KR101692539B1 - NPC decision making modelling method in dynamic terrain analysis by prospect theory - Google Patents

Abstract

The present invention applies Prospect Theory in behavioral economics to game artificial intelligence techniques to make decisions and actions of Non Player Character (NPC) NPC decision modeling method in dynamic terrain analysis based on perspective theory is presented. The NPC decision modeling method in the prospect theory based dynamic terrain analysis according to the present invention is based on the prospect theory based decision algorithm which performs the dynamic terrain analysis using the prospect theory to select the moving target point of the NPC, It is possible to design a game with a three-dimensional and intelligent nature, and it is possible to increase the enthusiastic interest and interest of the game user in the game, And a technical feature that can be effectively applied to a sports game to improve the quality of the game.

Description

{NPC decision making modeling method in dynamic terrain analysis based on prospect theory}

The present invention applies Prospect Theory in behavioral economics to game artificial intelligence techniques to make decisions and actions of Non Player Character (NPC) Based decision modeling method for dynamic geomorphic analysis based on the prospect theory to be modeled. More specifically, the present invention relates to a dynamic geomorphological analysis using a prospect theory, Each NPC can make different decisions through the algorithm, so that various NPC behaviors are generated. Accordingly, it is possible to design a three-dimensional and intelligent game, and it is possible to increase the immersion and interest of the game user , Especially for real-time strategy games and sports games, Outlook relates to the NPC decision on the modeling method of theory-based analysis of dynamic terrain.

Due to the remarkable development of the video game industry in recent years, the technology used in the game has also been diversified and completed to a very high level. In the past, game development technology was mainly focused on computer graphics, which produces realistic images in real time. In recent years, game artificial intelligence techniques for a rich and rich experience that can be provided by games have been revolutionizing the game . As a patented technology related thereto, Korean Registered Patent Publication No. 10-1231798 entitled " Game Difficulty Adjusting Apparatus and Method "

On the other hand, the game artificial intelligence field is mainly focused on the representation of behaviors of NPCs present in the game, and research and development is actively conducted to model NPC decision making in various environments. Representative NPC decision modeling techniques that have been recognized as being industrially useful include Finite State Machine, Hierarchical Finite State Machine, Behavior Tree, Utility- based System, Goal Oriented Action Planner (GOAP), and Hierarchical Task Network (HTNs).

Based on Utility Theory, which is widely used in fields such as game theory and economics, the utility-based system imposes a utility-based score on all possible actions in a given situation, As shown in FIG. Bernoulli, who advocated utility theory, argued that decision makers tend to avoid risk in situations where the marginal utility of wealth diminishes. His theory, however, can not explain the decision to risk gambling in a loss situation. To overcome these shortcomings, Carnarvon and Tverski discovered that decision makers are using different points of reference in the context of profit and loss, and have advocated the view theory as a framework. Carnerman 's prospect theory outlines a rational human hypothesis in traditional economics, and presents a view of how human beings choose and act and, as a result, what social phenomena they represent.

On the other hand, in a real-time strategy game or a sports game, an agent such as an NPC is approached strategically tactically at a specific position in an activated frame area of a game image constructed according to virtual external environment information. For example, it is possible to induce an agent's behavior by providing the agent with position point information that is determined as a hidden point that is not exposed to an attack in the battlefield, and a sniper point where enemy attack is easy. For this purpose, a planner can manually select a specific location point manually in the game production process. This is because, when the game environment is vast, it has difficulty in manual work and difficulty in coping with a dynamically changing environment. Therefore, dynamic topography analysis in Killzone FPS game released by PS Vita {A. As shown in FIG. 1, which illustrates the dynamic terrain analysis algorithm of Beij, "Killzone's AI: Dynamic Procedural Tactics ", Game Developer Conference Europe, 2005.}, a technology for automatically selecting a specific location point as artificial intelligence And the development of an algorithm design for increasing the operation efficiency of the dynamic terrain analysis has been variously demanded.

(Patent Document 1) Korean Registered Patent Publication No. 10-1231798 "Game Difficulty Adjusting Apparatus and Method"

The present invention is based on the three cognitive characteristics of reference point dependency, diminishing sensitivity, and loss avoidance (see, for example, In order to implement the NPC decision model that reflects the loss aversion, the decision theory is applied to the game artificial intelligence technique so that the decision is made by measuring the utility of the state change The purpose of this paper is to provide a method of NPC decision modeling in a new type of viewpoint based dynamic topography analysis in which NPC decisions and behaviors can be modeled in close proximity to human actual selection patterns and behavior patterns.

Also, according to the present invention, various NPC behaviors are induced by allowing the respective NPCs to make different decisions through a prospective theory-based decision algorithm that performs dynamic topography analysis by applying the prospect theory to select a moving target point of the NPC Thus, it is possible to design a stereoscopic and intelligent game, and it is possible to increase the immersion and interest of the game users, and particularly to a new type that can be applied to a real-time strategy game or a sports game to improve the quality of the game Based decision modeling method for dynamic terrain analysis.

According to an aspect of the present invention, there is provided a computer game device for designing a game, the game device comprising: virtual environment information of a game image; non-player character (NPC) Is set by the game designer and input by the input device of the computer device while a plurality of position points at which the NPC is located within the active frame area of the game image are set by the game designer, A position point setting step of inputting the position point; A decision-causing element for inducing an NPC movement to any one position point (a moving target point) selected from a plurality of position points is set by a game designer and input to the computer device by an input device of the computer device, A consideration factor definition step of defining a decision inducing factor as a consideration factor i; A decision-making-inducing-strength calculating algorithm for calculating a decision-making-inducing strength at each position point for each decision-inducing element is set by a game designer, and the decision- An algorithm setting step; And calculating the decision-induced strength measurements x _i decisions caused strength measurement value calculated by quantification of through the decision-induced intensity calculation algorithm stored in the computer system for each location point to the considerations i of the NPC step; The value a function of the utility value v (x _i) on the measured values x _i of the considered element i has been set by the game designer for each considerations i doedoe by the input device of the computer device input to the computer device, the value Function is a value function set step that is used in the view theory; Calculating utility value step utility value v (x _i) of points at each position over the function value stored in the computer device is calculated for each element taken into account; An occurrence probability calculating algorithm for calculating an occurrence probability P _i for the consideration factor i is set by a game designer and input to the computer device by an input device of the computer device; A sensitivity sensitivity weighting reflection algorithm for assigning a weight to the occurrence probability P _i for reflection of the sensitivity bodily sensation is set by the game designer and input to the computer device by the input device of the computer device, ; On through the occurrence probability calculation algorithm stored in the computer equipment which the occurrence probability P _i of the considered element i of the NPC calculated for each location point and then, the probability of occurrence over a sensitivity element sensitivity weight reflect the algorithm stored in the computer unit P _i An occurrence probability correction value calculation step of calculating a weighted occurrence probability correction value w (P _i ); An evaluation value for each location point is calculated with respect to the entire decision-making inducing element by the evaluation function defined by the equation (1) stored in the computer device; An NPC decision algorithm that evaluates the evaluation value at each location point calculated by the computer device in the set analysis pattern and selects one location point is set by the game designer and input to the computer device by the input device of the computer device And an NPC decision step in which a location point selected through the NPC decision algorithm is set as a moving target point of the NPC. The NPC decision modeling method in the dynamic terrain analysis based on the view theory is provided.

[Equation 1]

In the NPC decision modeling method in the dynamic terrain analysis based on the prospective theory-based dynamic terrestrial model according to the present invention, the step of setting the sensitivity body sensitivity weighting reflection algorithm may include the steps of selecting one of the sensitivity body sensitivity weight function table and the sensitivity body sensitivity weight heuristic setting algorithm Wherein the generation probability correction value calculating step is configured to calculate the probability of occurrence correction value using a sensitivity characteristic weight function function table and a sensitivity sensitivity weightedness heuristic setting algorithm by using the input device of the computer device, So that the value w (P _i ) is calculated.

In the NPC decision modeling method according to the present invention, the reference value is set for each consideration factor i so that the measurement value x _{i, which} is larger than the reference value, belongs to the profit area , And the measured value x _i smaller than the reference value can be set to a value function to belong to the damage area.

In the NPC decision modeling method in the prospect theory-based dynamic terrain analysis according to the present invention, the utility value calculation step may be such that the reference value of the value function is changed by adjusting the magnitude value of? X _i in the expression (1) can do.

In the NPC decision modeling method based on the prospect theory-based dynamic terrain analysis according to the present invention, the NPC decision algorithm of the NPC decision step is an algorithm for setting the location point having the maximum evaluation value as the moving target point of the NPC Or one of the one or more position points having the evaluation value equal to or higher than the set value may be set to the moving target point at random.

In the NPC decision modeling method based on the prospect theory-based dynamic terrain analysis according to the present invention, the consideration factor defining step includes consideration of proximity, LOF (line of fire), safety domain, and attack domain Element i, and the step of calculating the decision-making strength measurement value calculates the measured value x _i for the proximity, which is the consideration factor i, as the distance value between the NPC and the position point, and calculates the measured value x _i for the LOF at the position point The presence or absence of an obstacle between the NPCs is determined through a visual inspection and is calculated as a measurement value of the possibility of shooting and the measured value x _i for the safety domain is checked only when there is an obstacle between the NPCs at the position point It is calculated by applying a set value of the measured values x _i have a damage value, the attack succeeds and the intrinsic value of the vector towards the center friendly front-facing direction vector for attacking the castle area It can be calculated as.

According to the NPC decision modeling method in the dynamic terrain-based dynamic terrain analysis according to the present invention, the reference point dependency, the diminishing sensitivity, By applying the prospect theory to the game artificial intelligence technique so that the NPC decision model reflecting the loss aversion is implemented, the decision is made by measuring the utility of the state change rather than the effect of the decision result The decision making and behavior of the NPC are modeled close to the actual selection pattern and the behavior pattern of the human being.

In addition, according to the NPC decision modeling method in the prospect theory-based dynamic terrain analysis according to the present invention, the dynamic terrain analysis using the prospect theory is performed to determine the moving target point of the NPC, The NPCs of the NPCs can make different decisions, thereby causing various NPC behaviors. Accordingly, it is possible to design a stereoscopic and intelligent game, and it has the effect of increasing the immersion and interest of the game users. In particular, the present invention can be effectively applied to a real-time strategy game or a sports game to improve the quality of the game.

Figure 1 is an example of dynamic topography analysis in an FPS game Killzone released at PS Vita;
FIG. 2 is a block diagram illustrating a NPC decision modeling method in a prospective theory-based dynamic topography analysis according to an embodiment of the present invention; FIG.
FIG. 3 is a view for showing a screen on which a location point is set through a location point setting step of the NPC decision modeling method in the prospect theory-based dynamic terrain analysis according to the embodiment of the present invention;
FIGS. 4 and 5 are diagrams illustrating a value function used in the NPC decision modeling method in the dynamic terrain analysis based on the prospect theory, according to an embodiment of the present invention.

Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings 2 to 5. On the other hand, in the drawings and the detailed description, there are a variety of fields from general game artificial intelligence techniques, Prospect Theory, Non Player Character (NPC), NPC decision modeling method, The description and the operation of the structure and operation that the workers of the present invention can easily understand are omitted or simplified. In the drawings and specification, there are shown in the drawings and will not be described in detail, and only the technical features related to the present invention are shown or described only briefly. Respectively.

The NPC decision modeling method in the dynamic terrain analysis based on the prospect theory based on the embodiment of the present invention is a dynamic terrain analysis technique for analyzing the space to determine a point where strategic tactical approach is required in real time strategy game or sports game As shown in FIG. 2, the estimation method is applied to the location point setting step, the consideration factor defining step, the decision inducing intensity calculating algorithm setting step, the decision inducing intensity measurement value calculating step, the value function setting step, A probability calculation algorithm setting step, a sensitivity sensitivity weighting weighting algorithm setting step, an occurrence probability correction value calculating step, a location point evaluation value calculating step, and an NPC decision making step.

In the location point setting step, the virtual external environment information of the game image, the non-player character (NPC) and the internal variable of the NPC are set in the computer device for game design by the game designer, And a plurality of position points where the NPCs are positioned within the active frame region of the game image as shown in FIG. 3 are set by the game designer and input by the input device of the computer apparatus. Here, the virtual external environment information of the game image is information perceived through the visual system, and is information related to the image environment that is virtually displayed in the frame of the game image. Such virtual external environment information may include a terrain, a building or an obstacle disposed in a frame of the game image, as shown in FIG. And the internal variable of NPC is the characteristic value of each NPC that can influence the decision of the NPC, and the internal variable of such NPC can be set by the game designer. The internal variable of the NPC can be set differently for each NPC, and the internal variable set for the specific NPC can be changed as a function of time or space.

The consideration element defining step is a step in which a decision-making inducing element for inducing the movement of the NPC is set by the game designer and input to the computer device by the input device of the computer device, and the decision-inducing element is defined as the consideration factor i. Here, the decision-making inducing element is an element that induces the NPC movement to any one of the plurality of position points set in the above-mentioned position point setting step (the moving target point). In the design process of game such as real-time strategy game and sports game, consideration factor applied to NPC decision-making modeling in dynamic topography analysis is the proximity to the NPC, Considering the LOF (Line Of Fire) for the target location point, the safety domain for assessing the security level for the side threat components, and the attack domain for evaluating the degree of attack against the attack site location .

The decision-making strength calculation algorithm is set by a game designer so as to calculate a decision-making strength calculation algorithm for calculating the decision-making-inducing strength at each position point for each decision-inducing element, Respectively. Each decision-inducing-intensity calculation algorithm for each decision-inducing element is designed by the game designer according to the characteristics of the game or the planning intention.

The step of calculating the decision-making strength measurement value is a step of calculating the quantified decision-making induction intensity for each consideration factor i of the NPC by the decision-making strength calculation algorithm stored in the computer device as the measurement value x _i . The doctor calculated crystal induced intensity measurement step in accordance with an embodiment of the present invention calculates the measured value x _i for the considered element i proximity with NPC between the location point distance value, the measured value for the LOF x _i a location point The presence or absence of an obstacle between the NPCs in the NPCs is determined through a visual inspection and is calculated as a measurement value for the possibility of shooting and the measured value x _i for the safety domain is visually checked And the measured value x _i for the attack area property is calculated by applying a set value to the motion vector given to the corresponding NPC and the inner numerical value of the motion vector given to the NPC or the player character corresponding to the enemy among the other NPC and the player character It is calculated by multiplying the damage value (set by the game designer) when the attack of the NPC is successful.

Value function setting step is the utility value v value function of (x _i) on the measured values x _i of the considered element i has been set by the game designer for each considerations i input to the computer apparatus by the input device of the computer apparatus . In this case, the value function is used in the view theory and has a form as shown in FIG. 4. The reference value is set for each consideration factor i so that the measurement value x _i larger than the reference value belongs to the profit area and the measurement value x _i Is set to a value function that belongs to the loss area. In particular, the value function according to an embodiment of the present invention can be obtained by the curve fitting method of Alexander {B. This is set by using the Bucket Sampling method, which has a piecewise linear function form, which is set by using the above-described method (e.g., Alexander, " The Beauty of Response Curves ", AI Game Programming Wisdom, Charles River Media,

The utility value calculation step is a step in which the utility value v (x _i ) at each location point is calculated for each consideration factor through the value function stored in the computer device. The utility value v (x _i ) is computed by numerical solving the value function approximation through linear interpolation.

The occurrence probability calculation algorithm setting step is a step for calculating the occurrence probability calculation algorithm for calculating the occurrence probability P _i for each consideration factor i by the game designer for each consideration factor i and inputted to the computer device by the input device of the computer device . Such an occurrence probability calculation algorithm is designed in accordance with the characteristics of the game and the characteristics of the consideration factor i.

In the step of setting the sensitivity body sensitivity weighting reflection algorithm, a sensitivity body sensitivity weighting reflection algorithm for weighting the occurrence probability P _i for reflection of the sensitivity bodily sensation is set by the game designer and input to the computer device by the input device of the computer device . In the step of setting the sensitivity sensitive emotion weighting reflection algorithm, the sensory emotion weighting function table as shown in the following Table 1 proposed by Carnerman who has established the view theory is set in advance by the game designer, Or the sensitivity sensitive weighted heuristic setting algorithm is preset by the game designer and input to the computer device by the input device of the computer device, and then the sensitivity sensitive weighting algorithm is added to the sensitivity device Sensitivity weighting function table or sensitivity sensitivity weighting heuristic setting algorithm is applied.

probability (%) One 2 5 10 20 50 80 90 95 98 99 100 weight 5.5 8.1 13.2 18.6 26.1 42.1 60.1 71.2 79.3 87.1 91.2 100

The occurrence probability correction value calculation step calculates the occurrence probability P _i for the consideration element i of the NPC at each position point through the occurrence probability calculation algorithm stored in the computer device and then generates the occurrence probability P _i using the sensitivity body sensitivity weight reflection algorithm stored in the computer device And the occurrence probability correction value w (P _i ) to which the weight P _i is given is calculated. The generation probability correction value calculation step calculates the correction value w (P _i ) through a sensitivity body sensitivity weight function table or a sensitivity body sensitivity weighting heuristic setting algorithm applied to the sensitivity body sensitivity weight reflection algorithm.

In the evaluation value calculation step according to position point, the evaluation function defined by the equation (1) is stored in the computer device, and the evaluation value for each position point is calculated for the entire decision-making inducing element through the evaluation function.

Here, the reference value of the value function can be changed by adjusting the magnitude value of? X _i in [Equation 1]. When Δx _i is set to a positive value, as shown in FIG. 5A, the reference value at the value function (Δx _i = 0) set in the utility value calculation step moves to the left by Δx _i , Some areas are changed to profit areas, which increases risk-taking tendency in NPC decisions. In contrast, when Δx _i is set to a negative value, the reference value at the value function (Δx _i = 0) set in the utility value calculation step is shifted to the right by -Δx _i as shown in FIG. 5 (b) Some areas belonging to the area are changed into damage areas, and the tendency of seeking profit in decision making of the NPC increases.

Utility value calculation step has a utility value v value function of (x _i) on the measured values x _i of the considered element i has been set by the game designer for each considerations i is input to the computer apparatus by the input device of the computer apparatus , And the utility value v (x _i ) at each location point is calculated for each consideration factor through the value function stored in the computer device. 2, a reference value is set for each consideration factor i so that a measurement value x _i larger than the reference value belongs to the profit area, and a measurement value x _i smaller than the reference value is used Is set to a value function that belongs to the loss area.

The NPC decision step is a step in which an NPC decision algorithm, which evaluates the evaluation value at each location point calculated by the computer device in the set analysis pattern and selects one location point, is set by the game designer, And the position point selected by the NPC decision algorithm is set to the movement target point of the NPC. In this case, the NPC decision-making algorithm of the NPC decision stage can set the position point having the maximum evaluation value as the movement target point of the NPC. In this case, the decision pattern of the NPC is made in the same pattern, . In response to this, the NPC decision algorithm can set one of the one or more position points having the evaluation value equal to or higher than the set value at random as the movement target point, so that the decisions of the NPC can be diversified and the progress of the game can be complicated. Herein, the NPC decision algorithm in which one of the one or more position points having the evaluation value equal to or higher than the set value is set as the moving target point at random is disclosed in the bucketing technique {K. Dill, " A Game AI Approach to Autonomous Control of Virtual Characters ", Interservice / Industry Training, Simulation and Education Conference, pp4-5, 2011.} can be applied.

The NPC decision modeling method in the perspective topology based dynamic terrain analysis according to the embodiment of the present invention constituted as described above is based on the reference point dependency which is the cognitive characteristic of the human in the practical utility value judgment, (NPC) decision model that reflects diminishing sensitivity and loss aversion is applied to the game artificial intelligence technique, and the effectiveness of the state change is evaluated So that the NPC's decision making and behavior are modeled close to the human's actual selection pattern and behavior pattern. Also, the NPC decision modeling method in the prospect theory-based dynamic terrain analysis according to the embodiment of the present invention performs the dynamic terrain analysis using the prospect theory to select the moving target point of the NPC, And various NPC behaviors are generated by making decisions. Accordingly, it is possible to design a stereoscopic and intelligent game, and it has a feature of increasing a user's enthusiasm and interest in a game. Accordingly, the NPC decision modeling method based on the prospect theory-based dynamic terrain analysis according to an embodiment of the present invention can be effectively applied to a real-time strategy game or a sports game to improve the quality of the game.

Although the NPC decision modeling method in the prospect theory-based dynamic terrain analysis according to the embodiment of the present invention as described above has been described with reference to the above description and drawings, it is merely an example of the present invention, It will be understood by those skilled in the art that various changes and modifications may be made without departing from the scope of the present invention.

Claims (6)

The virtual external environment information of the game image, the non-player character (NPC) and the internal variable of the NPC are input to the computer device for game design by the input device of the computer device, A location point setting step in which a plurality of location points where the NPC is located within a frame area are input by the input device of the computer device;
A decision-inducing element for inducing an NPC movement to any one of a plurality of position points (a moving target point) is input to the computer device by an input device of the computer device, , LOF (Line Of Fire), safety domain, and attack domain;
A decision-making strength calculation algorithm for calculating a decision-making-inducing-strength calculation algorithm at each location point for each decision-causing element is input to the computer device by an input device of the computer device;
The quantified decision inducing intensity for the consideration factor i of the NPC is calculated as the measured value xi for each location point through the decision-making induction intensity calculation algorithm stored in the computer device, and the measured value xi Is calculated as the distance value between the NPC and the position point, and the measured value xi for the LOF is determined as the measured value for the possibility of shooting by discriminating the existence of the obstacle between the NPCs at the position point through the sight line inspection, The measured value xi is calculated by giving a set value only when there is an obstacle between NPCs at the position point through a sight line inspection, and the measured value xi for the attack area property is calculated as a vector And calculating a decision-inducing strength measurement value by multiplying the inner value of the attack value with the damage value when the attack is successful;
The value function of the utility value v (xi) for the measured value xi of the consideration factor i is input to the computer device by the input device of the computer device for each consideration factor i, A set of in-value functions;
The utility value v (xi) at each location point is calculated for each consideration factor through the value function stored in the computer device so that the reference value of the value function is changed by adjusting the magnitude value of [Delta] xi in [Equation 1] Calculating a utility value;
An occurrence probability calculation algorithm setting step in which an occurrence probability calculation algorithm for calculating an occurrence probability Pi for the consideration factor i is input to the computer device by an input device of the computer device;
A sensitivity body sensitivity weighting reflection algorithm for giving a weight to the occurrence probability Pi is input to the computer device by the input device of the computer device for reflecting the sensitivity bodily sensation, and the sensitivity body sensitivity weight function table and the sensitivity body sensitivity weighting heuristic setting algorithm A predetermined sensitivity characteristic weight reflecting algorithm setting step in which the selected one of the plurality of sensory weighting weight reflecting algorithm is preset and input to the computer device by the input device of the computer device;
An occurrence probability Pi for the consideration element i of the NPC is calculated for each position point through an occurrence probability calculation algorithm stored in the computer device and then a weight is assigned to the occurrence probability Pi through the sensitivity body sensitivity weight reflection algorithm stored in the computer device Calculating a correction value w (Pi) using the selected sensitivity probability weighting function table and sensitivity sensitivity weighting heuristic setting algorithm by calculating the given occurrence probability correction value w (Pi) ;
Calculating an evaluation value for each location point by an evaluation function defined by Equation 1 stored in the computer device and calculating an evaluation value for each location point with respect to the entire decision-making inducing element through the evaluation function;
An NPC decision algorithm for evaluating an evaluation value at each location point calculated by the computer device with a set analysis pattern and selecting one location point is input to the computer device by an input device of the computer device, And an NPC decision step in which a position point selected by the algorithm is set as a moving target point of the NPC,
Wherein the NPC decision algorithm of the NPC decision step is an algorithm for setting a position point having the maximum evaluation value as a moving target point of the NPC or one of at least one position point having an evaluation value equal to or higher than a set value, Wherein the NPC decision modeling method is based on a viewpoint-based dynamic topography analysis.
[Equation 1]

delete The method according to claim 1,
In the utility value calculation step, a value function is set such that a reference value is set for each consideration factor i, a measurement value x _i larger than the reference value belongs to the profit area, and a measurement value x _i smaller than the reference value belongs to the damage area NPC Decision Modeling Method in Perspective Dynamic Terrain Analysis Based on Perspective Theory. delete delete delete

Images