Mobile robot operation decision-making method and device based on large-scale simulation platform

Abstract

The invention provides a mobile robot operation decision-making method and device based on a large-scale simulation platform, and belongs to the technical field of robots, and the method comprises the steps: determining the mobility evaluation information of a robot based on the point cloud information of a target scene and the current position information of the robot; the moving direction and the moving step length of the robot are determined based on the mobility evaluation information of the robot, and moving control is conducted on the robot based on the moving direction and the moving step length; determining operability evaluation information of the robot based on the point cloud information of the target scene and the current position information of the robot in response to movement of the robot to the specified position; and determining an operation position of the robot on the target object based on the operability evaluation information. According to the mobile robot operation decision-making method and device based on the large-scale simulation platform, the mobile operation capability of the robot can be improved.

Description

Technical Field

[0001] This application belongs to the technical field of robotics. More specifically, it relates to a mobile robot operation decision-making method and device based on a large-scale simulation platform. Background Art

[0002] In order to make robots popular in every family and become an indispensable and reliable assistant in human life, robots need to master the ability to operate objects in various indoor environments. Considering the complexity of indoor environments, robots need to have a high level of mobile operation ability. Among them, the mobile operation ability includes, but is not limited to, navigating, identifying, and manipulating various items in a limited space. This is undoubtedly a challenging research topic. Summary of the Invention

[0003] The purpose of this application is to provide a mobile robot operation decision-making method and device based on a large-scale simulation platform to improve the mobile operation ability of robots.

[0004] In the first aspect of the embodiments of this application, a mobile robot operation decision-making method based on a large-scale simulation platform is provided, including: Determining mobility evaluation information of the robot based on the point cloud information of the target scene and the current position information of the robot; the point cloud information of the target scene includes the point cloud information of the target object and the occluder in the target scene; Determining the moving direction and moving step of the robot based on the mobility evaluation information of the robot, so as to perform movement control on the robot based on the moving direction and the moving step; In response to the robot moving to a specified position, determining operability evaluation information of the robot based on the point cloud information of the target scene and the current position information of the robot; the specified position is the position where the value of the mobility evaluation information is the largest; Determining the operation position of the robot on the target object based on the operability evaluation information.

[0005] In the second aspect of the embodiments of this application, a mobile robot operation decision-making device based on a large-scale simulation platform is provided, including: A first evaluation module, configured to determine mobility evaluation information of the robot based on the point cloud information of the target scene and the current position information of the robot; the point cloud information of the target scene includes the point cloud information of the target object and the occluder in the target scene, and the current position information is used to indicate the current position of the robot; A movement control module, configured to determine the moving direction and moving step of the robot based on the mobility evaluation information of the robot, so as to perform movement control on the robot based on the moving direction and the moving step; A second evaluation module, configured to determine operability evaluation information of the robot based on the point cloud information of the target scene and the current position information of the robot in response to the robot moving to a specified position; the specified position is the position where the value of the mobility evaluation information is the largest; An operation control module determines an operation position of the robot on a target object based on the operability evaluation information.

[0006] In a third aspect of the embodiments of the present application, an electronic device is provided, including a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the computer program, the steps of the above-mentioned mobile robot operation decision-making method based on a large-scale simulation platform are implemented.

[0007] In a fourth aspect of the embodiments of the present application, a computer-readable storage medium is provided. The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the above-mentioned mobile robot operation decision-making method based on a large-scale simulation platform are implemented.

[0008] The beneficial effects of the mobile robot operation decision-making method and device based on a large-scale simulation platform provided by the embodiments of the present application are as follows: The embodiments of the present application adopt an innovative method, that is, based on a given target scene, quantify the influence of the target object and the occluder on the movement of the robot at the current position, obtain mobility evaluation information, and according to the mobility evaluation information, the movement direction and movement step of the robot can be determined, so as to provide precise movement guidance for the robot and achieve efficient navigation of the robot.

[0009] On this basis, when the robot reaches the specified position, determine the operability evaluation information of the robot based on the point cloud information of the target scene and the current position information of the robot. The operability evaluation information not only reflects the geometric and functional characteristics of the target object itself, but also comprehensively considers the influence of the robot position and the occluder on the operation feasibility. Therefore, an optimal operation position on the object surface can be determined according to the operability evaluation information, so that the robot can operate the object more stably and improve the mobile operation ability of the robot. Description of the Drawings

[0010] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following will briefly introduce the drawings required for use in the embodiments or the description of the prior art. Obviously, the following drawings are only some embodiments of the present application. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0011] Figure 1Schematic flowchart of a mobile robot operation decision-making method based on a large-scale simulation platform provided by an embodiment of the present application; Figure 2 Block diagram of the structure of the first neural network provided by an embodiment of the present application; Figure 3 Visualization schematic diagram of the "mobile field" provided by an embodiment of the present application; Figure 4 Block diagram of the structure of a mobile robot operation decision-making device based on a large-scale simulation platform provided by an embodiment of the present application; Figure 5 Schematic block diagram of an electronic device provided by an embodiment of the present application. Detailed implementation manners

[0012] In the following description, specific details such as specific system architectures and technologies are presented for purposes of illustration rather than limitation, so as to thoroughly understand the embodiments of the present application. However, those skilled in the art should clearly understand that the present application can also be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted to avoid unnecessary details from obstructing the description of the present application.

[0013] To make the objectives, technical solutions, and advantages of the present application clearer, the following will be described through specific embodiments with reference to the accompanying drawings.

[0014] Please refer to Figure 1 , Figure 1 Schematic flowchart of a mobile robot operation decision-making method based on a large-scale simulation platform provided by an embodiment of the present application. The method includes: S101: Determine the mobility evaluation information of the robot based on the point cloud information of the target scene and the current position information of the robot; the point cloud information of the target scene includes the point cloud information of the target object and the occluder in the target scene.

[0015] In this embodiment, the target object is a specific object that the robot needs to operate (such as grasping, carrying, placing, etc.) in the target scene. The target scene is a specific three-dimensional space environment where the robot executes an operation task. The target scene usually includes the target object and occluders that hinder the movement of the robot. The target scene can be described by point cloud information, which includes the point cloud information of the target object and the occluder.

[0016] By analyzing the point cloud information of the target object and the occluder, and combining the current position information of the robot, the environmental conditions around the robot can be understood, and it can be judged which areas are passable and which areas have obstacles, so as to obtain the mobility evaluation information (which can also be called "mobile field" information), providing a basic judgment basis for the movement control of the robot.

[0017] Specifically, the mobility evaluation information can be characterized by a mobility field score. The higher the mobility field score, the better the mobility.

[0018] S102: Determine the moving direction and moving step length of the robot based on the mobility evaluation information of the robot, so as to perform moving control on the robot based on the moving direction and moving step length.

[0019] In this embodiment, according to the mobility evaluation information, it can be determined in which direction the robot moves to avoid obstacles and thus get closer to the target object. At the same time, an appropriate moving step length can ensure that the robot can reach the target object position efficiently on the premise of safe movement. For example, if the space ahead is open and the mobility is good, a larger step length can be selected to move forward quickly; if there are many obstacles around, the step length needs to be reduced to avoid collisions.

[0020] S103: In response to the robot moving to the specified position, determine the operability evaluation information of the robot based on the point cloud information of the target scene and the current position information of the robot; the specified position is the position where the value of the mobility evaluation information is the largest.

[0021] In this embodiment, during the movement of the robot, the mobility evaluation information (mobility field score) can be calculated in real time. When the mobility field score reaches the maximum value, it indicates that the robot reaches the optimal position (i.e., the specified position) where it can operate the target object.

[0022] After the robot reaches the specified position, the point cloud information of the target scene and the current position information of the robot can be used again to evaluate the feasibility of the robot operating on the target object, such as whether it can accurately grasp the target object and whether there is enough space for operation.

[0023] Specifically, the operability evaluation information can be characterized by an operability score. The higher the operability score, the higher the probability of successful operation.

[0024] S104: Determine the operation position of the robot on the target object based on the operability evaluation information.

[0025] In this embodiment, when the robot reaches the specified position, the optimal operation position on the target object can be determined through the operability evaluation information, which is beneficial for the robot to accurately execute the operation task. Specifically, the surface operation position of the target object with the highest operability score can be selected as the optimal operation position, thereby improving the success rate and accuracy of the robot operation.

[0026] As can be seen from the above, the embodiments of the present application take into account that the core challenge of robot movement operation lies in how to skillfully avoid various obstacles while navigating to the most suitable position for performing operations. Traditional research work usually relies on a detailed understanding of the scene and then derives corresponding movement strategies through complex motion planning control algorithms. This method may work in theory, but in a dynamic and unpredictable environment, it is often difficult to achieve effective generalization.

[0027] Therefore, the embodiments of the present application adopt an innovative method, that is, based on the given target scene, quantify the influence of the target object and obstacles on the movement of the robot at the current position, obtain mobility evaluation information, and determine the movement direction and movement step of the robot according to the mobility evaluation information, so as to provide precise movement guidance for the robot and achieve efficient navigation of the robot.

[0028] On this basis, when the robot reaches the specified position, determine the operability evaluation information of the robot based on the point cloud information of the target scene and the current position information of the robot. The operability evaluation information not only reflects the geometric and functional characteristics of the target object itself, but also comprehensively considers the influence of the robot position and obstacles on the operation feasibility. Therefore, an optimal object surface operation position can be determined according to the operability evaluation information, so that the robot can operate the object more stably and improve the movement operation ability of the robot.

[0029] In an embodiment of the present application, determining the mobility evaluation information of the robot based on the point cloud information of the target scene and the current position information of the robot includes: Input the point cloud information of the target object, the point cloud information of the obstacle, and the current position information of the robot into the first neural network to obtain the mobility evaluation information of the robot.

[0030] In this embodiment, inputting the point cloud information of the target object, the point cloud information of the obstacle, and the current position information of the robot into the first neural network can obtain the mobility evaluation information of the robot.

[0031] Specifically, please refer to Figure 2 , Figure 2 which is the structural schematic diagram of the first neural network provided by an embodiment of the present application. For the target object point cloud and the obstacle point cloud, two independent PointNet++ encoders are respectively used in this embodiment for feature extraction to obtain the features of the target object and the obstacle. Among them, PointNet++ is a deep learning model based on hierarchical point cloud processing, which can effectively capture the geometric structure and context information of the point cloud through local feature aggregation and multi-scale learning. These two encoders do not share weights to ensure that the features of the target object and the obstacle can be learned respectively.

[0032] For the current position information of the robot, in this embodiment, a multi-layer perceptron (MLP) encoder is used to extract features to obtain the robot position features, so as to capture the influence of the position on the operation.

[0033] The features of the target object, the features of the occluder, and the robot position features are concatenated and input into a mobility predictor (which can be implemented by an MLP specifically) to predict the movement field score. The loss function uses the L2 norm loss, and its calculation formula is as follows:

[0034] where M gt is the true movement field score of the current position, M pred is the movement field score of the current position predicted by the first neural network, and N is the number of samples.

[0035] In this embodiment, based on the mobility evaluation information output by the first neural network, "movement field" visualization information can be obtained, as Figure 3 shown. Among them, Figure 3 in, the area indicated by the mark "×" is the area with a lower movement field score, and the area indicated by the mark " " is the area with a higher movement field score.

[0036] It can be concluded from the above that this embodiment comprehensively considers the influences of the target object and all occluding objects, and the obtained mobility evaluation information can provide precise guidance for the robot's movement, ensuring its efficient navigation in a complex space.

[0037] In an embodiment of this application, the mobile robot operation decision-making method based on a large-scale simulation platform further includes: steps of training the first neural network, including: Determining the target movement position information of the robot based on the point cloud information of the target object in the historical data; Determining the first movement field score based on the current position information of the robot and the target movement position information of the robot in the historical data; Determining the second movement field score based on the current position information of the robot in the historical data and the point cloud information of the occluder in the historical data; Determining the third movement field score based on the difference between the first movement field score and the second movement field score; Determining the third movement field score as the true value corresponding to the output value of the first neural network; Constructing sample data based on the current position information of the robot in the historical data, the point cloud information of the occluder in the historical data, the point cloud information of the target object in the historical data, and the true value corresponding to the output value of the first neural network; Training the first neural network based on the sample data.

[0038] In this embodiment, the target object and the occluder can be randomly placed in the simulation environment, the movement operation task can be executed, and the operation results can be recorded to obtain historical data. Specifically, the following information is collected during the data collection process: the observed point cloud S of the current scene (i.e., the point cloud of the target scene, including the point cloud of the target object and the occluder), the current position R of the robot, and the ground truth M(x, y) of the "movement field" at the current position. These data provide comprehensive input and supervision signals for subsequent model training. Among them, each time data is collected, a target operation point is selected on the point cloud of the target object (such as a point on a certain drawer of a table), and based on this point, a search is performed in the target scene to find a position that does not collide with other objects (including the target object and the occluder) and is as close as possible to the target operation point. This position is used as the target movement position of the robot.

[0039] In this embodiment, the ground truth M(x, y) of the "movement field" at the current position comprehensively considers the influences of the target object and all occluding objects. Specifically, based on the current position information of the robot and the target movement position information in the historical data, the first movement field score can be determined, and based on the current position information of the robot and the point cloud information of the occluder in the historical data, the second movement field score can be determined.

[0040] On this basis, considering that the closer the robot is to the target movement position (the final position the robot needs to move to), the closer it is to the target object, and the higher the corresponding movement field score; the closer the robot is to the occluder, the greater the influence of the occluder on the robot's movement, and the lower the corresponding movement field score. Therefore, the influences of the target object and the occluder on the robot's movement are in opposite directions. The difference between the first movement field score and the second movement field score is used as the third movement field score, that is, the ground truth M(x, y) corresponding to the output value of the first neural network.

[0041] Specifically, the third movement field score can be calculated using the second formula, and the second formula is:

[0042] where, represents the third movement field score, represents the first movement field score, represents the second movement field score, represents the influence score of each occluder on the robot's movement.

[0043] On the basis of obtaining the true value corresponding to the output value of the first neural network, sample data can be constructed based on the current position information of the robot in the historical data, the point cloud information of the occluder in the historical data, the point cloud information of the target object in the historical data, and the true value corresponding to the output value of the first neural network, so as to train the first neural network based on the above sample data.

[0044] It can be concluded from the above that when constructing the sample data of the first neural network in this embodiment, the influence of the target object and each occluding object on the movement of the robot is comprehensively considered, which is conducive to obtaining accurate true values, thereby improving the training accuracy of the first neural network.

[0045] In an embodiment of the present application, determining the first movement field score based on the current position information of the robot in the historical data and the target movement position information of the robot includes: Determining a first distance between the current position of the robot and the target movement position of the robot based on the current position information of the robot in the historical data and the target movement position information of the robot; Determining the first movement field score based on the negative correlation between the first distance and the first movement field score.

[0046] In this embodiment, a specific implementation manner for calculating the first movement field score is given. First, a first distance between the current position of the robot and the target movement position of the robot can be obtained based on the current position information of the robot in the historical data and the target movement position information of the robot; then, considering that the farther the first distance between the current position of the robot and the target movement position of the robot is, the farther the distance between the robot and the target object is, and the lower the first movement field score is. Therefore, on the basis of obtaining the first distance, the first movement field score can be determined based on the negative correlation between the first distance and the first movement field score.

[0047] Specifically, the third formula can be used to calculate the first movement field score, and the third formula is:

[0048] where represents the first distance between the current position of the robot and the target movement position of the robot.

[0049] In an embodiment of the present application, determining the second movement field score based on the current position information of the robot in the historical data and the point cloud information of the occluder in the historical data includes: Calculating a fourth movement field score corresponding to each occluder based on the point cloud information of each occluder in the historical data and the current position information of the robot in the historical data; Accumulating the fourth movement field scores corresponding to multiple occluders to obtain the second movement field score; Among them, the calculation process of the fourth movement field score corresponding to any obstacle includes: Determine the second distance between the robot and any obstacle based on the point cloud information of any obstacle and the current position information of the robot in the historical data; If the second distance between the robot and any obstacle is less than or equal to the radius of any obstacle, calculate the fourth movement field score based on the negative correlation between the second distance and the fourth movement field score.

[0050] In this embodiment, a specific implementation method for calculating the second movement field score is given. First, calculate the influence score of each obstacle on the movement of the robot in the historical data, that is, the fourth movement field score, and then accumulate the fourth movement field scores corresponding to multiple obstacles to obtain the second movement field score.

[0051] For any obstacle, first determine the second distance between the robot and any obstacle based on the point cloud information of any obstacle and the current position information of the robot in the historical data; then, considering that when the second distance between the robot and any obstacle is less than or equal to the radius of any obstacle, the influence of the obstacle on the movement of the robot is greater, and the smaller the second distance, the greater the fourth movement field score. Therefore, the fourth movement field score can be calculated based on the negative correlation between the second distance and the fourth movement field score.

[0052] Specifically, the fourth formula can be used to calculate the fourth movement field score, and the fourth formula is:

[0053] Among them, represents the fourth movement field score (that is, the influence score of each obstacle on the movement of the robot), represents the second distance between the robot and any obstacle, represents the radius of any obstacle.

[0054] In an embodiment of the present application, determining the second movement field score based on the current position information of the robot in the historical data and the point cloud information of the obstacle in the historical data includes: Calculate the fourth movement field score corresponding to each obstacle based on the point cloud information of each obstacle in the historical data and the current position information of the robot in the historical data; Accumulate the fourth movement field scores corresponding to multiple obstacles to obtain the second movement field score; Among them, the calculation process of the fourth movement field score corresponding to any obstacle includes: Determine the second distance between the robot and any obstacle based on the point cloud information of any obstacle and the current position information of the robot in the historical data; If the second distance between the robot and any obstacle is greater than the radius of any obstacle, calculate the fourth movement field score based on the negative correlation between the radius of any obstacle and the fourth movement field score.

[0055] In this embodiment, if the second distance between the robot and any obstacle is greater than the radius of any obstacle, the robot is located outside the occlusion radius of the obstacle at this time, and the influence of the obstacle on the movement of the robot is small. At this time, the influence of the obstacle on the movement of the robot can be regarded as a fixed value determined by the radius of the obstacle. Therefore, the fourth movement field score can be calculated based on the negative correlation between the radius of any obstacle and the fourth movement field score.

[0056] Specifically, the fifth formula can be used to calculate the fourth movement field score. The fifth formula is: 。

[0057] It can be concluded from the above that this embodiment selects different calculation methods to calculate the fourth movement field score based on the relative size relationship between the second distance between the robot and any obstacle and the radius of any obstacle, which can more accurately represent the influence of the obstacle on the movement of the robot.

[0058] In an embodiment of the present application, determining the movement direction and movement step of the robot based on the mobility evaluation information of the robot includes: Calculating the gradient value of the mobility evaluation information relative to the current position information of the robot; Determining the movement direction and movement step of the robot based on the gradient value.

[0059] In this embodiment, by calculating the gradient value of the mobility evaluation information relative to the current position information of the robot, the rate and direction of the change of the mobility evaluation information near the current position can be obtained.

[0060] Among them, the direction of the gradient value indicates the direction in which the mobility evaluation information increases fastest. Controlling the robot to move in this direction helps the robot approach the target movement position or avoid obstacles more efficiently. The magnitude of the gradient value represents the degree of change of the mobility evaluation information. The larger the gradient value, the more obvious the change of the mobility evaluation information in the current direction, and the robot can adopt a larger movement step to reach a position with a larger mobility evaluation information faster; on the contrary, when the gradient value is small, in order to avoid excessive movement and falling into an unfavorable position, a smaller movement step can be taken.

[0061] Therefore, this embodiment dynamically adjusts the movement direction and movement step of the robot based on the gradient value, which can ensure the movement efficiency of the robot and improve the safety and accuracy of the robot movement at the same time.

[0062] In an embodiment of the present application, determining the moving direction and moving step length of the robot based on the gradient value includes: Determining the moving direction and moving step length of the robot based on the first formula, where the first formula is:

[0063] Wherein, represents the next moving position information of the robot, represents the current position information of the robot, represents the reference value of the moving step length, represents the change amount of the mobility evaluation information, represents the change amount of the current position information of the robot, represents the gradient value of the mobility evaluation information relative to the current position information of the robot, and the direction of the gradient value indicates the moving direction of the robot, and the magnitude of represents the moving step length of the robot.

[0064] In this embodiment, based on the above first formula, it can be ensured that the robot moves in the optimal direction of the "moving field" in each step, so as to achieve efficient navigation in a complex environment. In addition, by dynamically adjusting the moving step length, the robot can flexibly respond to dynamic changes in the environment, improving its adaptability and operation efficiency.

[0065] In an embodiment of the present application, determining the operability evaluation information of the robot based on the point cloud information of the target scene and the current position information of the robot includes: Determining the first occlusion vector based on the point cloud information of the occluder in the target scene and the current position information of the robot; Determining the second occlusion vector based on the point cloud information of the occluder in the target scene and the point cloud information of the target object; Calculating the occlusion field score based on the first occlusion vector and the second occlusion vector; Selecting the occluded area point cloud information from the point cloud information of the occluder based on the occlusion field score; Inputting the point cloud information of the target object, the occluded area point cloud information and the current position information of the robot into the second neural network to obtain the operability evaluation information of the robot.

[0066] In this embodiment, for robot operation, the occlusion problem is a common and complex challenge. Traditional methods often deal with this problem by inferring and restoring the shape of the occluder, but this method can be computationally very expensive and not flexible enough in a dynamic environment. To more effectively solve the object occlusion problem, this embodiment adopts a novel method, that is, from the perspective of robot operation, focusing on those occluders that directly affect the robot's operation of the target object at the current position. Through this method, this embodiment can accurately extract the part that has a substantial impact on the operation from the entire environment, which can be defined as the "occlusion field".

[0067] Specifically, the calculation of the "occlusion field" can be based on the concept of three-dimensional space, where represents the "occlusion field" from the current position of the robot to the position of the target object , (the first occlusion vector) and (the second occlusion vector) respectively represent the three-dimensional vectors from the points on the occluder to the current position of the robot and the position of the target object . The "occlusion field" (occlusion field fraction) can be calculated by the cross product of vectors:

[0068] Through the above calculation process, the influence of the occluder on the robot operation can be quantified, providing accurate information about the robot's operation environment for the robot.

[0069] Based on the above formula, it can be known that for the points in the three-dimensional space close to the robot position R and the target object position T, F R,T (x, y, z) is close to 0, and the greater the occlusion influence on the robot operation, that is:

[0070] Therefore, the points that are not important for the robot operation (the points with larger FR,T(x, y, z) values) can be removed by setting a threshold, and the points that are important for the robot operation are extracted as the point cloud information of the occlusion area. For example, by sorting the occlusion field fractions corresponding to all points, the 50 points with the lowest occlusion field fractions can be selected as the point cloud of the occlusion area that has the greatest influence on the robot operation in the "occlusion field".

[0071] On this basis, by inputting the point cloud information of the target object, the point cloud information of the occlusion area, and the current position information of the robot into the second neural network, the operability evaluation information of the robot can be obtained.

[0072] The operability evaluation information of the robot can be obtained by using the second neural network. The network structure of the second neural network is the same as that of the first neural network in the above embodiments, and will not be elaborated here.

[0073] Meanwhile, considering that the supervision signal of the second neural network is whether the final operation result is successful (the value is only 0 and 1), the loss function adopts the cross-entropy function, and its calculation formula is as follows:

[0074] where is the true label (1 for success and 0 for failure), is the operation success rate predicted by the model, and N is the number of samples.

[0075] Corresponding to the mobile robot operation decision-making method based on the large-scale simulation platform in the above embodiment, Figure 4 FIG. is a structural block diagram of a mobile robot operation decision-making device based on a large-scale simulation platform provided by an embodiment of the present application. For the sake of convenience of description, only the parts related to the embodiments of the present application are shown. Refer to Figure 4 As shown in, the mobile robot operation decision-making device 20 based on the large-scale simulation platform includes: a first evaluation module 21, a movement control module 22, and an operation control module 23. Among them, the first evaluation module 21 is configured to determine the mobility evaluation information of the robot based on the point cloud information of the target scene and the current position information of the robot; the point cloud information of the target scene includes the point cloud information of the target object and the occluder in the target scene, and the current position information is used to indicate the current position of the robot; The movement control module 22 is configured to determine the movement direction and movement step size of the robot based on the mobility evaluation information of the robot, so as to perform movement control on the robot based on the movement direction and movement step size; The second evaluation module 23 is configured to, in response to the robot moving to a specified position, determine the operability evaluation information of the robot based on the point cloud information of the target scene and the current position information of the robot; the specified position is the position where the value of the mobility evaluation information is the largest; The operation control module 24 determines the operation position of the robot on the target object based on the operability evaluation information.

[0076] In an embodiment of the present application, the first evaluation module 21 is specifically configured to: Input the point cloud information of the target object, the point cloud information of the occluder, and the current position information of the robot into the first neural network to obtain the mobility evaluation information of the robot.

[0077] In an embodiment of the present application, the first evaluation module 21 is further specifically configured to: Train the first neural network, including: Determine the target moving position information of the robot based on the point cloud information of the target object in the historical data; Determine the first movement field score based on the current position information of the robot and the target moving position information of the robot in the historical data; Determine the second movement field score based on the current position information of the robot in the historical data and the point cloud information of the occluder in the historical data; Determine the third movement field score based on the difference between the first movement field score and the second movement field score; Determine the third movement field score as the true value corresponding to the output value of the first neural network; Construct sample data based on the current position information of the robot in the historical data, the point cloud information of the occluder in the historical data, the point cloud information of the target object in the historical data, and the true value corresponding to the output value of the first neural network; Train the first neural network based on the sample data.

[0078] In an embodiment of the present application, the first evaluation module 21 is further specifically configured to: Determine the first distance between the current position of the robot and the target moving position of the robot based on the current position information of the robot and the target moving position information of the robot in the historical data; Determine the first movement field score based on the negative correlation between the first distance and the first movement field score.

[0079] In an embodiment of the present application, the first evaluation module 21 is further specifically configured to: Calculate the fourth movement field score corresponding to each occluder based on the point cloud information of each occluder in the historical data and the current position information of the robot in the historical data; Accumulate the fourth movement field scores corresponding to multiple occluders to obtain the second movement field score; Wherein, the calculation process of the fourth movement field score corresponding to any one occluder includes: Determine the second distance between the robot and any one occluder based on the point cloud information of any one occluder and the current position information of the robot in the historical data; If the second distance between the robot and any one occluder is less than or equal to the radius of any one occluder, calculate the fourth movement field score based on the negative correlation between the second distance and the fourth movement field score.

[0080] In an embodiment of the present application, the first evaluation module 21 is further specifically configured to: Calculate the fourth movement field score corresponding to each occluder based on the point cloud information of each occluder in the historical data and the current position information of the robot in the historical data; Accumulate the fourth movement field scores corresponding to multiple occluders to obtain the second movement field score; Among them, the calculation process of the fourth movement field score corresponding to any occluder includes: Determine the second distance between the robot and any occluder based on the point cloud information of any occluder and the current position information of the robot in the historical data; If the second distance between the robot and any occluder is greater than the radius of any occluder, calculate the fourth movement field score based on the negative correlation between the radius of any occluder and the fourth movement field score.

[0081] In an embodiment of the present application, the movement control module 22 is specifically configured to: Calculate the gradient value of the mobility evaluation information with respect to the current position information of the robot; Determine the movement direction and movement step of the robot based on the gradient value.

[0082] In an embodiment of the present application, the movement control module 22 is further specifically configured to: Determine the movement direction and movement step of the robot based on the gradient value, including: Determine the movement direction and movement step of the robot based on the first formula, and the first formula is:

[0083] Among them, represents the next movement position information of the robot, represents the current position information of the robot, represents the reference value of the movement step, represents the change amount of the mobility evaluation information, represents the change amount of the current position information of the robot, represents the gradient value of the mobility evaluation information with respect to the current position information of the robot.

[0084] In an embodiment of the present application, the second evaluation module 23 is specifically configured to: Determine the first occlusion vector based on the point cloud information of the occluder in the target scene and the current position information of the robot; Determine the second occlusion vector based on the point cloud information of the occluder in the target scene and the point cloud information of the target object; Calculate the occlusion field score based on the first occlusion vector and the second occlusion vector; Select the occluded area point cloud information from the point cloud information of the occluder based on the occlusion field score; Input the point cloud information of the target object, the point cloud information of the occlusion area, and the current position information of the robot into the second neural network to obtain the operability evaluation information of the robot.

[0085] See Figure 5 , Figure 5 which is a schematic block diagram of an electronic device provided by an embodiment of the present application. As Figure 5 shown, the electronic device 300 in this embodiment may include: one or more processors 301, one or more input devices 302, one or more output devices 303, and one or more memories 304. The above-mentioned processors 301, input devices 302, output devices 303, and memories 304 communicate with each other through a communication bus 305. The memory 304 is used to store computer programs, and the computer programs include program instructions. The processor 301 is used to execute the program instructions stored in the memory 304. Among them, the processor 301 is configured to call the program instructions to execute the functions of each module / unit in the above-mentioned device embodiments, such as Figure 4 the functions of the first evaluation module 21, the movement control module 22, the second evaluation module 23, and the operation control module 24 shown.

[0086] It should be understood that in the embodiments of the present application, the so-called processor 301 may be a central processing unit (CPU), and this processor may also be other general-purpose processors, digital signal processors (DSPs), application specific integrated circuits (ASICs), off-the-shelf programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or this processor may also be any conventional processor, etc.

[0087] The input device 302 may include a touchpad, a fingerprint acquisition sensor (for acquiring the fingerprint information and the direction information of the fingerprint of the user), a microphone, etc., and the output device 303 may include a display (such as an LCD), a speaker, etc.

[0088] The memory 304 may include a read-only memory and a random access memory, and provide instructions and data to the processor 301. A part of the memory 304 may also include a non-volatile random access memory. For example, the memory 304 may also store information about the device type.

[0089] In a specific implementation, the processor 301, input device 302, and output device 303 described in the embodiments of the present application may execute the implementation manners described in the first and second embodiments of the mobile robot operation decision-making method based on a large-scale simulation platform provided by the embodiments of the present application, and may also execute the implementation manner of the electronic device described in the embodiments of the present application, which will not be elaborated herein.

[0090] In another embodiment of the present application, a computer-readable storage medium is provided. The computer-readable storage medium stores a computer program, and the computer program includes program instructions. When the program instructions are executed by a processor, all or part of the processes in the above-described method of the embodiment are implemented. It may also be completed by instructing relevant hardware through the computer program. The computer program may be stored in a computer-readable storage medium. When the computer program is executed by the processor, the steps of the above-described method embodiments may be implemented. Among them, the computer program includes computer program code, and the computer program code may be in the form of source code, object code, executable file, or some intermediate form, etc. The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disc, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), electrical carrier signal, telecommunication signal, and software distribution medium, etc.

[0091] The computer-readable storage medium may be an internal storage unit of the electronic device in any of the foregoing embodiments, such as the hard disk or memory of the electronic device. The computer-readable storage medium may also be an external storage device of the electronic device, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc. equipped on the electronic device. Further, the computer-readable storage medium may also include both the internal storage unit and the external storage device of the electronic device. The computer-readable storage medium is used to store the computer program and other programs and data required by the electronic device. The computer-readable storage medium may also be used to temporarily store the data that has been output or will be output.

[0092] Those of ordinary skill in the art can realize that the units and algorithm steps of each example described in combination with the embodiments disclosed herein can be implemented by electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described according to functions in the above description. Whether these functions are executed in a hardware or software manner depends on the specific application and design constraints of the technical solution. Professional technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of this application.

[0093] Those skilled in the art can clearly understand that for the convenience and brevity of description, the specific working processes of the above-described electronic devices and units can refer to the corresponding processes in the foregoing method embodiments and will not be elaborated herein.

[0094] In several embodiments provided in this application, it should be understood that the disclosed electronic devices and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of units is only a logical function division, and there can be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the displayed or discussed couplings or direct couplings or communication connections to each other can be indirect couplings or communication connections through some interfaces or units, or can also be electrical, mechanical, or other forms of connection.

[0095] The units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they can be located in one place or distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of the embodiments of this application.

[0096] In addition, the functional units in each embodiment of this application can be integrated into one processing unit, or each unit can exist physically alone, or two or more units can be integrated into one unit. The above integrated units can be implemented in the form of hardware or in the form of software functional units.

[0097] The above is only the specific implementation manner of this application, but the protection scope of this application is not limited thereto. Any person skilled in the art within the technical scope disclosed in this application can easily think of various equivalent modifications or substitutions, and these modifications or substitutions should all be covered within the protection scope of this application. Therefore, the protection scope of this application should be subject to the protection scope of the claims.

Claims

1. A mobile robot operation decision-making method based on a large-scale simulation platform, characterized in that Including: Determining mobility evaluation information of the robot based on the point cloud information of the target scene and the current position information of the robot; The point cloud information of the target scene includes the point cloud information of the target object and the point cloud information of the occluder in the target scene; Determining the moving direction and moving step of the robot based on the mobility evaluation information of the robot, so as to perform movement control on the robot based on the moving direction and the moving step; In response to the robot moving to a specified position, determining operability evaluation information of the robot based on the point cloud information of the target scene and the current position information of the robot; the specified position is the position where the value of the mobility evaluation information is the largest; Determining the operation position of the robot on the target object based on the operability evaluation information.

2. The mobile robot operation decision-making method based on a large-scale simulation platform according to claim 1, characterized in that, The determining the mobility evaluation information of the robot based on the point cloud information of the target scene and the current position information of the robot includes: Inputting the point cloud information of the target object, the point cloud information of the occluder, and the current position information of the robot into a first neural network to obtain the mobility evaluation information of the robot.

3. The mobile robot operation decision-making method based on a large-scale simulation platform according to claim 2, characterized in that, It further includes: The step of training the first neural network includes: Determining the target moving position information of the robot based on the point cloud information of the target object in the historical data; Determining a first movement field score based on the current position information of the robot and the target moving position information of the robot in the historical data; Determining a second movement field score based on the current position information of the robot in the historical data and the point cloud information of the occluder in the historical data; Determining a third movement field score based on the difference between the first movement field score and the second movement field score; Determining the third movement field score as the true value corresponding to the output value of the first neural network; Constructing sample data based on the current position information of the robot in the historical data, the point cloud information of the occluder in the historical data, the point cloud information of the target object in the historical data, and the true value corresponding to the output value of the first neural network; Training the first neural network based on the sample data.

4. The method for operating decision-making of a mobile robot based on a large-scale simulation platform according to claim 3, characterized in that, Determining a first movement field score based on the current position information of the robot and the target moving position information of the robot in the historical data includes: Determining a first distance between the current position of the robot and the target moving position of the robot based on the current position information of the robot and the target moving position information of the robot in the historical data; Determining the first movement field score based on the negative correlation relationship between the first distance and the first movement field score.

5. The mobile robot operation decision-making method based on a large-scale simulation platform according to claim 3, characterized in that, The determining the second movement field score based on the current position information of the robot in the historical data and the point cloud information of the occluder in the historical data includes: Calculating a fourth movement field score corresponding to each occluder based on the point cloud information of each occluder in the historical data and the current position information of the robot in the historical data; Accumulating the fourth movement field scores corresponding to multiple occluders to obtain the second movement field score; Wherein, the calculation process of the fourth movement field score corresponding to any one occluder includes: Determining a second distance between the robot and the any one occluder based on the point cloud information of the any one occluder and the current position information of the robot in the historical data; If the second distance between the robot and any of the occluders is less than or equal to the radius of any of the occluders, calculate the fourth movement field score based on the negative correlation between the second distance and the fourth movement field score.

6. The mobile robot operation decision-making method based on a large-scale simulation platform according to claim 3, wherein The determining the second movement field score based on the current position information of the robot in the historical data and the point cloud information of the occluders in the historical data includes: Calculate the fourth movement field score corresponding to each occluder based on the point cloud information of each occluder in the historical data and the current position information of the robot in the historical data; Accumulate the fourth movement field scores corresponding to multiple occluders to obtain the second movement field score; Wherein, the calculation process of the fourth movement field score corresponding to any occluder includes: Determine the second distance between the robot and any of the occluders based on the point cloud information of any of the occluders and the current position information of the robot in the historical data; If the second distance between the robot and any of the occluders is greater than the radius of any of the occluders, calculate the fourth movement field score based on the negative correlation between the radius of any of the occluders and the fourth movement field score.

7. The mobile robot operation decision-making method based on a large-scale simulation platform according to claim 1, characterized in that The determining the movement direction and movement step length of the robot based on the mobility evaluation information of the robot includes: Calculate the gradient value of the mobility evaluation information relative to the current position information of the robot; Determine the movement direction and movement step length of the robot based on the gradient value.

8. The mobile robot operation decision-making method based on a large-scale simulation platform according to claim 7, characterized in that, The determining the movement direction and movement step length of the robot based on the gradient value includes: Determine the movement direction and movement step length of the robot based on the first formula, and the first formula is: Among them, represents the next movement position information of the robot, represents the current position information of the robot, represents the reference value of the movement step size, represents the change amount of the mobility evaluation information, represents the change amount of the current position information of the robot, represents the gradient value of the mobility evaluation information relative to the current position information of the robot.

9. The method for operating decision-making of a mobile robot based on a large-scale simulation platform according to claim 1, wherein The determining the operability evaluation information of the robot based on the point cloud information of the target scene and the current position information of the robot includes: Determine the first occlusion vector based on the point cloud information of the occluders in the target scene and the current position information of the robot; Determine the second occlusion vector based on the point cloud information of the occluders in the target scene and the point cloud information of the target object; Calculate the occlusion field score based on the first occlusion vector and the second occlusion vector; Select the occluded area point cloud information from the point cloud information of the occluders based on the occlusion field score; Input the point cloud information of the target object, the occluded area point cloud information and the current position information of the robot into the second neural network to obtain the operability evaluation information of the robot.

10. A mobile robot operation decision-making device based on a large-scale simulation platform, characterized in that, Includes: A first evaluation module for determining the mobility evaluation information of the robot based on the point cloud information of the target scene and the current position information of the robot; The point cloud information of the target scene includes the point cloud information of the target object and the point cloud information of the occluders in the target scene, and the current position information is used to indicate the current position of the robot; A movement control module for determining the movement direction and movement step length of the robot based on the mobility evaluation information of the robot, so as to perform movement control on the robot based on the movement direction and movement step length; A second evaluation module for, in response to the robot moving to a specified position, determining the operability evaluation information of the robot based on the point cloud information of the target scene and the current position information of the robot; the specified position is the position where the value of the mobility evaluation information is the largest; The operation control module determines the operation position of the robot on the target object based on the operability evaluation information.

Images