Decision-making is divided into strategic and tactical layers, defining state space, action space, and reward function. Task component combinations are generated based on the current state to reduce computational complexity, ensure macro-level narrative, and achieve precise micro-level matching. The goal of each decision step (generating a task) is to optimize the entire "chess game" (the overall experience of all players).

[0075] Methods for multi-agent simulation using MARL models (such as RayRLlib or OpenAIGym) include: the MARL model selects a combination of task elements from the action space that maximizes the total reward based on the current state space, generates a personalized task, and assigns it to the most suitable player. Specifically:

[0076] Define the state space, action space, and reward function. Select a combination of task elements from the action space that maximizes the total reward to generate a personalized task and assign it to the most suitable player.

[0077] The state space employs a multi-layered structure, including a strategic layer and a tactical layer. This breaks down massive decisions into strategic (macro) and tactical (micro) layers, avoiding the "computational disaster" and narrative inconsistencies. Specifically, the strategic layer is responsible for setting narrative and scheduling goals from a global perspective, while the tactical layer is responsible for breaking down high-level instructions into specific tasks.

[0078] The action space is defined as selecting actions from the task component library;

[0079] The total reward of the reward function is defined as the weighted sum of experience reward, efficiency reward, story reward, and adaptability reward (e.g., weights set as 40% experience, 30% efficiency, 20% story, and 10% adaptability). Experience reward is a negative matching score used to quantify the compatibility between the task and the player's abilities. For example, matching score = (task difficulty - player ability difference)^2 negative (where task difficulty is a preset value of 0-100, and player ability difference is the absolute difference; negative values ​​encourage minimizing the difference). Player ability difference refers to XX. Efficiency reward is a negative congestion penalty used to quantify the regional load within the scenic area. = -Density*10 (Density is players / square meter, multiplied by 10 as a penalty coefficient to ensure that the total reward is significantly reduced when the density is high), the plot reward is a positive logical consistency score, used to quantify the coherence of the task sequence. For example, NLP (such as the BERT model) is used to check logical consistency (BERT calculates the semantic similarity of the task sequence, if the score is greater than 0.8, it is +10 points, otherwise -5 points), the adaptive reward is a positive environmental adjustment bonus, used to quantify the response to variables. Here, the variable refers to XX, for example, successful weather adjustment is +5 points (e.g., if the task is successfully moved from outdoors to indoors, it is +5 points; if it fails, it is 0 points).

[0080] The experiential rewards are designed to match high-level players with challenging tasks and low-level players with easy tasks. Efficiency rewards guide players away from congested areas and towards less crowded ones. If a player density in an area is too high, the system will reduce the reward value for tasks assigned to that area. Story rewards ensure that task combinations align with the core plot logic, avoiding inconsistencies. Adaptability rewards adjust tasks based on weather conditions (for example, replacing the "reading the stone tablet outdoors" task with "obtaining information by talking to an NPC inside the pavilion" when it rains) to yield higher rewards.

[0081] Based on player profiles, an initial difficulty coefficient is maintained for each player. When the tactical layer agent generates a task, a pre-designed plot security guard mechanism is used to perform rule verification on the task output by the tactical layer. If the verification passes, the task is sent to the task generation module; otherwise, it is regenerated.

[0082] The system continuously monitors the player's task completion status. If the player succeeds, the positive counter is incremented by 1; if the player fails, the negative counter is incremented by 1. The individual difficulty coefficient is adjusted based on the counter value, and the component is selected based on the difficulty coefficient in the next task.

[0083] Task completion status refers to the results of task execution, including success rate, time taken, frequency of requests for help, and number of errors. For example, success = completion rate > 80% and time taken < expected time, failure = requests for help > 3 times or timeout.

[0084] The method for selecting components based on difficulty coefficient is as follows: When combining components at the tactical level, select components with matching difficulty from the component library (for example, if the difficulty coefficient is >0.7, select high-difficulty challenge components, such as complex puzzles; if the difficulty coefficient is <0.4, select low-difficulty components, such as simple hints). At the same time, use a filtering function, that is, when the component difficulty is ≤ difficulty coefficient × 1.2, to ensure gradual adjustment.

[0085] The specific steps for maintaining an initial difficulty coefficient for each player based on their profile are as follows: obtain the player profile (e.g., reasoning ability of 80) from the multimodal environment perception module, calculate the initial difficulty coefficient = (average profile score / 100) × maximum difficulty threshold. For example, if the average profile score is 70, the initial difficulty coefficient is calculated as 0.7 and stored in the data center as a player attribute. Initially, the average profile score is maintained using a weighted average formula: average profile score = (reasoning ability × 0.4 + observation ability × 0.3 + ...), and is locked at the start of the game.

[0086] The strategic layer's responsibility is to formulate narrative and operational goals from a holistic perspective, specifically including:

[0087] The strategy layer is a high-level intelligent agent that doesn't concern itself with specific task details, such as "finding the stone tablet" or "connecting with the dragons." Its mission is to formulate narrative and scheduling objectives from a global perspective for the current stage (the current stage refers to a plot chapter or time period in a murder mystery game, for example, a segment from "opening investigation" to "climax confrontation," such as a total game time of 60 minutes, with the first 20 minutes being stage 1). Narrative objectives refer to the macro-direction of plot progression, such as "driving faction conflict or revealing clues," for example, ensuring the "detective faction" collects 3 / 5 of the clues within stage 2. Scheduling objectives refer to macro-strategies for resource allocation, such as "optimizing player distribution to avoid congestion," for example, diverting players from high-density areas to low-density areas. The method for formulating these objectives involves running Q-Learning through the MARL strategy layer model, simulating game theory (Nash equilibrium) after inputting the macro-state, and selecting the instruction that maximizes the reward (e.g., prioritizing narrative objectives if plot progress is <50%).

[0088] The input (state space) is defined to focus on macro-level states, including overall plot progress, heatmaps, weather conditions, and remaining game time. The output (action space) is defined as high-level instructions, which are high-level guiding commands. Examples include: advancing the plot: guiding players in the 'detective' faction to discover crucial evidence left by the 'murderer' in the west district within 15 minutes; balancing the load: diverting at least 30% of players from the central plaza (congestion > 85%) to the east garden (congestion < 40%); and adapting to the environment: moving all 'outdoor writing' mission objectives to the nearest shelter due to the start of rainfall. A custom decision frequency is defined, typically low, such as once every 5-10 minutes.

[0089] The process of generating advanced commands based on overall story progress, heatmaps, weather conditions, and remaining game time includes:

[0090] Step A1: Obtain data from the multimodal environment perception module (e.g., plot progress: 40%, heat map: central congestion 0.8, weather: rainfall 0.6, remaining time: 30min) to form a state vector;

[0091] Step A2: MARL runs a Q-Learning variant to simulate a multi-round game to evaluate rewards (e.g., if the remaining time is <20 minutes and congestion is high, prioritize scheduling targets);

[0092] Step A3: Select the maximum reward instruction (e.g., if the rainfall is >0.5, generate the "Adapt to Environment" instruction);

[0093] Step A4: Store the basis for decision-making.

[0094] The tactical layer's responsibility is to break down high-level instructions into specific tasks, including:

[0095] Receive high-level instructions from the strategic layer and break them down into specific, executable task elements. Each tactical agent is responsible for a player group or an area.

[0096] Decomposition and translation refer to breaking down high-level instructions into sub-goals and mapping them to components. For example, "driving the plot" is decomposed into "clue search sub-task", which is translated as {action: search, location: West District}.

[0097] The inputs are defined as strategic instructions and micro-states. Strategic instructions are high-level guidance, such as "guide the detective team to find evidence." Micro-states include each player's task progress and precise location, as well as the updated player profile. The output is defined as a personalized task assembled for a single player from the task component library.

[0098] Methods for creating personalized missions for individual players from a mission component library include:

[0099] Step X1: Query the task component library (MongoDB);

[0100] Step X2: Match player profiles (e.g., players with high observation skills select the "Search" task element);

[0101] Step X3: Apply a difficulty level filter (e.g., select medium difficulty with a coefficient of 0.7);

[0102] Step X4: Optimize rewards (e.g., maximize experience rewards);

[0103] Step X5: Generate a combination (e.g., {Action: Search, Location: Rockery, Challenge: None}).

[0104] Customizable decision-making frequency, higher than that of the strategic layer, facilitates the generation of instructions for players who need new tasks in real time.

[0105] By introducing hierarchical decision-making to reduce computational complexity, the massive decision-making problem is broken down into two more manageable sub-problems, avoiding the curse of dimensionality. Simultaneously, it safeguards the macro-narrative; the strategic layer ensures the main storyline isn't derailed by a massive number of micro-tasks, maintaining its correct direction. Furthermore, it achieves precise micro-matching, allowing the tactical layer to utilize player profiles more precisely, enabling highly personalized task allocation.

[0106] Task elements refer to atomic-level task components, such as {Action: Search}, {Location: Ancient Pavilion}, and {Challenge: Puzzle}, which can be combined to form a complete task. The task element library refers to a pre-designed database of atomic-level task elements, for example, using MongoDB to store more than 1000 elements (such as {Target Location: Ancient Pavilion} and {Action: Search}), supporting querying and combination, facilitating dynamic assembly at the tactical level.

[0107] The rules for the positive and negative counters are as follows: if the player succeeds once, the positive counter increments by 1. If the player succeeds consecutively, the positive counter increments by 1 until the player fails, at which point the positive counter stops counting and its value is reset to zero. At the same time, the negative counter increments by 1. If the player fails consecutively, the negative counter increments by 1 until the player succeeds, at which point the negative counter stops counting and its value is reset to zero.

[0108] The Story Security Guardian mechanism is a lightweight, rule-based filter module (implemented using a scripting language such as Python) embedded after the tactical layer output as the final verification checkpoint, including logical consistency, story lock, and spatial rationality checks; ensuring that the combinations generated by MARL are free of vulnerabilities.

[0109] Logical consistency refers to the absence of contradictions between task rewards or clues and the player's existing information. For example, check whether the clues rewarded in the task are duplicates or conflict with clues that the player already possesses. Key fragment B should not be rewarded to a player who already has it. Database queries and comparisons can be used.

[0110] Story locks refer to quests depending on the unlock status of story nodes. For example, to check if a quest depends on a story node that has not yet been unlocked, such as before the king was killed, no player could receive the coronation quest. The story node status can be verified through the global story state machine.

[0111] Spatial rationality testing refers to the feasibility of a task in terms of physical space and time. For example, to check the possibility that a task requires a player to move from the northernmost point of a scenic area to the southernmost point within 5 minutes, LBS distance calculation and average movement speed estimation can be used. If the distance is greater than 500m and the time is less than 5min, it is rejected.

[0112] Rule validation executes the if-then rule chain sequentially. For example, it first checks logical consistency. If that passes, it checks plot locks. If that passes, it checks spatial rationality. If any one of these fails, the validation fails. An automated script compares the task combination with a predefined rule base (stored in a JSON file). The entire process takes less than 1 second, and the validation results are logged.

[0113] Task generation module: Receives the output of the dynamic decision module, assembles and renders it to form the final task, and sends it to the player's terminal. The player receives the task and interacts with it. The task completion status is automatically verified using computer vision (CV) and sensor fusion, and feedback is given to the multimodal environment perception module to form a loop.

[0114] In computer vision (CV) and sensor fusion, CV uses the YOLO model to recognize uploaded photos (e.g., leaf shape matching >90%); sensor fusion combines LBS to confirm location and CV to verify objects. If verification fails, a prompt or help button is provided and recorded as feedback data.

[0115] Methods for assembling and rendering the final task include: using dynamic narrative generation techniques to fill task elements into a template and generate multimodal output.

[0116] Dynamic narrative generation technology, which fills task elements into templates, refers to using a GPT-like model (such as fine-tunedChatGPT) to populate templates, ensuring personalized narratives (e.g., adding time limits to high-ability player tasks). Template example: {Character}, in {location} {action} {object}, obtain {reward} through {challenge}.

[0117] Multimodal output includes voice using TTS tools such as Google Text-to-Speech, with customized voice timbres (e.g., a deep male voice for a traditional Chinese NPC); AR integration with Unity AR Foundation, allowing for the overlay of arrows / highlights in the app (e.g., glowing targets). It supports multilingual output to cater to international tourists.

[0118] The player's terminal app is developed using React Native and supports push notifications and real-time chat (enhanced social interaction). AR devices are compatible with smartphones / glasses (such as Apple Vision Pro).

[0119] Example 2

[0120] Please see Figure 2 The present invention also provides an electronic device. Figure 2 This is a structural diagram of an electronic device 20 according to an exemplary embodiment. The content of the diagram should not be construed as limiting the scope of the invention. Specifically, the electronic device 20 may include: at least one processor 21, at least one memory 22, a power supply 23, a communication interface 24, an input / output interface 25, and a communication bus 26. The memory 22 stores a computer program, which is loaded and executed by the processor 21 to implement the relevant steps in the high-reliability memory chip data read / write verification method disclosed in any of the foregoing embodiments. Furthermore, the electronic device 20 in this embodiment may specifically be an electronic computer.

[0121] In this embodiment, the power supply 23 is used to provide operating voltage for each hardware device on the electronic device 20; the communication interface 24 can create a data transmission channel between the electronic device 20 and external devices, and the communication protocol it follows can be any communication protocol applicable to the technical solution of this invention, and is not specifically limited here; the input / output interface 25 is used to acquire external input data or output data to the outside world, and its specific interface type can be selected according to specific application needs, and is not specifically limited here.

[0122] In addition, the memory 22, as a carrier for resource storage, can be a read-only memory, random access memory, disk or optical disk, etc. The resources stored thereon can include operating system 221, computer program 222, etc., and the storage method can be temporary storage or permanent storage.

[0123] The operating system 221 is used to manage and control the various hardware devices on the electronic device 20 and the computer program 222, which may be Windows Server, Netware, Unix, Linux, etc. In addition to including a computer program capable of performing the high-reliability memory chip data read / write verification method disclosed in any of the foregoing embodiments, the computer program 222 may further include computer programs capable of performing other specific tasks.

[0124] Furthermore, the present invention also provides a computer-readable storage medium for storing a computer program; wherein, when the computer program is executed by a processor, it implements the aforementioned high-reliability memory chip data read / write verification method. Specific steps of this method can be found in the corresponding content disclosed in the foregoing embodiments, and will not be repeated here.

[0125] The above description is merely a preferred embodiment of the present invention, and the scope of protection of the present invention is not limited to the above embodiments. All technical solutions falling within the scope of the present invention's concept are within the scope of protection of the present invention. It should be noted that for users of ordinary technical skills, any improvements and modifications made without departing from the principles of the present invention should also be considered within the scope of protection of the present invention.

Claims

1. An AI-powered intelligent agent task generation system is applied to a scenic area script-based murder mystery game system, characterized by: include: Multimodal environment perception module: It acquires the type of murder mystery game and generates an evaluation method based on the type. It assesses the player's relevant abilities under this type, generates a player profile, and perceives the current environmental state to generate an environmental profile. Dynamic Decision Module: Assigns initial tasks to each player based on player profiles and environment profiles, uses the MARL model for multi-agent simulation, sets narrative and scheduling goals at the strategic level, decomposes them into specific tasks at the tactical level, generates personalized tasks by combining reward functions and story security guard mechanisms, and dynamically adjusts the player's difficulty coefficient. Task generation module: Receives the output of the dynamic decision module, assembles and renders it to form the final task, and sends it to the player's terminal. The player receives the task and interacts with it. The task completion status is automatically verified and fed back to the multimodal environment perception module, forming a loop.

2. The AI ​​intelligent agent generation task system according to claim 1 is applied to a scenic area script-based murder mystery game system, characterized in that, The method for generating player profiles by evaluating players' relevant abilities within a specific type based on the evaluation criteria includes: The evaluation method includes mapping evaluation dimensions and test adjustment direction; mapping evaluation dimensions include the player evaluation dimensions for each type and the corresponding evaluation rules for each dimension; test adjustment direction includes the adjustment dimensions for the basic mini-games for each type and the adjustment rules for each dimension. Design a basic mini-game, adjust it based on testing, generate an updated mini-game, and push it to the mobile device. During the player's gameplay, behavioral data is acquired, and then processed according to the corresponding mapping evaluation dimensions to obtain an evaluation report; A unique code is generated for each player, and the evaluation report is linked with the code to form a player profile.

3. The AI ​​intelligent agent generation task system according to claim 2 is applied to a scenic area script-based murder mystery system, characterized in that, The method for perceiving the current environmental state and generating an environmental profile includes: Obtain the visitor flow, distribution density, and weather conditions of the scenic area, and standardize them into numerical vectors; Using GIS tools, density is mapped onto a scenic area map, with color gradients representing risks, thus creating a heat map; Combine the density vector and weather vector into a multidimensional vector; add metadata to the multidimensional vector, including timestamps and risk scores. Heatmaps and multidimensional vectors are saved to the data center and output via API interface.

4. The AI ​​intelligent agent generation task system according to claim 3 is applied to a scenic area script-based murder mystery system, characterized in that, The method for multi-agent simulation using the MARL model includes: Define the state space, action space, and reward function. Select a combination of task elements from the action space that maximizes the total reward to generate a personalized task and assign it to the most suitable player. The state space uses a multi-layered structure, including a strategic layer and a tactical layer. The strategic layer is responsible for setting narrative and scheduling goals from a global perspective, while the tactical layer is responsible for breaking down high-level instructions into specific tasks. The action space is defined as selecting actions from the task component library; The total reward of the reward function is defined as the weighted sum of experience reward, efficiency reward, story reward, and adaptive reward. Among them, experience reward is a negative matching score, which is used to quantify the fit between the task and the player's ability; efficiency reward is a negative congestion penalty, which is used to quantify the regional load within the scenic area; story reward is a positive logical consistency score, which is used to quantify the coherence of the task sequence; and adaptive reward is a positive environmental adjustment bonus, which is used to quantify the response to variables. Based on player profiles, an initial difficulty coefficient is maintained for each player. When the tactical layer agent generates a task, a pre-designed plot security guard mechanism is used to perform rule verification on the task output by the tactical layer. If the verification passes, the task is sent to the task generation module; otherwise, it is regenerated. The system continuously monitors the player's task completion status. If the player succeeds, the positive counter increments by 1; if the player fails, the negative counter increments by 1. The system adjusts the individual's difficulty coefficient based on the counter value and selects components based on the difficulty coefficient in the next task.

5. The AI ​​intelligent agent generation task system according to claim 4 is applied to a scenic area script-based murder mystery system, characterized in that, The strategic layer's responsibility is to formulate narrative and scheduling goals from a holistic perspective, specifically including: The strategic layer is used to formulate the narrative goals and scheduling goals for the current stage; the narrative goals refer to the macro direction of the plot's progression, and the scheduling goals refer to the macro strategy for resource allocation; The inputs are defined as overall story progress, heatmap, weather conditions, and remaining game time. The output is defined as high-level instructions, which are high-level guiding commands, and the frequency of custom decisions is allowed.

6. The AI ​​intelligent agent generation task system according to claim 5 is applied to a scenic area script-based murder mystery system, characterized in that, The responsibility of the tactical layer is to break down high-level instructions into specific tasks, including: Receive high-level instructions from the strategic layer and break them down into specific, executable task elements. Each tactical agent is responsible for a player group or an area. The inputs are defined as strategic instructions and micro-states. Strategic instructions are high-level guidance, and micro-states include each player's task progress and precise location, as well as the updated player profile. The output is defined as a personalized task assembled for a single player from the task component library. Customizable decision-making frequency, higher than the decision-making frequency at the strategic level.

7. The AI ​​intelligent agent generation task system according to claim 6 is applied to a scenic area script-based murder mystery system, characterized in that, The task component library includes: Task elements refer to atomic-level task features, and the task element library refers to a pre-designed database of atomic-level task features.

8. The AI ​​intelligent agent generation task system according to claim 7 is applied to a scenic area script-based murder mystery system, characterized in that, The rules for the positive and negative counters are as follows: if the player succeeds once, the positive counter increments by 1; if the player succeeds consecutively, the positive counter increments by 1 continuously until the player fails, at which point the positive counter stops counting and its value is reset to zero, and the negative counter increments by 1; if the player fails consecutively, the negative counter increments by 1 continuously until the player succeeds, at which point the negative counter stops counting and its value is reset to zero.

9. The AI ​​intelligent agent generation task system according to claim 8 is applied to a scenic area script-based murder mystery system, characterized in that, The plot security guard mechanism is embedded after the tactical layer output, serving as the final verification checkpoint, including logical consistency, plot lock, and spatial rationality detection; Logical consistency refers to the absence of contradictions between task rewards or clues and the player's existing information; plot lock refers to the unlocking status of a task depending on a plot node; spatial rationality check refers to the feasibility of a task in terms of physical space and time. If the rule validation passes, the if-then rule chain is executed sequentially.

10. The AI ​​intelligent agent generation task system according to claim 9 is applied to a scenic area script-based murder mystery system, characterized in that, The method for forming the final task through assembly and rendering includes: using dynamic narrative generation technology to fill task elements into a template and generate multimodal output.

Images