JP4182892B2 - Behavior simulation system - Google Patents

Description

  The present invention relates to a behavior simulation system that simulates human behavior in a virtual space during a disaster or the like.

Conventionally, various systems for simulating human behavior in a virtual space generated by a computer have been proposed. As this type of system, for example, an action toward the target state from the initial state is obtained by using a knowledge base in which a rule of the form “if (condition) then (action)” is given by giving an initial state and a target state. There is known one provided with a planner that searches a knowledge base for rules for generating a plan that is a series of (see, for example, Patent Document 1). Patent Literature 1 describes how the real-world state changes depending on the action candidate by applying the action candidate issued by the planner to the real-world simulator (ie, virtual space) for the assumed scenario. Is evaluated. That is, in the assumed scenario, a plan that cannot be derived from the rules of the knowledge base is generated by inquiring the real world simulator about the contents not described in the knowledge base.
JP 2003-242434 A (paragraph 0073-0089)

  By the way, although the system described in Patent Document 1 described above can generate a sequence of actions to be taken by a person in a disaster or the like as a plan, the action that is considered to be highly effective for reaching the target state. Since a sequence is generated, it is impossible to simulate how a person behaves in response to environmental conditions that change every moment during a disaster or the like. For example, it cannot be used for the purpose of verifying how people evacuate in response to changes in various environmental conditions during a disaster.

  The present invention has been made in view of the above-mentioned reasons, and its purpose is to simulate human behavior with respect to changes in environmental conditions by using a virtual human that corrects behavior in response to environmental conditions that change every moment. The object is to provide a behavior simulation system that can be used.

The invention of claim 1 is an environment simulation unit that generates an environment of a virtual space in which a virtual human behaves according to an environment scenario given in advance, a position specifying unit that specifies an initial position and a target final position of the virtual human, An action database storing action rules in which a plurality of types of actions that can be performed by a virtual human in the virtual space generated by the environment simulation means and an environment condition that activates each action are stored; a sensor means that detects the environment condition; The type of action is determined by collating the environmental conditions detected by the sensor means with the action database, and the virtual human is determined from the current position of the virtual human, the final position specified by the position specifying means, and the environmental conditions detected by the sensor means. Action decision to determine action vector including direction of action The environment simulation means generates a virtual human that executes the action instructed by the action instruction means and the action instruction means that instructs the virtual human to execute the action according to the action type and action vector determined by the action determination means Output means for displaying on the screen of the display means together with the virtual space, the sensor means perceives the presence or absence of an object that becomes an obstacle when the virtual human heads to the final position, and the virtual human at the final position. At least a function of perceiving the presence or absence of another person who acts together when heading, wherein the behavior determination means includes a vector heading from the current position of the virtual human to the final position, and a vector for avoiding an object detected by the sensor means When the distance to the object is within the first threshold, the closer the distance is, the larger the distance is The vector multiplied by the weighting factor of 1 and the vector detected by the sensor means that approaches the other person are multiplied by the second weighting factor that increases as the distance increases when the distance to the other person is equal to or greater than the second threshold. The action vector is determined by combining the obtained vector, the virtual human performs the action instructed by the action instruction means, and the sensor means detects the environmental condition every unit time until the virtual human reaches the final position. It is characterized by repeating.

According to this configuration, the virtual human performs an action according to the environmental condition detected by the sensor means, and further, the virtual human performs an action according to the environmental condition detected by the sensor means after a unit time. Since it repeats every unit time until the human reaches the final position, the behavior of the virtual human will be reviewed every unit time, even if the surrounding environment of the virtual human changes over time like a disaster The behavior of the virtual human can be corrected according to changes in the surrounding environment, and the virtual human in the virtual space can be made to act autonomously. That is, since the virtual human behaves autonomously so as to respond to changes in the surrounding environment, the human behavior in the real space can be accurately simulated by the virtual human. In addition, the type of action of the virtual human is determined according to the environmental conditions detected by the sensor means, and the action vector including the direction of the action is determined using the current position, final position, and environmental conditions of the virtual human. , If the behavior rule specifies avoiding when there is an obstacle on the route from the current position to the final position, or there are multiple virtual humans in the virtual space For example, if the behavior rules stipulate that you want to follow and act after catching up with other virtual humans, use an action vector that includes the direction of the action to avoid obstacles and follow other people. The behavior of virtual humans is defined by simple information of action types and action vectors. It is possible.
Furthermore, since the avoidance behavior for obstacles and the behavior approaching others are represented by vectors, and these vectors are represented by behavior vectors by multiplying them by weighting factors, the avoidance behavior for obstacles and the behavior approaching others are Can be reflected in the action vector. In other words, the closer to the obstacle, the greater the weighting factor for the vector that avoids the obstacle, so the direction of the action vector is set to avoid the obstacle, while the distance from the other person is closer to the other person. Since the weight coefficient for the vector becomes large, the action vector is set to approach another person in a short time. As described above, it is possible to realize not only the goal of reaching the final position but also the obstacle avoidance action and the action of approaching another person only by using the action vector representing the moving direction of the virtual human. In addition, a threshold is set for the distance to the obstacle and the distance to the other person, and when the distance to the obstacle is within the first threshold, an action target of avoiding the obstacle is started, and the distance to the other person is Since the action target of approaching another person is activated when it is equal to or greater than the second threshold value, the distance from the obstacle is not affected by the obstacle when the distance to the obstacle is equal to or greater than the first threshold value, and the distance from the other person is equal to the second threshold value. If it is within the range, no further approach is taken, and appropriate action rules are adopted according to the distance to the obstacles and others.

  According to a second aspect of the present invention, in the first aspect of the invention, the virtual space generated by the environment simulation means is three-dimensional, and mapping means for mapping the virtual space to a two-dimensional virtual plane is added, and the action determination The means is characterized in that an action vector is determined in a virtual plane obtained by the mapping means.

  According to this configuration, the action vector is determined in a two-dimensional virtual plane, and the movement of a person in a room or the like is normally performed in a two-dimensional plane. Compared with the case where the movement vector is determined using three-dimensional information while using sufficient information, the processing load is reduced, and as a result, the time required for determining the action vector is shortened. That is, the processing from the detection of the environmental condition by the sensor means to the determination of the behavior vector can be performed in real time, and the virtual human can be smoothly acted on the screen of the display means.

According to a third aspect of the present invention, in the first or second aspect of the present invention, the sensor means includes at least a function of perceiving a specific environmental condition that defines a behavioral goal of a virtual human, and the behavior determining means includes a sensor When the means detects the presence of the environmental condition, another action is prohibited until an action goal corresponding to the environmental condition is achieved.

  According to this configuration, the action target of the virtual human can be defined when a specific environmental condition occurs, and the action target of reaching the final position can be switched to another action target. For example, in the case of simulating an earthquake in virtual space, when it is a normal goal to reach the evacuation exit as the final position, under certain environmental conditions that there is a desk near the virtual human In order to avoid falling objects, it is possible to start the action of entering under the desk as an action target. In this case, the action of moving to the evacuation exit is prohibited until the action of entering under the desk is completed. By completing the action of entering under the desk, the goal of moving to the evacuation exit becomes unnecessary if the earthquake stops while avoiding falling objects, so if the earthquake is no longer detected by the sensor means, evacuation It is possible to cancel the goal of moving to the mouth. Alternatively, when an environmental scenario is set so that a fire is caused by an earthquake, the action of moving to the evacuation exit by detecting the fire by the sensor means may be activated again. As is apparent from the above-described operation, it is possible to set an intermediate target until the final goal is achieved, and allow the virtual human to act so as to approach the final target after achieving the intermediate target.

In the invention of claim 4, in the invention of claim 1 or claim 2 , the environment simulating means generates an environment at the time of fire in the virtual space and changes with time in the smoke concentration distribution in the virtual space at the time of fire. The sensor means has a function of detecting the smoke density, and the behavior database has a smoke density within a specified range centered on the current position of the virtual human being equal to or higher than a specified threshold value. It stores an action rule that associates an action of moving to a lower smoke concentration with respect to the environmental condition.

  According to this configuration, when simulating a fire in a virtual space, pay attention to the smoke density distribution, and when the smoke density around the virtual human exceeds the specified threshold, evacuate to the lower smoke density. Since the behavior is defined, it is possible to simulate the behavior of avoiding smoke during a fire.

According to a fifth aspect of the present invention, in the first or second aspect of the invention, the environment simulation means has a function of generating an environment at the time of an earthquake in a virtual space, and the sensor means determines a virtual human action target. It has a function of detecting a desk as a specific environmental condition to be defined, and the behavior database stores an action rule that associates an action of entering the desk with respect to the environmental condition of the presence of the desk, The behavior determining means is characterized in that when the sensor means detects the presence of the desk, other actions are prohibited until the behavior goal of entering under the desk is achieved.

  According to this configuration, when an earthquake is simulated in a virtual space and there is a desk in the vicinity of the virtual human, the action is prescribed to enter under the desk to avoid falling objects. The behavior of avoiding falling objects during an earthquake can be simulated. In addition, other actions are prohibited until the action of entering under the desk is completed, so the action of entering under the desk does not conflict with other actions. After completion, other actions are possible, such as moving to an evacuation exit according to the environmental conditions detected by the sensor means.

According to a sixth aspect of the present invention, in the first to fifth aspects of the invention, a virtual space editor that edits a virtual space generated by the environment simulating means, and an environment scenario of a virtual space that is generated by the environment simulating means is edited. An environment scenario editor, a virtual human editor that edits virtual human attributes displayed on the display means in the output means, and an action editor that edits action types and action rules stored in the action database, To do.

  According to this configuration, since the virtual space and the environmental scenario can be edited, for example, a building designed for building design or the like can be used as a virtual space, and the user can edit the virtual space. The environment assumed in the above can be expressed in the virtual space by the user, and information that can be used to improve the design can be obtained by monitoring the behavior of the virtual human in these environments. In addition, since the virtual human attributes and action rules can be edited, the virtual human attributes are given as attributes that induce other virtual humans, or conversely, the induced person is guided by the inducer. Or the like, the evacuation guidance is simulated, and the evacuation behavior can be simulated. In addition, with regard to action rules, the type of action that is activated according to environmental conditions, such as not including the run assuming the elderly, is restricted or enhanced, or the action vector is set according to the environmental scenario. It is possible to simulate under various conditions such as adjusting the method of obtaining the value.

In the invention of claim 7, in the invention of claim 6 , the virtual human editor can arrange a plurality of virtual humans in a virtual space generated by the environment simulating means, and an attribute of another virtual human can be used as an attribute of the virtual human. It is possible to select a guider that induces an action.

  According to this configuration, it is possible to simulate evacuation guidance and the like by setting attributes as a guider to a plurality of virtual humans existing in the virtual space.

According to the configuration of the present invention, the virtual human performs an action according to the environmental condition detected by the sensor unit, and the virtual human performs an action according to the environmental condition detected by the sensor unit after a unit time. Because it repeats every unit time until the virtual human reaches the final position, the behavior of the virtual human will be reviewed every unit time, and the surrounding environment of the virtual human will change over time like a disaster Even in this case, the behavior of the virtual human can be corrected according to changes in the surrounding environment, and the virtual human in the virtual space can be made to act autonomously. As a result, since the virtual human behaves autonomously in response to changes in the surrounding environment, there is an advantage that the human behavior in the real space can be accurately simulated by the virtual human. In addition, the type of action of the virtual human is determined according to the environmental conditions detected by the sensor means, and the action vector including the direction of the action is determined using the current position, final position, and environmental conditions of the virtual human. , If the behavior rule specifies avoiding when there is an obstacle on the route from the current position to the final position, or there are multiple virtual humans in the virtual space For example, if the behavior rules stipulate that you want to follow and act after catching up with other virtual humans, use an action vector that includes the direction of the action to avoid obstacles and follow other people. The behavior of virtual humans is defined by simple information of action types and action vectors. That there is an effect that it is.
Furthermore, since the avoidance behavior for obstacles and the behavior approaching others are represented by vectors, and these vectors are represented by behavior vectors by multiplying them by weighting factors, the avoidance behavior for obstacles and the behavior approaching others are Can be reflected in the action vector. In other words, the closer to the obstacle, the greater the weighting factor for the vector that avoids the obstacle, so the direction of the action vector is set to avoid the obstacle, while the distance from the other person is closer to the other person. Since the weight coefficient for the vector becomes large, the action vector is set to approach another person in a short time. As described above, it is possible to realize not only the goal of reaching the final position but also the obstacle avoidance action and the action of approaching another person only by using the action vector representing the moving direction of the virtual human. In addition, a threshold is set for the distance to the obstacle and the distance to the other person, and when the distance to the obstacle is within the first threshold, an action target of avoiding the obstacle is started, and the distance to the other person is Since the action target of approaching another person is activated when it is equal to or greater than the second threshold value, the distance from the obstacle is not affected by the obstacle when the distance to the obstacle is equal to or greater than the first threshold value, and the distance from the other person is equal to the second threshold value. If it is within the range, no further approach is taken, and appropriate action rules are adopted according to the distance to the obstacles and others.

  The present embodiment is mainly composed of a computer. As shown in FIG. 1, an environment simulation unit 11 for generating a three-dimensional virtual space environment according to a predetermined environment scenario, and an environment simulation unit 11 The sensor means 12 as a sensory organ of a virtual human acting in the virtual space generated by the virtual human, and the type and behavior of the virtual human depending on the environmental conditions detected by the sensor means 12 while the virtual human acts in the virtual space The simulation apparatus 1 includes an action determination unit 13 that determines an action vector including the direction of the action, and an action instruction unit 14 that instructs the virtual human to execute the action according to the type of action determined by the action determination unit 13 and the action vector. It is provided.

  In addition, the image forming apparatus includes an output unit 21 that displays on the screen of the display unit 22 a virtual human that executes an action instructed by the behavior instruction unit 14 provided in the simulation apparatus 1 together with the virtual space generated by the environment simulation unit 11. A device 2 is provided. The output unit 21 has a function of giving a motion for an action instructed by the action instruction unit 14 to a virtual human having an appearance selected by the user. Data relating to the appearance and movement of the virtual human is stored in the motion database 23 provided in the imaging apparatus 2, and the output means 21 stores data necessary for the movement corresponding to the action instructed by the action instruction means 14 in the motion database. By acquiring from 23, the virtual human is displayed on the display means 22. The display unit 22 may be any display unit that can display a moving image on a screen, and a known unit such as a CRT or a liquid crystal display can be used. Moreover, if the display means 22 that enables stereoscopic viewing using parallax can be used, the sense of reality can be enhanced, and a hemispherical concave screen is displayed on the screen that has a wider range than the human field of view. If stereoscopic display is performed using the display means 22, an immersive feeling can be obtained and the presence can be further enhanced.

  The simulation device 1 will be described in more detail. The present embodiment mainly simulates human behavior in an abnormal state such as a fire, an earthquake, a flood, and a traffic jam by a virtual human, and the virtual human detects the environment of the virtual space generated by the environment simulation means 11. Thus, the simulation device 1 realizes the function of the virtual human to determine the behavior autonomously. In such an environment, the purpose of virtual human behavior is to move from the current position to the target position in a short time, or to ensure safety, so as an initial condition for simulating human behavior, The position, the type of the first action and the direction of the first action are given. Usually, the final position that should be reached as a target is set. In other words, the main goal is to move from the initial position to the final position. Such initial conditions and final positions are given by the position specifying means 16 having an input means using a keyboard and a mouse.

  The environment simulation means 11 specifies the location and range of the three-dimensional virtual space, specifies the type and position of the object existing in the virtual space, and further considers the influence of the object on the behavior of the virtual human. Specify attributes. In an abnormal state, the environment often changes with the passage of time, so the environment simulating means 11 also has a function of realizing the time change of the environment according to the environment scenario. The environmental time changes assumed in the environmental scenario include, for example, the temporal change in the smoke density distribution around the virtual human that occurs during a fire, and the position of an obstacle in the virtual human's action path due to furniture falling over during an earthquake. Including time changes that occur. The environmental scenario is appropriately set according to the abnormal state or place to be simulated.

  As described above, the environment of the virtual space that is the range of action of the virtual human may change over time even if the virtual human does not act, and if the virtual human moves in the virtual space, the distance from the object The conditions that affect the behavior of virtual humans change. In this way, the conditions for changing the behavior of the virtual human in the virtual space, regardless of whether the environment is generated by the environmental scenario or the environment generated by the virtual human behavior, are as follows: Call.

  An object included in the virtual space generated by the environment simulation unit 11 is represented by an object name and an attribute attached to the object name. The nodes represented by the object names include composite nodes that are hierarchized in a parent-child relationship and isolated nodes that are not hierarchized. Hierarchization in a composite node means, for example, a case where “kitchen” is a parent node and “door”, “storage shelf”, “sink” is a child node. That is, the parent node uses the name of a space such as a room or place as an object name, and the attribute of the parent node is associated with the object name of the child node. The child node uses the name of an object such as a facility or tool existing in the space represented by the object name of the parent node as the object name. In this way, by layering nodes, it is possible to recognize what kind of objects exist around by simply representing the location of the virtual human by the object name of the parent node, and environmental conditions. Easy to set up.

  The attributes of the child node and the isolated node include at least the position of each object represented by the object name, the occupied area (size), and the direction of the object. The position that is the attribute of the isolated node and the child node represents the position of the representative point of the object represented by the object name. The position has three-dimensional position information, and is represented by, for example, three-dimensional coordinates (X, Y, Z). As will be described later, the virtual human normally does not recognize the height information (Z), and uses information related to the virtual plane which is the XY plane when moving. The occupied area of the object is represented by a rectangular area in a two-dimensional virtual plane, and is represented by dimensions in two directions (X direction and Y direction) perpendicular to each other along the virtual plane. Here, a chair may be lifted and moved by a person, and a large fixture may fall, so when simulating an environment that involves the movement or fall of a chair or fixture, the orientation of the object By changing the occupancy area of the object using the height information, it is possible to fold the change in the environment accompanying the movement or the fall of the object into the virtual space.

  Therefore, in order to perceive the environmental conditions described above, the simulation apparatus 1 is provided with sensor means 12. The sensor means 12 includes a plurality of types of sensors 12a, 12b,... So that a plurality of types of human sensations can be simulated, and each sensor 12a, 12b,. To do. The types of sensors 12a, 12b,... Are set according to the target environment. Therefore, the environmental condition output by the sensor means 12 is a set of a plurality of sensory data. Detection of the environmental condition by the sensor means 12 is performed every unit time (for example, 1 second). Specific examples of sensory data obtained by the sensors 12a, 12b,... Will be described later.

  The environmental condition detected by the sensor means 12 is given to the action determining means 13, and the action determining means 13 determines the type of action by comparing the environmental condition with the action database 15. That is, the behavior database 15 stores behavior rules in which a plurality of types of behaviors that can be performed by a virtual human in the virtual space generated by the environment simulation means 11 and environmental conditions that activate each behavior are associated with each other. The behavior determining unit 13 determines the type of behavior by checking the behavior rule of the behavior database 15 when the environmental condition is given from the sensor unit 12. The action determination unit 13 detects whether or not an action target that the virtual human should preferentially achieve with respect to the movement from the initial position to the final position occurs with respect to the environmental condition given from the sensor unit 12. A detection unit 13a and a behavior calculation unit 13b that determines a behavior vector including a direction in which the virtual human acts with respect to the environmental condition given from the sensor unit 12 are provided. The behavior target detected by the behavior target detection unit 13a and the behavior vector calculated by the behavior calculation unit 13b will be described later.

  In the action determination means 13, if the action target is not detected even if the action target detection unit 13a collates the environmental conditions detected by the sensor means 12 every unit time with the action database 15, the action type is not changed and the action is not changed. Only the behavior vector is calculated by the calculation unit 13b. Further, when an action target that should be preferentially performed before reaching the final position when the environmental condition from the sensor means 12 is collated with the action database 15, the action database 15 is set so as to achieve the action target. The type of action is determined based on the collation result, and the action calculation unit 13b calculates an action vector for the action. When the main goal is to reach the final position from the initial position, the goal achieved in preference to the main goal can be called an intermediate goal. In the action determining means 13, when the achievement of the intermediate target is requested by the action target detection unit 13a, other actions are prohibited until the intermediate goal is achieved. For example, when an earthquake is assumed and there is a desk in the room where the virtual human exists, entering the desk under the desk to avoid falling objects can be set as an intermediate target (action target). . In this case, in an earthquake environment scenario, when there is an environmental condition in which a virtual human exists in a room and there is a desk where a person can enter under the room, the action of entering under the desk The target is activated.

  By the way, the behavior calculation unit 13b uses the sensory data obtained from the sensors 12a, 12b,... As a reference and obtains a behavior vector by adding (synthesizing) vectors that determine the direction and speed of movement of the virtual human. . That is, the weighting coefficient is determined based on sensory data obtained from the sensors 12a, 12b,..., And the weighting coefficients are respectively assigned to a plurality of vectors defined in relation to the environments detected by the sensors 12a, 12b,. The action vector is determined by multiplying and adding. Since the vectors for obtaining the behavior vectors are related to the environment detected by the sensors 12a, 12b,..., These vectors will be described later together with the types of the sensors 12a, 12b,.

  When the action type and the action vector are determined by the action determining unit 13 as described above, the action instruction unit 14 instructs the virtual human using the action type and the action vector. For example, the action instruction unit 14 sets a function having the action type as a function name and an action vector as a parameter, and causes the virtual human to perform an action corresponding to the function name with the direction and speed specified by the action vector. Here, as a function name for movement, two types of “walking” and “running” can be selected, and “walking” and “running” are displayed not only on the speed difference but also on the display means 22. It is used as information for changing the form of the virtual human. That is, since it is unnatural to “run” in the “walking” posture, the “running” posture is extracted from the motion database 23 in the output means 21 when “running”. Here, “walking” or “running” includes speed information, and is selected according to the magnitude of the action vector obtained by the action calculating unit 13b.

  When the action determination means 13 determines the action type and action vector, and the action instruction means 14 instructs the contents of the action to the virtual human, and the virtual human executes the action, the sensors 12a, 12b,... Changes in the environment.

  For example, if there is a desk that cannot be moved easily in the room where the virtual human acts, and the virtual human tries to go straight from the initial position to the final position, the desk becomes an obstacle. It is only necessary to go straight to the final position until it approaches, but in the vicinity of the desk, the direction of movement must be changed so as to avoid the desk that is an obstacle. Therefore, the sensor 12a provided in the sensor means 12 needs to detect an obstacle such as a desk. It is assumed that the sensor 12a detects the distance from the current position of the virtual human to the obstacle. In the behavior target detection unit 13a, a threshold is defined for the distance detected by the sensor 12a. When the distance detected by the sensor 12a is within the threshold, it is determined that there is an environmental condition for starting the behavior, and the detection is performed by the sensor 12a. An action vector is calculated using a weighting factor associated with the distance. However, when the distance detected by the sensor 12a exceeds the threshold, the output of the sensor 12a is invalid for the calculation of the action vector. That is, as shown in FIG. 2A, a threshold value Th1 is set for the distance, and when the distance is within the threshold value Th1, a weighting factor w1 (0 ≦ w1 ≦ 1) that increases as the distance decreases is set. In the range where the distance exceeds the threshold Th1, the weighting factor w1 is 0. When the distance to the obstacle is far, the distance to the obstacle is independent of the direction in which the virtual human moves. The threshold Th1 is set to a distance that allows the weighting factor w1 to be selected in the range of 0.6 to 0.8.

  Further, in evacuation behavior such as a disaster, when there are other people nearby, the sensor means 12 often needs a sensor 12b for detecting the existence of others. The sensor 12b has a function of detecting the distance to each other virtual human. When there are three or more virtual humans in the virtual space, the distance to the closest virtual human is adopted. A threshold is set for the distance. If the distance detected by the sensor 12b is equal to or greater than the threshold, it is determined that there is an environmental condition for starting the action, and a weighting factor associated with the distance detected by the sensor 12b is set. To calculate the action vector. However, when the distance detected by the sensor 12b is smaller than the threshold value, the output of the sensor 12b becomes invalid for the calculation of the action vector. That is, as shown in FIG. 2B, a threshold value Th2 is set for the distance, and when the distance is greater than or equal to the threshold value Th2, a weighting factor w2 (0 ≦ w2 ≦ 1) that increases as the distance increases is set. In the range where the distance is smaller than the threshold Th2, the weighting factor w2 is 0, and when the distance to the other virtual human is short, the distance to the other virtual human is irrelevant to the direction in which the virtual human moves. However, when the distance is closer than the threshold value Th2, an action rule is adopted in which the virtual human moves closest to the final position among the adjacent virtual humans. The threshold Th2 is set to a distance that can be selected in the range where the weight coefficient w2 is 0.1 to 0.3.

  By the way, assuming the case of simulating the environment at the time of fire, the smoke density also affects the behavior of the virtual human. When the sensor unit 12 is provided with the sensor 12c for detecting the smoke density, as shown in FIG. 3, an appropriate shape centered on the current position P1 of the virtual human (in the illustrated example, a square) ) And the neighborhood area DA is further divided into a plurality (9 in the illustrated example) of unit areas D11 to D13, D21 to D23, and D31 to D33. Since the position in the virtual space is generally expressed using an orthogonal coordinate system, in the illustrated example, squares are adopted as the shapes of the unit areas D11 to D13, D21 to D23, and D31 to D33 in order to simplify the processing. However, the size and shape of one unit region D11-D13, D21-D23, D31-D33 can be appropriately set according to the environment to be simulated.

  The sensor 12c detects the smoke density for each of the unit areas D11 to D13, D21 to D23, and D31 to D33. If the smoke density of all the unit areas D11 to D13, D21 to D23, D31 to D33 is equal to or less than the prescribed threshold, the unit areas D11 to D13, D21 to D23, D31 can be moved in all directions from the current position P1. When the smoke density of any one of -D33 exceeds the threshold value, the smoke density can be moved in the minimum direction. Such behavior rules for the smoke density are stored in the behavior database, and the smoke density has a threshold value in any one of the unit areas D11 to D13, D21 to D23, and D31 to D33 in the behavior target detection unit 13a. When it is determined that it exceeds, the weighting factor w3 is set to 1 so as to increase the moving direction vector determined by the smoke concentration distribution as described above. The weight coefficient w3 is set to 0. Here, binary values of 0 and 1 are adopted as the weighting factor w3, but the weighting factor w3 may be changed according to the smoke density.

  The threshold Th1 related to the distance to the obstacle, the threshold Th2 related to the distance to other virtual humans, and the threshold related to the smoke density are all action rule switching points, and the threshold Th1 is the action rule of the action target of avoiding the obstacle. The switching point to be activated, the threshold value Th2, is a switching point for activating an action rule for an action target to approach another virtual human, and the threshold value for smoke density is a switching point for activating the action rule for an action target to escape from smoke. For example, an action rule of approaching another virtual human is adopted in a section having a distance larger than the threshold Th2, but an action rule of following another virtual human is adopted in a section having a distance equal to or less than the threshold Th2. Different action rules are adopted depending on the section. That is, when an action rule that follows another virtual human is adopted, the vector for the action rule that approaches another virtual human is 0, but the vector in the action rule that follows another virtual human is 0. Instead, it becomes the same action vector as other virtual humans.

Now, let V1 be a vector for avoiding obstacles, V2 be a vector approaching other virtual humans, V3 be a vector determined by the smoke density distribution, and use these vectors V1 to V3 to make a virtual human action vector. If V is to be determined, the behavior calculation unit 13b combines the vectors V1 to V3 with the following calculation formula.
V = w1, V1 + w2, V2 + w3, V3
By defining the direction and speed of the movement of the virtual human using such an action vector V, the behavior of the virtual human can be satisfied so as to avoid obstacles, follow other virtual humans, and escape from smoke. Can be determined.

  Since the above-described action vector V can be determined in a two-dimensional virtual plane, when the action of the virtual human is determined only by the above three conditions, the virtual human in the three-dimensional virtual space in which the virtual human behaves is determined. Mapping means 17 for extracting only information of a two-dimensional virtual plane on which the person acts is provided, and the action vector V is determined on the virtual plane obtained by the mapping means 17. By such processing, the amount of calculation can be greatly reduced as compared with the case where the above-described calculation is performed using a three-dimensional virtual space, and an action vector can be determined in real time.

  As described above, the type of action (two types of action, “running” and “walking” are basically selected in the event of a fire) and the action vector are determined, and the action is given to the virtual human through the action instruction means 14. , The virtual human moves at the speed indicated in the direction indicated by the action vector. In the present invention, since the sensor means 12 detects the environmental condition every unit time, the virtual human performs the action determined by the environmental condition detected by the sensor means 12 for the unit time, and after the action, the sensor means 12 The operation of performing the action determined by the environmental condition detected again for a unit time is repeated. That is, the detection of the environmental condition and the action are repeated every unit time, and if the environmental condition detected by the sensor means 12 changes due to the environmental scenario or the action of the virtual human, it corresponds to the change of the environmental condition. Can change the behavior. Such actions of perception of environmental conditions and modification of behavior according to perceived environmental conditions simulate the processing in the brain when humans actually act with high accuracy, and virtual humans in virtual space This makes it possible to accurately simulate human behavior in real space.

  In the above example, the case where the behavior at the time of a fire is simulated has been described. However, the present invention can simulate the behavior of a person in an abnormal state such as an earthquake, a flood, and a traffic jam in addition to a fire. Thus, for example, in the event of an earthquake, the position of an obstacle can be changed due to furniture falling down, or a desk that becomes an obstacle in the evacuation route can be used as an evacuation site in the event of a fire. In addition, since it is only indoors that can be used as an evacuation area under the desk, there is no desk in an automobile or the like, so it is necessary to apply behavior rules different from those in the room. Since the difference between indoors and cars is determined by the environmental scenario, this difference is dealt with by setting a set of action rules corresponding to the environmental scenario in the action database 15. Therefore, when simulating an earthquake, the sensors 12a, 12b,... Provided in the sensor means 12 can be detected as obstacles to avoid not only installed furniture but also fallen furniture, etc. It is necessary to have something that detects the presence of desks that can be used, and something that detects the presence of other virtual humans as well as in the event of a fire. Also, during an earthquake, actions such as “stop the gas tap” and “drop the breaker” may be required, so it may be necessary to detect the presence of an object of this type of action. . Furthermore, since fires often occur during an earthquake, it may be necessary to detect the smoke concentration distribution as in the case of a fire.

  Now, an action rule of “entering under the desk” is defined in the event of an earthquake, and the sensor means 12 searches whether there is a desk that can enter under the room, and the presence is detected, and When the sensor means 12 detects that the seismic intensity of the earthquake set by the environmental scenario is equal to or greater than the threshold, the behavior target detection unit 13a activates (ignites) this behavior rule. When this action rule is activated, the action target is “enter under the desk”, so other types of actions are activated until the action target “enter under the desk” is achieved. First, an action vector is set so as to approach the desk. The goal of “entering under the desk” is completed when there is no earthquake vibration in the state under the desk. After the completion of the action target, the action vector is generated so that the action returns to the normal action and the virtual human can move to the final position.

  In the event of an earthquake, fallen furniture and falling objects may become obstacles, but these obstacles are not known and are caused by the occurrence of an earthquake. Therefore, the virtual human cannot have knowledge about the position of these obstacles in advance, and is perceived after the obstacle occurs. In view of this, a sensor that detects this type of obstacle has a smaller distance that serves as a switching point of the action rule than the sensor 12a that detects a normal obstacle. In other words, since a person recognizes this kind of obstacle in real space only after approaching the obstacle, the switching of action rules for this kind of obstacle is performed at a shorter distance than when a normal obstacle is recognized. It is set to a point.

  The environment generated by the environment simulating means 11 is held by the virtual human as knowledge, and the virtual human perceives the environment (distance etc.) relative to the current position by the sensor means 12. In addition, for environments that change over time in the virtual space, such as smoke density, the environment for each unit time for each location in the virtual space is calculated in advance within the simulated time range, and virtual humans are calculated from the calculated data. The data of the area corresponding to the current position is extracted by the sensor means 12.

  A user interface 3 is connected to the simulation apparatus 1 so that the user can set the environment and behavior to be simulated in the simulation apparatus 1. The user interface 3 includes a virtual space editor 31 that enables creation and editing of a virtual space generated by the environment simulation unit 11 and an environment scenario editor 32 that enables creation and editing of an environment scenario set in the environment simulation unit 11. Provided. The virtual space and the environment scenario may be provided from a separate application 30. For example, data in a virtual space provided on an Internet website can be used. Further, the function of the sensor means 12 that is an attribute of the virtual human can be selected and edited by the sensor editor 33, and the contents of the motion database 23 that is the attribute of the virtual human can be created and edited by the motion editor 34. Yes. That is, the sensor editor 33 and the motion editor 34 constitute a part of a virtual human editor (not shown). The virtual human editor also has a function of arranging a plurality of virtual humans in a virtual space. When a plurality of virtual humans are arranged, an attribute can be assigned to each virtual human. As an attribute of the virtual human, a guider that induces the behavior of another virtual human can be selected. Here, the guider has a role of instructing other virtual humans in the virtual space in the virtual space, and if the other virtual human has an instruction from the guider, the other person instructs the action given priority over other actions. Do. Also, in the virtual human editor, when a plurality of virtual humans are arranged in the virtual space, a function for identifying the virtual human to be focused on, when a plurality of virtual humans approach, or another virtual human is followed. It also has a function to select whether to follow. Furthermore, the user interface 3 is provided with an action editor 35 for creating and editing action types and action rules in the action database 15.

  As described above, since each unit of the simulation apparatus 1 and the imaging apparatus 2 can be externally given data or edited, the simulation apparatus 1, the imaging apparatus 2, and the user interface 3 are realized. The program may be provided so that the user himself / herself can set the virtual space, the environmental scenario, the behavior type and behavior rule, and the virtual human attribute. For example, when used to verify whether or not an evacuation route at the time of a fire is secured in a construction company, the environment simulation means 11 obtains the drawing data from the application 30 such as a CAD that creates drawing data for designing the building. If the virtual space is set by taking in the data and the data necessary for the fire simulation is set using the user interface 3, the evacuation behavior in the building can be simulated. Therefore, the problem of evacuation behavior can be found easily, which helps to optimize the design for securing the evacuation route and the placement of guiders.

  Further, a part of the simulation apparatus 1 and the imaging apparatus 2 excluding the display unit 22 and a program for realizing the user interface 3 are provided in the network server, and the display unit 22 is provided in a terminal connected to the network server and the terminal. If the user interface 3 can be operated, it is possible to provide a virtual human behavior simulation system service as an application service provider (ASP) using a network such as the Internet.

It is a block diagram which shows embodiment of this invention. It is operation | movement explanatory drawing same as the above. It is principle explanatory drawing regarding perception of smoke density in the same as the above.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 Simulation apparatus 2 Imaging apparatus 3 User interface 11 Environment simulation means 12 Sensor means 12a, 12b, ... Sensor 13 Action determination means 13a Action target detection part 13b Action calculation part 14 Action instruction means 15 Action database 16 Position designation means 17 Mapping Means 21 Output means 31 Virtual space editor 32 Environmental scenario editor 33 Sensor editor 34 Motion editor 35 Action editor

Claims (7)

Environment simulation means for generating an environment of a virtual space in which a virtual human behaves according to a predetermined environment scenario, position specifying means for specifying the initial position of the virtual human and a final position to be a target, and virtual generated by the environment simulation means An action database storing action rules in which a plurality of types of actions that can be performed by a virtual human in space and an environmental condition that activates each action are stored, sensor means for detecting the environmental condition, and environmental conditions detected by the sensor means The action type including the direction of the virtual human action is determined from the current position of the virtual human, the final position specified by the position specifying means, and the environmental condition detected by the sensor means. Action decision means for determining vectors and action decision The action instruction means for instructing the virtual human to execute the action according to the action type and action vector determined in the step, and the virtual human for executing the action instructed by the action instruction means are displayed together with the virtual space generated by the environment simulation means. and output means for displaying on the device screen, said sensor means has a function of virtual human perceives the object existence of an obstacle in towards the final position, both actions when the virtual human is directed to the final position At least a function of perceiving the presence or absence of another person, wherein the action determination means is a distance from the object to a vector from the current position of the virtual human to the final position and a vector to avoid the object detected by the sensor means. Multiplied by the first weighting factor that increases as the distance is shorter when is within the first threshold And a vector obtained by multiplying the vector detected by the sensor means and approaching the other person by a second weighting factor that increases as the distance increases when the distance to the other person is equal to or greater than the second threshold. The action vector is determined, and the operation in which the virtual human performs the action instructed by the action instructing means and the environmental condition is detected by the sensor means is repeated every unit time until the virtual human reaches the final position. Behavior simulation system.   The virtual space generated by the environment simulating means is three-dimensional, and mapping means for mapping the virtual space to a two-dimensional virtual plane is added, and the action determining means adds an action vector in the virtual plane obtained by the mapping means. The behavior simulation system according to claim 1, wherein the behavior simulation system is determined. The sensor means has at least a function of perceiving a specific environmental condition that defines a behavioral goal of a virtual human, and the behavior determination means corresponds to the environmental condition when the sensor means detects the presence of the environmental condition. 3. The behavior simulation system according to claim 1, wherein other behavior is prohibited until the behavior goal is achieved . The environment simulation means has a function of generating an environment at the time of fire in the virtual space and simulating a temporal change in the smoke concentration distribution in the virtual space at the time of the fire, and the sensor means is a function of detecting the smoke density The behavior database corresponds to the behavior of moving to a lower smoke concentration with respect to an environmental condition that the smoke concentration within a specified range centered on the current position of the virtual human is equal to or higher than a specified threshold. 3. The behavior simulation system according to claim 1, wherein the attached behavior rule is stored . The environment simulation means has a function of generating an environment at the time of an earthquake in a virtual space, and the sensor means has a function of detecting a desk as a specific environmental condition that defines a behavioral target of a virtual human, The database stores an action rule that associates the action of entering under the desk with respect to the environmental condition that the desk is present, and the action determining means is located under the desk when the sensor means detects the presence of the desk. claim 1 or claim 2 behavioral simulation system according to and inhibits the other actions until behavioral goals are achieved that fall. A virtual space editor for editing the virtual space generated by the environment simulation means, an environment scenario editor for editing an environment scenario of the virtual space generated by the environment simulation means, and a virtual human attribute displayed on the display means by the output means The behavior simulation according to claim 1, further comprising: a virtual human editor that edits an action, and an action editor that edits an action type and an action rule stored in the action database. system. The virtual human editor is capable of arranging a plurality of virtual humans in a virtual space generated by an environment simulating means, and selecting a guider that induces another virtual human action as a virtual human attribute. behavioral simulation system of claim 6, wherein the.

Images