A three-dimensional map quality assessment method and system

Abstract

A three-dimensional map quality assessment method and system, the method comprising the following steps: converting the three-dimensional map into an octree map; setting a corresponding intelligent agent according to the user type of the three-dimensional map, and setting a reward function and a behavior function of the intelligent agent; placing the intelligent agent in the octree map, the intelligent agent moving in the octree map according to its behavior function, and calculating the quality value of the map grid in the octree map according to the interaction between the intelligent agent and the octree map and the reward function; and assessing the quality of the three-dimensional map relative to the user type based on the sum of the quality values ​​of all the map grids in the octree map.

Description

Technical Field

[0001] The present invention relates to the field of three-dimensional map technology, and in particular to a three-dimensional map quality assessment method and system. Background Art

[0002] As one of the most cutting-edge technologies in graphics and multimedia, the core of three-dimensional maps is the spatial layout design of the virtual world. Three-dimensional digital maps have significant spatial design significance. They use a method consistent with geographical theory to integrate reality and fiction and attach them to various applications such as factory layout, workshop layout, production line planning, etc. Therefore, three-dimensional digital maps are an important part of digital design as a spatial background for production operations to carry various resources. Among them, ensuring that the virtual avatar representing the real person can move in an open and feasible space and reach the expected destination is an important part of three-dimensional map path planning. At present, the basic pathfinding algorithms include Dijkstra algorithm, A* pathfinding algorithm, B* pathfinding algorithm, etc., which are usually used for the efficiency of three-dimensional map accessibility testing.

[0003] At present, on the one hand, path-finding algorithms all plan the shortest or optimal path with clear starting and ending points. They regard three-dimensional maps as fixed environments, which is not very helpful for multi-target and multi-type map design. On the other hand, there has been a lack of effective tools for the level evaluation of three-dimensional map design, especially for three-dimensional maps for human activities. Their testing can generally only rely on real people to roam and explore the map. The test is highly subjective and difficult to evaluate objectively. And with the development of PCG (Procedural Content Generation) technology, large-scale three-dimensional maps can be generated randomly and in batches. If manual testing of map quality is still used, the efficiency is low and the map quality cannot be objectively evaluated. Summary of the invention

[0004] In view of the above analysis, an embodiment of the present invention aims to provide a three-dimensional map quality assessment method system to solve the problem that the existing manual testing of map quality is inefficient and cannot objectively assess map quality.

[0005] On the one hand, an embodiment of the present invention provides a three-dimensional map quality assessment method, comprising the following steps:

[0006] Converting the three-dimensional map into an octree map;

[0007] Setting a corresponding intelligent agent according to the user type of the three-dimensional map, and setting a reward function and a behavior function of the intelligent agent;

[0008] The agent is placed in the octree map, the agent moves in the octree map according to its behavior function, and the quality value of the map grid in the octree map is calculated according to the interaction between the agent and the octree map and the reward function;

[0009] The quality of the three-dimensional map relative to the user type is evaluated based on the sum of the quality values ​​of all map grids in the octree map.

[0010] The beneficial effects of the above technical solution are as follows: by converting the three-dimensional map into an octree map, different types of maps can be quickly normalized, thereby providing a unified scale standard for subsequent quality evaluation. By using an intelligent agent to replace real human users for map testing, the quality of the map grid is calculated based on the interaction between the intelligent agent and the spatial map, and the quality of the three-dimensional map for the user type is comprehensively evaluated through the quality values ​​of all map grids, so as to quickly and objectively obtain the quality of the three-dimensional map for a certain type of user. The method is simple and easy to implement, which improves the efficiency and accuracy of the three-dimensional map quality evaluation.

[0011] Based on the further improvement of the above method, the reward function of the agent is set, including:

[0012] The reward function of the agent is the reward value of the current map grid obtained when the agent reaches the current map grid of the octree map for the first time. The task goal of the agent is to maximize the cumulative reward value; the reward value of each map grid in the octree map is the same.

[0013] The beneficial effects of the above technical solution are as follows: by setting the reward value of each map grid to be the same, the intelligent agent obtains the reward value of the map grid when it interacts with the map grid for the first time. The task goal of the intelligent agent is to maximize the cumulative reward value, thereby driving the intelligent agent to continuously explore more map space through learning, and then whether the intelligent agent can obtain rewards, the convenience of obtaining rewards and the magnitude correspond to the quality of the spatial map.

[0014] Furthermore, setting the behavior function of the agent includes:

[0015] According to the user type corresponding to the agent, the random behavior selection probability of selecting the next behavior when the agent is in an obstacle-free environment is set, and

[0016] The probability of choosing different behaviors when the agent is in different obstacle environments.

[0017] The beneficial effects of the above technical solution are as follows: by analyzing the behavior patterns of user types, setting the behavior functions of the intelligent agents to correspond to the behavior patterns of the corresponding user types, so that the intelligent agents can replace the corresponding user types, thereby obtaining objective map quality evaluation results.

[0018] Further, calculating the quality value of the map grid in the octree map according to the interaction between the agent and the octree map and the reward function includes:

[0019] When the agent roams to the current map grid for the first time, the quality value of the current map grid is calculated as Among them, Q i Represents the quality value of the current map grid, r represents the reward value of each map grid, γ represents the discount rate, and k represents the maximum number of map grids that the agent can remember.

[0020] The beneficial effects of the above technical solution are as follows: by setting the maximum memory number of the corresponding intelligent agent according to the memory capacity of the user type, and by associating the map grid quality with the reward and memory capacity of the map grid, different memory capacities can be used to experience different map quality, thereby achieving map quality evaluation for different user types.

[0021] Furthermore, for different three-dimensional maps of the same scale, within the same test time, the sum of the quality values ​​of all map grids of each three-dimensional map relative to the same user type is obtained. The higher the sum of the quality values, the higher the quality relative to the user type.

[0022] The beneficial effects of the above technical solution are as follows: for three-dimensional maps of the same scale, within the same test time, the three-dimensional map with a higher total quality value has a higher quality for the user type, thereby achieving rapid and objective evaluation of high-quality maps among multiple three-dimensional maps of the same scale.

[0023] Further, based on the sum of the quality values ​​of all map grids in the octree map, evaluating the quality of the three-dimensional map relative to the user type includes:

[0024] For the same 3D map, within the same test time, the sum of the quality values ​​of all map grids of the 3D map relative to different user types is obtained. The higher the sum of the 3D map quality values ​​corresponding to the user type, the higher the quality relative to the user type.

[0025] The beneficial effects of the above technical solution are as follows: for the same three-dimensional map, within the same test time, the sum of the quality values ​​of all map grids of the three-dimensional map relative to different user types is obtained. The higher the sum of the three-dimensional map quality values ​​corresponding to the user type, the higher the quality relative to the user type, thereby achieving a rapid and objective evaluation of the quality values ​​of the three-dimensional map for different users, and then targeted user sorting and push, etc.

[0026] On the other hand, an embodiment of the present invention provides a three-dimensional map quality assessment system, including the following modules:

[0027] A map conversion module, used for converting the three-dimensional map into an octree map;

[0028] A setting module, used to set a corresponding intelligent agent according to the user type of the three-dimensional map, and set a reward function and a behavior function of the intelligent agent;

[0029] A testing module, used for placing the agent in the octree map, the agent moving in the octree map according to its behavior function, and calculating the quality value of the map grid in the octree map according to the interaction between the agent and the octree map and the reward function;

[0030] An evaluation module is used to evaluate the quality of the three-dimensional map relative to the user type based on the sum of the quality values ​​of all map grids in the octree map.

[0031] Furthermore, the setting module sets the reward function of the agent, including:

[0032] The reward function of the agent is the reward value of the current map grid obtained when the agent reaches the current map grid of the octree map for the first time. The task goal of the agent is to maximize the cumulative reward value; the reward value of each map grid in the octree map is the same.

[0033] Furthermore, setting the behavior function of the agent includes:

[0034] According to the user type corresponding to the agent, the random behavior selection probability of selecting the next behavior when the agent is in an obstacle-free environment is set, and

[0035] The probability of choosing different behaviors when the agent is in different obstacle environments.

[0036] Furthermore, it is characterized in that the test module calculates the quality value of the map grid by the following method:

[0037] When the agent roams to the current map grid for the first time, the quality value of the current map grid is calculated as Among them, Q i Represents the quality value of the current map grid, r represents the reward value of each map grid, γ represents the discount rate, and k represents the maximum number of map grids that the agent can remember.

[0038] In the present invention, the above-mentioned technical solutions can also be combined with each other to achieve more preferred combination solutions. Other features and advantages of the present invention will be described in the subsequent description, and some advantages can become obvious from the description, or can be understood by practicing the present invention. The purpose and other advantages of the present invention can be realized and obtained through the contents particularly pointed out in the description and the drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] The drawings are only for the purpose of illustrating particular embodiments and are not to be considered limiting of the present invention. Like reference symbols denote like components throughout the drawings.

[0040] Figure 1 A flowchart of a three-dimensional map quality assessment method according to an embodiment of the present invention;

[0041] Figure 2 The structure diagram of the three-dimensional map quality assessment system according to the embodiment of the present invention is shown in FIG. DETAILED DESCRIPTION

[0042] The preferred embodiments of the present invention are described in detail below in conjunction with the accompanying drawings, wherein the accompanying drawings constitute a part of this application and are used together with the embodiments of the present invention to illustrate the principles of the present invention, but are not used to limit the scope of the present invention.

[0043] With the development of programmatic content generation technology, the scale of maps continues to expand. The quality of maps is mainly based on subjective evaluation and is difficult to measure objectively. Especially when the scale of maps is large and the number is large, it becomes increasingly difficult to evaluate maps with higher interactive quality for a certain type of user. The manual testing method has a long cycle, low efficiency and strong subjective factors.

[0044] In view of this, a specific embodiment of the present invention discloses a three-dimensional map quality assessment method, such as Figure 1 The method comprises the following steps:

[0045] S1. Convert the three-dimensional map into an octree map.

[0046] A three-dimensional map is a three-dimensional space. The three-dimensional space modeling is divided into many small cubes. If each face of each small cube is cut into two pieces, then the small cube will become eight small cubes of the same size. This step is repeated until the final cube size reaches the highest accuracy of modeling. In this process, eight sub-nodes are expanded from one node. Then, the process of subdividing the entire space from the largest space to the smallest space is an octree, which completely frames the objects of different scales in the original various maps into the sub-nodes of the corresponding level. Therefore, ordinary three-dimensional maps can be converted into octree maps through 3D software. The process of the octree dividing the subspace layer by layer also contains the spatial information of the object, which also makes the subsequent processing more efficient. When implementing, you can use Minecraft's supporting tool MCedit to convert the three-dimensional map into an octree map.

[0047] By converting the three-dimensional map into an octree map, different types of spatial maps can be simply normalized, which is convenient for subsequent calculation and processing and map quality evaluation. The number of map grids in the octree map represents the scale of the three-dimensional map.

[0048] S2. Setting a corresponding intelligent agent according to the user type of the three-dimensional map, and setting a reward function and a behavior function of the intelligent agent.

[0049] Reinforcement learning is a learning method in which an agent learns by trial and error. The agent is guided by rewards or punishments obtained through interaction with the environment. The goal is to maximize the reward for the agent. The agent is trained through a reward and punishment mechanism. The agent will be rewarded for correct behavior and punished for incorrect behavior. In this way, the agent will try to minimize its incorrect behavior and maximize its correct behavior. The trained agent can be applied to corresponding tasks, such as autonomous driving, chess and card games, etc.

[0050] This application combines reinforcement learning with map quality assessment, and uses an agent to test map quality instead of real users. Since the purpose of this application is to test map quality rather than to train the agent to perform the most correct behavior, there is no penalty for the agent.

[0051] The same map has different interaction qualities for different types of users. By using intelligent agents to replace the corresponding user types, the quality of the 3D map to be evaluated for different types of users can be quickly and objectively measured.

[0052] Specifically, users with different spatial interactive roaming time and different types of space contact have different spatial direction decision-making ability, behavior randomness, and memory ability of spatial terrain. Cluster analysis of user behavior patterns can be performed from three perspectives: spatial direction decision-making ability, behavior randomness, and memory ability of spatial terrain, thereby obtaining different user types. Specific user types can be obtained based on research and analysis of target users of three-dimensional maps.

[0053] After getting the user type, you can set the agent corresponding to the user type. Agents of different types do not share maps and any behavior parameters.

[0054] Specifically, the agent's behaviors in the map include turning left, turning right, turning back, jumping, moving forward, moving backward, and stopping.

[0055] The agent's behavior strategy function can be considered as outputting an action a through internal processing when the spatial map state is s. The agent's behavior function corresponding to the user type can be set from the two dimensions of directional decision-making ability and behavior randomness, and the agent's behavior strategy can be matched with the behavior pattern of the user type.

[0056] Directional decision-making ability refers to the ability to infer directional associations under specific spatial states. Strong directional decision-making ability is manifested in the ability to infer the movable direction under specific spatial states. For example, when encountering an L-shaped roadblock, one can quickly decide to turn to the right front. If the decision-making ability is weak, it is manifested as a directional decision without direct association, and the user is more like rolling dice to choose the direction of movement.

[0057] Behavioral randomness refers to the probability of map roaming behaviors, such as moving forward, backward, stopping, and jumping, remaining consistent when there is no stimulation from a specific spatial state (for example, when there are no obstacles in front, behind, left, and right, and all can be passed). The stronger the randomness, the more unpredictable the map roaming behaviors in each time segment are, which is close to random roaming.

[0058] According to the behavior pattern of the user type, the behavior function of the corresponding agent is set as follows:

[0059] Setting the random behavior selection probability of the agent to select the next behavior when it is in the obstacle-free map state, and

[0060] The probability of choosing different behaviors when the agent is in different obstacle environments.

[0061] Specifically, different types of users have different randomness in their behavior choices in barrier-free environments. For example, users with more spatial work experience have stronger personal operating habits, which leads to stronger behavioral logic, so the probability of their random behavior is very low. On the contrary, users with less spatial roaming work experience have greater randomness in their behavior choices due to their less experience.

[0062] Therefore, according to the behavior patterns of different user types, the random selection probability of the corresponding agent's next behavior in an accessible environment can be set.

[0063] The smaller the random behavior probability of the agent, the greater the probability of keeping the last action. The greater the random probability of behavior, the greater the probability that the agent will choose a different action each time it moves. For example, if the random behavior probability of the agent is 0.7, then on average 3 out of 10 actions are continuous, that is, the agent keeps the same behavior as the last time. If the random behavior probability of the agent is 0.1, then on average 9 out of 10 actions are continuous. The specific random probability comes from the user type analysis.

[0064] Different types of users have different behavior choices in an environment with obstacles. For example, for a type of user with a strong spatial direction perception ability, the probability of having a clear direction of action when encountering an obstacle is higher, such as the probability of choosing to retreat when encountering a U-shaped obstacle is higher; while for a type of user with a weaker direction perception ability, the probability of having a clear direction of action when encountering an obstacle is lower; for a type of user with rich experience in parkour space roaming, the probability of choosing to jump when encountering an obstacle is higher. During implementation, for example, for a certain type of user, the probability of choosing to retreat when encountering a U-shaped obstacle is 80%, and the corresponding intelligent agent is set to have a probability of choosing to retreat when encountering a U-shaped obstacle of 80%. Then, when the intelligent agent encounters a U-shaped obstacle 10 times, it will choose to retreat 8 times on average.

[0065] Therefore, according to the behavior patterns of different types of users, the probability of the corresponding agent choosing different behaviors for the next action in an obstacle environment can be set.

[0066] Setting the agent's reward function includes:

[0067] The reward function of the agent is to obtain the reward of the current map grid when the agent reaches the current map grid of the octree map for the first time. The task goal of the agent is to maximize the cumulative reward value; the reward value of each map grid in the octree map is the same.

[0068] The agent obtains rewards through interaction with the environment (spatial map), which is generally in the form of r(a), that is, the reward for the current action. For this application, since only the friendliness of the spatial map to the agent, that is, the quality of the map rather than the training of the agent, is considered, the various settings and devices on the spatial map, such as the lounge and studio, are not considered. The impact on the welfare of the agent representing the user, then the spatial map should theoretically be homogeneous, and the rewards or benefits to the user should be the same, that is, the reward value of each map grid on the map is the same.

[0069] When the agent reaches a map grid for the first time, it obtains the reward for the current map grid. The agent's task goal is to maximize the cumulative reward value, thereby driving the agent to explore more map space.

[0070] During implementation, the Malmo reinforcement learning platform can be used to define the agent and its behavior functions. An agent XML can be used as a runtime configuration file. The configuration file example is as follows:

[0071] S3. placing the agent in the octree map, wherein the agent moves in the octree map according to its behavior function, and calculating the quality value of the map grid in the octree map according to the interaction between the agent and the octree map and the reward function.

[0072] The agent is placed in the octree map to roam. The agent moves in the octree map according to its behavior function. The agent obtains a reward value according to the interaction with the map grid, and the map grid obtains a quality value.

[0073] Whether the agent can obtain rewards, the convenience and magnitude of obtaining rewards illustrate the design quality of the spatial map. The overall design quality of the spatial map comes from the evaluation of the quality of each grid. Although the quality of each grid seems to be related to the rewards given directly (that is, the rewards directly obtained by the user), according to many psychological studies, the user's sense of welfare comes from reality and memory. Therefore, the quality of the single-grid map of this application is also related to the user's memory ability, that is, the quality of the map grid is related to the reward value obtained by the agent and the memory ability of the user type corresponding to the agent. The stronger the user's memory ability, the better the interactive experience obtained, and the greater the quality of the corresponding map grid. Therefore, the map quality is relative to a certain type of user, rather than a universal map quality.

[0074] Specifically, the quality of the map grid is calculated by the following method:

[0075] When the agent roams to the current map grid for the first time, the quality value of the current map grid is calculated as Among them, Q i represents the quality value of the current map grid, r represents the reward value of each map grid, γ represents the discount rate, and k represents the maximum number of map grids that the agent can remember, corresponding to the memory capacity of the user type. γ is the discount rate, also known as the forgetting rate, which represents the discount of the contribution of the map grid that the agent has walked to the quality of the current map grid. The farther the map grid that the agent has walked is from the current map grid (i), the smaller the contribution to the quality of the current map grid. γ varies according to the design requirements of different spatial maps, and can be taken as 0.8 during implementation.

[0076] When the agent explores the grid for the first time, it gets the reward for the grid and the grid gets the quality value. When the agent passes through the grid again, it no longer gets the reward for the grid and the quality of the grid is no longer calculated.

[0077] The stronger the user's memory ability, the better the map experience he or she will get. Conversely, the quality of the same map will be different for users with different memory abilities.

[0078] S4. Evaluate the quality of the three-dimensional map relative to the user type based on the sum of the quality values ​​of all map grids in the octree map.

[0079] Specifically, for different 3D maps of the same scale, within the same test time, the sum of the quality values ​​of all map grids of each 3D map relative to the same user type is obtained. The higher the sum of the quality values, the higher the quality relative to the user type.

[0080] If the agent runs for an infinite amount of time, it may roam any grid of the entire map, so the evaluation of the map quality should be done within the same testing period.

[0081] When a large number of maps of the same size are generated through programmatic content generation (PCG), the three-dimensional map with the highest quality for a certain type of user can be quickly evaluated according to the quality evaluation method of the present application.

[0082] For the same three-dimensional map, during the same test time, the sum of the quality values ​​of all map grids of the three-dimensional map relative to different user types is obtained. The higher the sum of the three-dimensional map quality values ​​corresponding to the user type, the higher the quality relative to the user type. For the same three-dimensional map, during the same test time, since the quality values ​​for different types of users are different, the method of this application can also be used to evaluate the quality values ​​of the three-dimensional map for different users, so as to perform targeted user sorting and push, etc.

[0083] A specific embodiment of the present application discloses a three-dimensional map quality assessment system, such as Figure 2 The following modules are shown:

[0084] A map conversion module, used for converting the three-dimensional map into an octree map;

[0085] A setting module, used to set a corresponding intelligent agent according to the user type of the three-dimensional map, and set a reward function and a behavior function of the intelligent agent;

[0086] A testing module, used for placing the agent in the octree map, the agent moving in the octree map according to its behavior function, and calculating the quality value of the map grid in the octree map according to the interaction between the agent and the octree map and the reward function;

[0087] An evaluation module is used to evaluate the quality of the three-dimensional map relative to the user type based on the sum of the quality values ​​of all map grids in the octree map.

[0088] Preferably, the setting module sets the reward function of the agent, including:

[0089] The reward function of the agent is the reward value of the current map grid obtained when the agent reaches the current map grid of the octree map for the first time. The task goal of the agent is to maximize the cumulative reward value; the reward value of each map grid in the octree map is the same.

[0090] Preferably, setting the behavior function of the agent includes:

[0091] According to the user type corresponding to the agent, the random behavior selection probability of selecting the next behavior when the agent is in an obstacle-free environment is set, and

[0092] The probability of choosing different behaviors when the agent is in different obstacle environments.

[0093] Preferably, the test module calculates the quality value of the map grid by the following method:

[0094] When the agent roams to the current map grid for the first time, the quality value of the current map grid is calculated as Among them, Q i Represents the quality value of the current map grid, r represents the reward value of each map grid, γ represents the discount rate, and k represents the maximum number of map grids that the agent can remember.

[0095] The above method embodiments and system embodiments are based on the same principle, and their related parts can be used for reference, and can achieve the same technical effect. The specific implementation process refers to the above embodiment, which will not be repeated here.

[0096] Those skilled in the art will appreciate that all or part of the processes of the above-mentioned embodiments can be implemented by instructing related hardware through a computer program, and the program can be stored in a computer-readable storage medium, wherein the computer-readable storage medium is a disk, an optical disk, a read-only storage memory, or a random access memory, etc.

[0097] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by any technician familiar with the technical field within the technical scope disclosed by the present invention should be covered within the protection scope of the present invention.

Claims

1. A three-dimensional map quality assessment method, characterized in that: The following steps are involved: Converting the three-dimensional map into an octree map; Setting a corresponding intelligent agent according to the user type of the three-dimensional map, and setting a reward function and a behavior function of the intelligent agent; The agent is placed in the octree map, the agent moves in the octree map according to its behavior function, and the quality value of the map grid in the octree map is calculated according to the interaction between the agent and the octree map and the reward function; Evaluating the quality of the three-dimensional map relative to the user type based on the sum of the quality values ​​of all map grids in the octree map; Setting the agent's reward function includes: The reward function of the agent is the reward value of the current map grid obtained when the agent first arrives at the current map grid of the octree map, and the task goal of the agent is to maximize the cumulative reward value; the reward value of each map grid in the octree map is the same; Setting the behavior function of the agent includes: According to the user type corresponding to the agent, the random behavior selection probability of selecting the next behavior when the agent is in an obstacle-free environment is set, and The probability of choosing different behaviors when the agent is in different obstacle environments.

2. The three-dimensional map quality assessment method according to claim 1, characterized in that: Calculating the quality value of the map grid in the octree map according to the interaction between the agent and the octree map and the reward function, including: When the agent roams to the current map grid for the first time, the quality value of the current map grid is calculated as ,in, Indicates the quality value of the current map grid. Indicates the reward value of each map grid, represents the discount rate, Indicates the maximum number of map grids that the agent can remember.

3. The three-dimensional map quality assessment method according to claim 1, characterized in that: Evaluating the quality of the three-dimensional map relative to the user type based on the sum of the quality values ​​of all map grids in the octree map, including: For different three-dimensional maps of the same scale, within the same test time, the sum of the quality values ​​of all map grids of each three-dimensional map relative to the same user type is obtained. The higher the sum of the quality values, the higher the quality relative to the user type.

4. The three-dimensional map quality assessment method according to claim 1, characterized in that: Evaluating the quality of the three-dimensional map relative to the user type based on the sum of the quality values ​​of all map grids in the octree map, including: For the same 3D map, within the same test time, the sum of the quality values ​​of all map grids of the 3D map relative to different user types is obtained. The higher the sum of the 3D map quality values ​​corresponding to the user type, the higher the quality relative to the user type.

5. A three-dimensional map quality assessment system, characterized in that: Includes the following modules: A map conversion module, used for converting the three-dimensional map into an octree map; A setting module, used to set a corresponding intelligent agent according to the user type of the three-dimensional map, and set a reward function and a behavior function of the intelligent agent; A testing module, used for placing the agent in the octree map, the agent moving in the octree map according to its behavior function, and calculating the quality value of the map grid in the octree map according to the interaction between the agent and the octree map and the reward function; An evaluation module, configured to evaluate the quality of the three-dimensional map relative to the user type based on the sum of the quality values ​​of all map grids in the octree map; The setting module sets the reward function of the agent, including: The reward function of the agent is the reward value of the current map grid obtained when the agent first arrives at the current map grid of the octree map, and the task goal of the agent is to maximize the cumulative reward value; the reward value of each map grid in the octree map is the same; Setting the behavior function of the agent includes: According to the user type corresponding to the agent, the random behavior selection probability of selecting the next behavior when the agent is in an obstacle-free environment is set, and The probability of choosing different behaviors when the agent is in different obstacle environments.

6. The three-dimensional map quality assessment system according to claim 5, characterized in that: The test module calculates the quality value of the map grid by the following method: When the agent roams to the current map grid for the first time, the quality value of the current map grid is calculated as ,in, Indicates the quality value of the current map grid. Indicates the reward value of each map grid, represents the discount rate, Indicates the maximum number of map grids that the agent can remember.

Images