Mechanical hand gripping posture generation, sample construction method, device, equipment and medium

Abstract

The application discloses a mechanical hand gripping posture generation and sample construction method and device, equipment and medium, and relates to the field of dexterous hand control technology. By inputting the obtained point cloud data of the object set to be gripped into a target gripping posture generation model, the target gripping posture data corresponding to the object set to be gripped can be quickly output, and the whole process takes several seconds or even less than ten seconds. The method for quickly generating the gripping posture can be applied to a real-time robot system. Further, through the target gripping posture data, the mechanical hand can grip multiple objects, and the ability of the mechanical hand to stably grip multiple objects is improved.

Description

Technical Field

[0001] This invention relates to the field of dexterous hand control technology, and in particular to a method, apparatus, device and medium for generating and constructing a robotic hand grasping posture. Background Technology

[0002] With the development of robotic grasping technology, traditional two-finger grasping methods have made great progress, but they have certain limitations in handling complex objects. Therefore, dexterous hand grasping, especially five-finger grasping, has emerged, providing a richer and more diverse range of grasping methods.

[0003] Current technologies require a considerable amount of time (e.g., tens of minutes) to generate the gripping posture of a robotic arm for grasping objects. Clearly, such time-consuming methods are unsuitable for real-time robotic systems. Therefore, there is an urgent need to propose a new method for generating gripping postures. Summary of the Invention

[0004] This invention aims to at least partially solve one of the technical problems in related technologies. To this end, this invention proposes a method, apparatus, device, and medium for generating gripping postures and constructing samples for a robotic arm.

[0005] This invention provides a method for generating a gripping posture for a robotic hand, the method comprising:

[0006] Obtain point cloud data of a set of objects to be grasped; wherein, the set of objects to be grasped includes multiple objects to be grasped;

[0007] The point cloud data of the set of objects to be grasped is input into the target grasping posture generation model to obtain the corresponding target grasping posture data, so that the robotic arm can grasp the set of objects to be grasped.

[0008] This invention provides a method for generating a gripping posture for a robotic hand. The robotic hand is used to grip a set of objects, the set of objects including multiple objects to be gripped, and the surface of the robotic hand has contact points with the objects to be gripped. The method includes:

[0009] The force closure estimation error of the contact point to the contacted object is integrated to obtain the summation error of the force closure of the robot arm to the set of objects; wherein, the contacted object is the object to be grasped at the contact point.

[0010] The initial gripping energy of the robot arm on the object set is determined based on the penetration depth data corresponding to the object set and the force closure summary error.

[0011] Target grip posture data is generated based on the initial grip energy.

[0012] This invention provides a method for constructing object grasping data samples, the method comprising:

[0013] The point cloud data of the object set and the target grasping posture data of the robotic arm grasping the object set are obtained by any of the methods described in the above embodiments.

[0014] Using the point cloud data and the target grasping posture data, an object grasping data sample is constructed for training a grasping posture generation model; wherein, the point cloud data serves as the input to the grasping posture generation model, and the target grasping posture data serves as the label corresponding to the point cloud data.

[0015] This invention provides a device for generating a gripping posture for a robotic hand, the device comprising:

[0016] A point cloud data acquisition module is used to acquire point cloud data of a set of objects to be grasped; wherein, the set of objects to be grasped includes multiple objects to be grasped.

[0017] The grasping posture generation module is used to input the point cloud data of the set of objects to be grasped into the target grasping posture generation model to obtain the corresponding target grasping posture data, so that the robotic arm can grasp the set of objects to be grasped.

[0018] This invention provides a device for generating a gripping posture for a robotic hand. The robotic hand is used to grip a set of objects, the set of objects including multiple objects to be gripped, and the surface of the robotic hand has contact points for gripping the objects. The device includes:

[0019] The force closure error integration module is used to integrate the force closure estimation error of the contact point to the contacted object to obtain the summary force closure error of the robot arm to the set of objects; wherein, the contacted object is the object to be grasped where the contact point is located.

[0020] The initial energy determination module is used to determine the initial gripping energy of the robot arm on the object set based on the penetration depth data corresponding to the object set and the force closure summary error.

[0021] The target posture generation module is used to generate target grip posture data based on the initial grip energy.

[0022] This invention provides an object grasping data sample construction device, the device comprising:

[0023] A point cloud pose acquisition module is used to acquire point cloud data of an object set and target grasping pose data of a robotic arm grasping the object set; wherein the target grasping pose data is obtained by any of the methods described in the above embodiments.

[0024] The grasping sample construction module is used to construct object grasping data samples for training a grasping posture generation model using the point cloud data and the target grasping posture data; wherein, the point cloud data serves as the input of the grasping posture generation model, and the target grasping posture data serves as the label corresponding to the point cloud data.

[0025] The present invention provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of any of the methods described above.

[0026] The present invention provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of any of the methods described above.

[0027] In this invention, by inputting the point cloud data of the acquired set of objects to be grasped into the target grasping posture generation model, the target grasping posture data corresponding to the set of objects to be grasped can be output quickly. The entire process takes only a few seconds or even less than ten seconds. This method of rapidly generating grasping postures can be applied to real-time robotic systems. Furthermore, using the target grasping posture data, the robotic arm can grasp multiple objects, improving its ability to stably grasp multiple objects.

[0028] Additional aspects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0029] Figure 1a A schematic diagram illustrating the generation of a gripping posture based on a target gripping posture generation model provided in one embodiment of this specification;

[0030] Figure 1b This is a schematic diagram of a set of sample objects for training a target grasping posture generation model according to an embodiment of this specification;

[0031] Figure 2 This is a flowchart of a method for generating a gripping posture of a robotic arm according to an embodiment of this specification;

[0032] Figure 3 This is a flowchart illustrating the generation model of obtaining a target gripping posture according to an embodiment of this specification;

[0033] Figure 4a This is a flowchart of obtaining sample grip posture data according to an embodiment of this specification;

[0034] Figure 4bThis is a schematic diagram of sample gripping posture data provided according to an embodiment of this specification;

[0035] Figure 5 This is a flowchart of a method for generating a gripping posture of a robotic arm according to another embodiment of this specification;

[0036] Figure 6 This is a flowchart for determining the penetration depth data according to an embodiment of this specification;

[0037] Figure 7 A flowchart of a method for generating a gripping posture of a robotic arm according to another embodiment of this specification;

[0038] Figure 8 This is a flowchart of a method for constructing object grasping data samples according to an embodiment of this specification;

[0039] Figure 9 This is a schematic diagram of the gripping posture generation device for a robotic arm provided according to an embodiment of this specification.

[0040] Figure 10 This is a schematic diagram of the structure of a gripping posture generation device for a robotic arm according to another embodiment of this specification;

[0041] Figure 11 This is a schematic diagram of the structure of an object grasping data sample construction device provided according to an embodiment of this specification. Detailed Implementation

[0042] Embodiments of the present invention are described in detail below, examples of which are illustrated in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain the present invention, and should not be construed as limiting the present invention.

[0043] In related technologies, multi-fingered dexterous hands, when performing grasping tasks, mostly only consider and can only grasp a single object. For robotic hands that can only grasp a single object, on the one hand, the grasping efficiency needs to be improved, and on the other hand, they cannot support the robot's complex operations, such as playing with walnuts or using chopsticks, thus wasting the kinematic redundancy provided by the multiple fingers of the dexterous hand.

[0044] The main steps involved in using dexterous hands to perform object grasping in related technologies include "dexterous grasp generation". Specifically, given a target object and a specified multi-fingered dexterous hand structure, the robot's posture (position, rotation, joint position) is determined to enable it to grasp the object and ensure that the object can resist external disturbances from any direction within a reasonable range.

[0045] In related technologies, the main problems faced by control algorithms related to dexterous grasping include: 1) Difficulty in planning grasping posture: Multi-fingered dexterous hands have a high degree of freedom (5 fingers usually have 6 spatial degrees of freedom and about 20 joint degrees of freedom), making it infeasible to plan grasping posture using search methods; modeling robot grasping posture planning as an optimization problem will face the highly non-sudden nature of the optimization objective, making it difficult to complete using conventional methods; 2) Lack of grasping data: Also due to the high degree of freedom of multi-fingered dexterous hands, and the fact that they may not have the same morphological structure as human hands, it is very difficult, almost impossible, for humans to manually label grasping datasets.

[0046] Based on this, considering the kinematic redundancy of multi-fingered dexterous hands, which enables them to grasp multiple objects, this specification proposes a method for generating the grasping posture of a robotic hand through a target grasping posture generation model. This generates a grasping posture capable of grasping multiple objects, enabling the application of multi-fingered dexterous hands in multi-object grasping scenarios. For details, please refer to... Figure 1a A target grasping posture generation model is obtained by training the model using a dataset. When multiple objects need to be grasped, the multiple objects are regarded as a set of objects to be grasped. The point cloud data of the set of objects to be grasped is obtained and input into the target grasping posture generation model. The target grasping posture generation model outputs the corresponding target grasping posture data. Based on the target grasping posture data, the robot can grasp the set of objects to be grasped.

[0047] After determining the target grasping posture data, the target grasping posture for the robotic arm to stably grasp the set of objects to be grasped is determined. If it is necessary to move the set of objects to be grasped from the current position to the target position, the movement task of the set of objects to be grasped can continue. The movement task can be divided into two stages. In the first stage, the motion trajectory of the robotic arm approaching and reaching the target grasping posture is determined. In the second stage, the observation space data of the observation space corresponding to the robotic arm after passing the motion trajectory is determined. Based on the observation space data, the movement action of the robotic arm on the set of objects to be grasped is generated to move the set of objects to be grasped to the target position. Among them, the target neural network used in the object lifting stage can be learned through reinforcement learning. The observation space data is input into the target neural network to obtain the movement action of the robotic arm on the set of objects to be grasped. This enables the robotic arm to accurately guide the autonomous execution of the grasping task based on the target grasping posture. For example, taking the lifting of objects as an example, the grasping execution process of the robotic arm is divided into two stages, including: approach grasping stage and lifting stage. During the approach grasping phase, motion planning can be used to determine the approach trajectory from the initial grasping posture to the target grasping posture. It is evident that during the approach grasping phase, the robotic arm has transitioned from the initial grasping posture to a stable target grasping posture, firmly holding the set of objects to be grasped. It can then proceed to the lifting phase, which involves raising the set of objects to be grasped to the target position. A target neural network predicts the corresponding movement based on the observed spatial data. This enables the robotic arm to accurately guide the autonomous execution of the lifting task based on the target grasping posture.

[0048] Furthermore, the training process of the target grasping pose generation model is illustrated by way of example. For details, please refer to [link to relevant documentation]. Figure 1b An initial grasping posture generation model is constructed. Several sample object sets are built, each containing multiple sample objects and corresponding to sample point cloud data. This allows for the generation of sample grasping posture data for each sample object set. A training dataset is built using these sample object sets. This training dataset is then used to train the initial grasping posture generation model. The sample grasping posture data is used as labels, and the sample point cloud data is input into the initial grasping posture generation model for training until the model stops training, resulting in the target grasping posture generation model. In essence, based on the sample grasping posture data, the robotic arm can grasp the sample object set.

[0049] Please see Figure 2 This specification provides a method for generating a gripping posture for a robotic arm, which may include the following steps:

[0050] S210. Obtain point cloud data of the set of objects to be grasped; wherein, the set of objects to be grasped includes multiple objects to be grasped.

[0051] The set of objects to be grasped can be multiple objects stacked together. When the distance between multiple objects needs to be sufficiently close, these multiple objects can be considered as a set of objects to be grasped. If multiple objects are detected and the distance between any two adjacent objects exceeds a preset distance threshold, one of these two adjacent objects can be moved to bring them sufficiently close, and then the set of objects to be grasped is formed based on the detected multiple objects.

[0052] Specifically, in one embodiment, the target grasping posture generation model can be deployed on the controller of the robotic arm. The controller can be connected to sensing units (such as LiDAR, cameras, etc.) to collect data on the set of objects to be grasped, thereby obtaining point cloud data of the set of objects to be grasped. In another embodiment, the point cloud data of the set of objects to be grasped can be obtained from a computer device that is communicatively connected to the controller.

[0053] S220. Input the point cloud data of the set of objects to be grasped into the target grasping posture generation model to obtain the corresponding target grasping posture data so that the robot can grasp the set of objects to be grasped.

[0054] The target grasping posture generation model can be a generative model, a type of machine learning model whose main task is to learn the distribution of data and generate new data similar to the training data. For example, in this embodiment, a conditional diffusion model (DDPM) can be used. The target grasping posture data output by the target grasping posture generation model can include the target position, target orientation, and target joint angle corresponding to the state of the robotic arm grasping the set of objects to be grasped; or, in other words, the posture data corresponding to the grasping posture that matches the shape and placement of the set of objects to be grasped.

[0055] Specifically, after acquiring the point cloud data of the set of objects to be grasped, the acquired point cloud data is input into the target grasping posture generation model. Since the target grasping posture generation model learns the distribution of a large amount of training data, the acquired point cloud data is similar to a large amount of training data. Therefore, the target grasping posture generation model can directly output the corresponding target grasping posture data based on the input point cloud data, so that the robotic arm can grasp the set of objects to be grasped.

[0056] In the above embodiments, by inputting the point cloud data of the acquired set of objects to be grasped into the target grasping posture generation model, the target grasping posture data corresponding to the set of objects to be grasped can be output quickly. The entire process takes only a few seconds or even less than ten seconds. This method of rapidly generating grasping postures can be applied to real-time robotic systems. Furthermore, through the target grasping posture data, the robotic arm can grasp multiple objects, improving its ability to stably grasp multiple objects.

[0057] In some implementations, please refer to Figure 3 The target grip posture generation model is obtained through the following methods:

[0058] S310. Build an initial gripping posture generation model.

[0059] S320. Obtain sample point cloud data of the sample object set and sample gripping posture data of the robotic arm gripping the sample object set; wherein, the sample object set includes multiple sample objects.

[0060] S330. Using sample gripping posture data as labels, input the sample point cloud data into the initial gripping posture generation model for model training to obtain the target gripping posture generation model.

[0061] The initial gripping posture generation model can be a Conditional Diffusion Model. A Conditional Diffusion Model is a type of diffusion model that learns to generate corresponding data based on input conditions, such as generating gripping posture data from input object point cloud data. A diffusion model is a specific type of generative model with good performance. A diffusion model consists of two processes: a forward process and a reverse process, where the forward process is also called the diffusion process. Both the forward and reverse processes are parameterized Markov chains, with the reverse process used to generate data.

[0062] Specifically, the process begins by building an initial grasping posture generation model. Next, a dataset is constructed to train this model. Finally, the user-built dataset is used to train the initial grasping posture generation model until training stops. The constructed dataset includes several sets of sample objects. Each set contains multiple sample objects, and each set corresponds to sample point cloud data and sample grasping posture data. The sample grasping posture data includes the target position, target orientation, and target joint angles corresponding to the state when the robotic arm grasps the sample object set; in other words, the posture data corresponding to the grasping posture that matches the shape and placement of the sample object set.

[0063] To enable the initial gripping posture generation model to learn the data distribution, sample point cloud data from the sample object set is input into the initial gripping posture generation model. This generates predicted gripping posture data corresponding to the sample object set. Based on the predicted gripping posture data output by the initial gripping posture generation model and the sample gripping posture data of the robotic arm gripping the sample object set, the parameters in the initial gripping posture generation model are updated, resulting in an intermediate gripping posture generation model with updated parameters. This process is repeated, inputting the sample object set from the constructed dataset into the intermediate gripping posture generation model for training, until the model stops training when the training condition is met, resulting in the target gripping posture generation model.

[0064] In the above implementation, an initial grasping posture generation model is built, and sample grasping posture data is used as labels. Sample point cloud data is input into the initial grasping posture generation model for model training to obtain a target grasping posture generation model. The target grasping posture generation model can match complex multi-object grasping and can output diverse grasping postures that match the shape, pose, and combination of objects. High-quality grasping can be achieved without complex operational planning, and each object can be kept stable during the grasping process as much as possible.

[0065] In some implementations, please refer to Figure 4a Sample grip posture data were obtained through the following methods:

[0066] S410. Obtain the summation error of the force closure of the robotic arm on the sample object set and the corresponding penetration depth data of the sample object set.

[0067] S420. Determine the initial sample gripping energy of the robot arm on the sample object set based on the penetration depth data and the force closure summary error.

[0068] S430. Generate sample grasping posture data based on the initial sample grasping energy.

[0069] The sample object set includes multiple sample objects. The force closure summary error can be obtained by integrating the force closure errors corresponding to the sample objects in the sample object set, or in other words, the force closure summary error can be the minimum force closure error when the sample object set is stably gripped as a whole by the manipulator. Penetration depth data is used to characterize penetration energy. Penetration depth data can be determined based on the distance function between a point on the manipulator surface and the object surface, the distance function between a point on one surface of two adjacent objects and the other object surface, and the distance function between a point on the manipulator surface and the bearing surface of the sample object set.

[0070] In some cases, when the sample object set is treated as a whole, the force closure summation error of the robot arm on the sample object set varies under different grasping postures, as does the corresponding penetration depth data. Consequently, the robot arm will also have different posture data, each corresponding to different sample grasping energy. By optimizing the robot arm's posture data with the goal of minimizing sample grasping energy, the optimal sample grasping posture data for the robot arm to effectively grasp the sample object set can be determined.

[0071] Specifically, the sample grasping posture data corresponds to the target grasping posture that the robotic arm can grasp the set of sample objects. Before determining the sample grasping posture data, the robotic arm has an initial grasping posture. Under this initial grasping posture, the force closure summation error of the robotic arm on the set of sample objects and the corresponding penetration depth data of the set of sample objects are determined. Based on the force closure summation error and penetration depth data, the initial sample grasping energy of the robotic arm on the set of sample objects can be calculated. In order for the robotic arm to grasp the set of sample objects and to find the minimum sample grasping energy, the initial sample grasping energy is optimized to determine the sample grasping posture data that the robotic arm can grasp the set of sample objects, thereby obtaining the target grasping posture that can grasp the set of sample objects.

[0072] For example, please refer to Figure 4b Based on the penetration depth data and force closure summarization error, sample grasping posture data capable of grasping multiple sample objects can be generated. This includes using... Figure 4b A five-finger robotic hand grasps one ( Figure 4b (columns 1 and 2) and two ( Figure 4b Columns 3 and 4, three ( Figure 4b (Column 5) The pose of the object. This method can theoretically be applied to any robotic arm structure and object.

[0073] In this embodiment, sample grasping posture data is generated by summing the force closure error of the robotic arm on the sample object set and the corresponding penetration depth data of the sample object set. This allows the construction of a dataset for training the initial grasping posture generation model. It should be noted that the method for generating sample grasping posture data in this embodiment takes several tens of minutes, but the quality of the generated sample grasping posture data is reliable. Therefore, using the sample grasping posture data generated in this embodiment to construct a dataset improves the accuracy of the target grasping posture data output by the target grasping posture generation model.

[0074] In some implementations, the surface of the robot arm has contact points with the sample objects. The force closure summary error is determined by integrating the force closure estimation errors of the contacted objects based on the contact points to obtain the force closure summary error of the robot arm on the set of sample objects; wherein the contacted objects are the sample objects where the contact points are located.

[0075] In the field of robotics, force closure estimation error refers to the inability of a robot's manipulator to maintain a force closure state between itself and an object. This means the manipulator cannot apply sufficient contact force to hold the object in place, leading to displacement or slippage during grasping or manipulation. This error affects the robot's accuracy and stability; therefore, reducing force closure estimation error is crucial for improving a robot's grasping and manipulation capabilities.

[0076] Specifically, the sample object set includes multiple sample objects. These sample objects contact the surface of the robotic arm, creating contact points. The sample object containing the contact point is denoted as the contacted object. For any contacted object, there are multiple contact points between the robotic arm and that object. With the aid of friction, the resultant force provided by these multiple contact points can resist external forces or torques in any direction, enabling the robotic arm to stably grasp the contacted object. In this state, there is a force closure estimation error of the contact points on the contacted object. For a sample object set including multiple sample objects, for each contacted object, the force closure estimation error of the contact points on the contacted object needs to be integrated using any of the following methods: equal summation, weighted summation, or product. This yields the total force closure error of the robotic arm on the sample object set, determining whether the robotic arm can stably grasp the sample object set, i.e., resist external forces or torques in any direction.

[0077] For example, for each contact point, each contact point has coordinates, and each contact point has a corresponding normal vector on the surface of the object being contacted. Based on the coordinates of each contact point and the corresponding normal vector, it is possible to estimate whether the object is in a force-closed state, thus obtaining the force closure estimation error.

[0078] In some implementations, the force closure estimation error of the contact point on the contacted object is integrated to obtain the sum of force closure errors of the robot arm on the sample object set. This includes: summing the force closure estimation errors of the contact point on the contacted object to obtain the sum of force closure errors of the robot arm on the sample object set, which is used as the sum of force closure errors.

[0079] In this embodiment, by integrating the force closure estimation error of the contact point to the contacted object, the summative force closure error of the robot arm to the sample object set is obtained. This provides a data basis for accurately determining the gripping energy of the robot arm to grasp the sample object set, and further enables the generation of accurate sample gripping posture data, which is beneficial for training the target gripping posture generation model.

[0080] In some implementations, please refer to Figure 5 Before determining the initial sample grasping energy, the method may further include:

[0081] S510: Obtain joint over-limit data for the robotic arm when the joint angle exceeds the joint limit.

[0082] The initial sample gripping energy of the robot arm on the sample object set is determined based on the penetration depth data and the force closure summation error, including:

[0083] S520: Perform a weighted summation operation on the joint exceeding the limit data, the force closure summary error, and the penetration depth data to obtain the initial sample grasping energy.

[0084] Among them, the joint exceeding the limit data can be understood as the amount by which the robot's joints exceed their angle limits.

[0085] In some cases, to ensure the proper posture of the robotic arm, constraints can be imposed on its pose. For example, this can be achieved by using joint over-limit data. Therefore, it is necessary to obtain joint over-limit data for the robotic arm's angles exceeding joint limits, and determine the initial sample gripping energy based on this data, force closure summation error, and penetration depth data. Specifically, the gripping energy of the gripping posture data H for the set of gripping sample objects O is determined using the following formula:

[0086]

[0087] Where, n o Let S(H) be the number of objects to be grasped in the sample object set O, and let S(H) represent the surface area of ​​the robotic arm. j For the j-th object O on the surface S(H) of the robotic arm j The contact point, E FC (x j O j User represents contact point xj For the j-th object O j Force closure estimation error, E pen (H,O) represents the penetration depth data corresponding to the sample object set O, E q (H) is used to represent joint overshoot data of the robot arm beyond the joint limit angle, λ pen , λ q These are the weighting coefficients.

[0088] In this embodiment, the initial sample gripping energy is obtained by performing a weighted summation operation on the joint exceeding the limit data, the force closure summary error, and the penetration depth data. This allows for the accurate determination of the gripping energy of the robotic arm gripping the sample object set, which is beneficial for generating accurate sample gripping posture data.

[0089] In some implementations, please refer to Figure 6 The penetration depth data is determined in the following ways:

[0090] S610. Determine the first penetration depth between two adjacent sample objects in the sample object set, the second penetration depth between the robot and the sample object, and the third penetration depth between the robot and the bearing surface of the sample object set.

[0091] S620. Determine the molding depth data based on the first molding depth, the second molding depth, and the third molding depth.

[0092] Specifically, the sample object set contains several pairs of adjacent sample objects. The first penetration depth between two adjacent sample objects is calculated. Since the robotic arm grasps the sample object set, penetration also exists between the robotic arm and at least some of the sample objects; therefore, the second penetration depth between the robotic arm and the sample objects is calculated. When the sample object set is placed on a support surface, penetration also exists between the robotic arm and the support surface when the robotic arm grasps the sample object set from the support surface; therefore, the third penetration depth between the robotic arm and the support surface is calculated. The first penetration depth, second penetration depth, and third penetration depth are summed or weighted summed to obtain the penetration depth data. Furthermore, the self-penetrating data of the robotic arm can also be considered; then, the penetration depth data is determined based on the self-penetrating data of the robotic arm, the first penetration depth, the second penetration depth, and the third penetration depth.

[0093] In this embodiment, by determining the penetration depth data based on the first penetration depth, the second penetration depth, and the third penetration depth, a data basis is provided for accurately determining the gripping energy of the robotic arm gripping the sample object set. Furthermore, accurate sample gripping posture data can be generated, which is beneficial for training the target gripping posture generation model.

[0094] In some implementations, the force closure estimation error is determined by: acquiring initial gripping posture data of the robotic arm; and determining the force closure estimation error based on sample point cloud data and the initial gripping posture data.

[0095] In this process, there is a contact surface between the robot and the object being contacted. The target contact point combination is found on the contact surface. The target contact point combination includes several contact points. The coordinates of each contact point in the target contact point combination and the corresponding normal vector can be used to estimate the minimum force closure error, that is, the force closure estimation error of the contact point on the object being contacted.

[0096] Specifically, the robotic arm has initial gripping posture data. Based on the initial gripping posture data and sample point cloud data, a contact surface can be determined between the robotic arm and the object being contacted. Multiple contact points that minimize the force closure error are identified within the contact surface. Based on the coordinates of these contact points and their corresponding normal vectors, the minimum force closure error, i.e., the force closure estimation error, can be estimated.

[0097] In this embodiment, by accurately determining the force closure estimation error of the contact point on the contacted object, an accurate force closure summary error can be obtained, thereby accurately determining the gripping energy provided by the robotic arm to grasp the sample object set.

[0098] In some implementations, generating sample grasping posture data based on initial sample grasping energy includes: optimizing the initial sample grasping energy to obtain a target grasping energy that meets preset energy conditions; and using the grasping posture data corresponding to the target grasping energy as sample grasping posture data.

[0099] Specifically, due to the complexity of the robot's kinematics, the gripping energy of the robot is used as the optimization objective for first-order gradient optimization. To avoid the optimization getting trapped in local optima due to the highly non-convex energy definition, the MALA (Metropolis-adjusted Langevin algorithm) algorithm can be used for additional adjustments.

[0100] In some implementations, GPU parallel computation is used to improve efficiency. For example, parallel optimization can be performed starting from initial gripping energy initialization corresponding to multiple initial gripping poses (e.g., the initialization object is gripping pose H), and results exceeding certain acceptance thresholds can be discarded. For example, an initial gripping pose H can be randomly generated, and the initial gripping pose H can be optimized with the minimum gripping energy to begin convergence to obtain the target gripping pose.

[0101] In some implementations, the target grip posture can be modeled as sampling from a Gibbs distribution. The initial grip energy is optimized using the following formula:

[0102]

[0103] Where O represents the set of sample objects, H represents the gripping posture data used to generate the set of sample objects, p(H|O) represents the probability that the set of sample objects O can be gripped using the gripping posture data H, Z is the normalization constant, e is the natural constant, and E(H,O) represents the gripping energy of the set of sample objects O gripped by the gripping posture data H.

[0104] Please see Figure 7 This specification provides a method for generating a gripping posture for a robotic hand. The robotic hand is used to grasp a set of objects, the set of objects including multiple objects to be grasped, and the surface of the robotic hand has contact points with the objects to be grasped. The method may include the following steps:

[0105] S710. Integrate the force closure estimation error of the contact point to the contacted object to obtain the summation error of the force closure of the robot arm to the set of objects; where the contacted object is the object to be grasped where the contact point is located.

[0106] S720. Determine the initial gripping energy of the robot arm on the object set based on the penetration depth data and force closure summary error corresponding to the object set.

[0107] S730 generates target grip posture data based on initial grip energy.

[0108] The object set includes multiple objects to be grasped. The force closure summary error can be obtained by integrating the force closure estimation errors corresponding to the objects to be grasped in the object set. In other words, the force closure summary error can be the minimum force closure error when the object set is stably grasped as a whole by the manipulator. Penetration depth data is used to characterize penetration energy. Penetration depth data can be determined based on the distance function between a point on the manipulator surface and the object surface, the distance function between a point on one surface of two adjacent objects and the other object surface, and the distance function between a point on the manipulator surface and the bearing surface of the sample object set.

[0109] In some cases, when considering a collection of objects as a whole, the force closure summation error of the robotic arm on the collection of objects varies under different grasping postures, as does the corresponding penetration depth data. Consequently, the robotic arm will also have different posture data, each corresponding to different sample grasping energy. By optimizing the robotic arm's posture data with the goal of minimizing sample grasping energy, the target grasping posture data that the robotic arm can grasp the collection of objects can be determined.

[0110] Specifically, the target gripping posture data corresponds to the target gripping posture that the robotic arm can grasp a set of objects. Before determining the target gripping posture data, the robotic arm has an initial gripping posture. Under this initial gripping posture, the force closure summation error of the robotic arm on the set of objects and the corresponding penetration depth data of the set of objects are determined. Based on the force closure summation error and penetration depth data, the initial gripping energy of the robotic arm on the set of objects can be calculated. In order for the robotic arm to grasp the set of objects and to find the minimum gripping energy, the initial gripping energy is optimized to determine the target gripping posture data that the robotic arm can grasp the set of objects, thereby obtaining the target gripping posture that can grasp the set of objects.

[0111] In this embodiment, target grasping posture data is generated by summing the force closure error of the robotic arm on the object set and the penetration depth data corresponding to the object set. This enables the generation of grasping posture data for multiple objects, and the target grasping posture data matching multiple objects is accurate and reliable. Furthermore, in the scenario of multi-object grasping, the penetration depth data and force closure error corresponding to the object set conform to the complex contact situation between the robotic arm and multiple objects, solving the technical problem that grasping execution methods in traditional technologies almost all fail due to overly coarse planning of dynamics, and enabling each of the multiple objects to remain stable during the grasping process.

[0112] In some implementations, the force closure estimation error of the contact point to the contacted object is integrated to obtain the sum of force closure errors of the robot arm to the set of objects. This includes: summing the force closure estimation errors of the contact point to the contacted object to obtain the sum of force closure errors of the robot arm to the set of objects, which is used as the sum of force closure errors.

[0113] In the field of robotics, force closure estimation error refers to the inability of a robot's manipulator to maintain a force closure state between itself and an object. This means the manipulator cannot apply sufficient contact force to hold the object in place, leading to displacement or slippage during grasping or manipulation. This error affects the robot's accuracy and stability; therefore, reducing force closure estimation error is crucial for improving a robot's grasping and manipulation capabilities.

[0114] Specifically, the object set includes multiple objects to be grasped. These objects contact the surface of the robotic arm, creating contact points. The objects containing these contact points are denoted as the contacted objects. For any contacted object, there are multiple contact points between the robotic arm and that object. With the aid of friction, the resultant force provided by these multiple contact points can resist external forces or torques in any direction, enabling the robotic arm to stably grasp the contacted object. In this state, there is a force closure estimation error between the contact points and the contacted objects. For an object set including multiple objects to be grasped, for each contacted object, the force closure estimation error between the contact points and the contacted object needs to be integrated using any of the following methods: equal summation, weighted summation, or product. This yields the total force closure error of the robotic arm on the object set, determining whether the robotic arm can stably grasp the sample object set, i.e., resist external forces or torques in any direction.

[0115] For example, for each contact point, each contact point has coordinates, and each contact point has a corresponding normal vector on the surface of the object being contacted. Based on the coordinates of each contact point and the corresponding normal vector, it is possible to estimate whether the object is in a force-closed state, thus obtaining the force closure estimation error.

[0116] In this embodiment, by summing the force closure estimation error of the contact point on the contacted object, the summation error of the force closure of the robot arm on the set of objects is obtained, which provides a data basis for accurately determining the gripping energy of the robot arm in grasping the set of objects, and further enables the generation of accurate target gripping posture data.

[0117] In some implementations, before determining the initial gripping energy, the method may further include: acquiring joint over-limit data of the manipulator exceeding the joint limit angle; accordingly, determining the initial gripping energy of the manipulator on the object set based on the mold penetration depth data corresponding to the object set and the force closure summary error may include: performing a weighted summation operation on the joint over-limit data, the force closure summary error, and the mold penetration depth data to obtain the initial gripping energy.

[0118] Among them, the joint exceeding the limit data can be understood as the amount by which the robot's joints exceed their angle limits.

[0119] In some cases, to ensure the proper posture of the robotic arm, certain methods can be used to constrain its posture. For example, this can be achieved by using joint over-limit data. Therefore, it is necessary to obtain joint over-limit data for the robotic arm's angles exceeding joint limits, and determine the initial gripping energy based on this data, force closure summation error, and penetration depth data.

[0120] In some implementations, the gripping energy of the gripping posture data H and the set of objects O is determined by the following formula:

[0121]

[0122] Where, n o Let O be the number of objects to be grasped in the object set O, and S(H) represent the surface of the manipulator. j For the j-th object O on the surface S(H) of the robotic arm j The contact point, E FC (x j O j User represents contact point x j For the j-th object O j Force closure estimation error, E pen (H,O) is used to represent the clipping depth data corresponding to the object set O, E q (H) is used to represent joint overshoot data of the robot arm beyond the joint limit angle, λ pen , λ q These are the weighting coefficients.

[0123] In this embodiment, the initial gripping energy is obtained by performing a weighted summation operation on the joint exceeding the limit data, the force closure summation error, and the penetration depth data, so as to accurately determine the gripping energy of the set of objects grasped by the robot arm, which is conducive to generating accurate target gripping posture data.

[0124] In some implementations, the penetration depth data is determined by: determining a first penetration depth between two adjacent objects to be grasped in the object set, a second penetration depth between the robot and the object to be grasped, and a third penetration depth between the robot and the bearing surface of the object set; and determining the penetration depth data based on the first penetration depth, the second penetration depth, and the third penetration depth.

[0125] Specifically, the object set contains several pairs of adjacent objects to be grasped. The first clipping depth between two adjacent objects to be grasped is calculated. Since the robotic arm grasps the object set, clipping also exists between the robotic arm and at least some of the objects to be grasped; therefore, the second clipping depth between the robotic arm and the objects to be grasped is calculated. When the object set is placed on a support surface, clipping also exists between the robotic arm and the support surface when the robotic arm grasps the object set from the support surface; therefore, the third clipping depth between the robotic arm and the support surface is calculated. The first, second, and third clipping depths are summed or weighted summed to obtain the clipping depth data. Furthermore, the robotic arm's self-clipping data can also be considered; then, the clipping depth data is determined based on the robotic arm's self-clipping data, the first, second, and third clipping depths.

[0126] In this embodiment, by determining the molding depth data based on the first molding depth, the second molding depth, and the third molding depth, a data basis is provided for accurately determining the gripping energy of the robotic arm grasping the set of objects. Furthermore, accurate target gripping posture data can be generated, thereby enabling the robotic arm to stably grasp multiple objects.

[0127] In some implementations, generating target grip posture data based on initial grip energy includes: optimizing the initial grip energy to obtain target grip energy that meets preset energy conditions; and using the grip posture data corresponding to the target grip energy as target grip posture data.

[0128] Specifically, due to the complexity of the robot's kinematics, the gripping energy of the robot is used as the optimization objective for first-order gradient optimization. To avoid the optimization getting trapped in local optima due to the highly non-convex energy definition, the MALA (Metropolis adjusted Langevin algorithm) algorithm can be used for additional adjustments.

[0129] In some implementations, GPU parallel computation is used to improve efficiency. For example, parallel optimization can be performed starting from initial gripping energy initialization corresponding to multiple initial gripping poses (e.g., the initialization object is gripping pose H), and results exceeding certain acceptance thresholds can be discarded. For example, an initial gripping pose H can be randomly generated, and the initial gripping pose H can be optimized with the minimum gripping energy to begin convergence to obtain the target gripping pose.

[0130] In some implementations, the initial gripping energy is optimized using the following formula:

[0131]

[0132] Where O represents the set of objects, H represents the gripping posture data used by the generated set of objects to be gripped, p(H|O) is the probability of gripping the set of objects O with the gripping posture data H, Z is the normalization constant, e is the natural constant, and E(H,O) represents the gripping energy of gripping the set of objects O with the gripping posture data H.

[0133] Please see Figure 8 This specification provides a method for constructing object grasping data samples, which may include the following steps:

[0134] S810: Acquire point cloud data of the object set and target gripping posture data of the robotic arm gripping the object set.

[0135] The target gripping posture data is obtained through any of the methods described in the above embodiments;

[0136] S820: Construct object grasping data samples for training the grasping posture generation model using point cloud data and target grasping posture data.

[0137] The point cloud data serves as the input to the grip posture generation model, and the target grip posture data serves as the label corresponding to the point cloud data.

[0138] For detailed specifications on the method for constructing object grasping data samples, please refer to the description of the method for generating grasping postures of robotic arms above.

[0139] Please see Figure 9 This specification provides a gripping posture generation device 900 for a robotic hand, which includes a point cloud data acquisition module 910 and a gripping posture generation module 920.

[0140] The point cloud data acquisition module 910 is used to acquire point cloud data of a set of objects to be grasped; wherein, the set of objects to be grasped includes multiple objects to be grasped.

[0141] The gripping posture generation module 920 is used to input the point cloud data of the set of objects to be gripped into the target gripping posture generation model to obtain the corresponding target gripping posture data, so that the robotic arm can grip the set of objects to be gripped.

[0142] Please see Figure 10 This specification provides a gripping posture generation device 1000 for a robotic hand. The robotic hand is used to grip a set of objects, the set of objects including multiple objects to be gripped, and the surface of the robotic hand has contact points with the objects to be gripped. The device may include:

[0143] The force closure error integration module 1010 is used to integrate the force closure estimation error of the contact point to the contacted object to obtain the summary force closure error of the robot arm to the set of objects; wherein, the contacted object is the object to be grasped where the contact point is located.

[0144] The initial energy determination module 1020 is used to determine the initial gripping energy of the robot arm on the object set based on the penetration depth data corresponding to the object set and the force closure summary error.

[0145] The target posture generation module 1030 is used to generate target grip posture data based on the initial grip energy.

[0146] Please see Figure 11 This specification provides an object grasping data sample construction device 1100, which may include:

[0147] The point cloud pose acquisition module 1110 is used to acquire point cloud data of an object set and target grasping pose data of a robotic arm grasping the object set; wherein the target grasping pose data is obtained by any of the methods described in the above embodiments.

[0148] The grasping sample construction module 1120 is used to construct object grasping data samples for training a grasping posture generation model using the point cloud data and the target grasping posture data; wherein, the point cloud data serves as the input of the grasping posture generation model, and the target grasping posture data serves as the label corresponding to the point cloud data.

[0149] This specification also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a computer, causes the computer to perform the method described in any of the above embodiments.

[0150] This specification also provides a computer program product containing instructions that, when executed by a computer, cause the computer to perform the methods described in any of the above embodiments.

[0151] It is understood that the specific examples in this document are only intended to help those skilled in the art better understand the embodiments described herein, and are not intended to limit the scope of the invention.

[0152] It is understood that in the various embodiments described in this specification, the sequence number of each process does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments described in this specification.

[0153] It is understood that the various implementation methods described in this specification can be implemented individually or in combination, and the implementation methods in this specification are not limited in this respect.

[0154] Unless otherwise stated, all technical and scientific terms used in the embodiments of this specification have the same meaning as commonly understood by one of ordinary skill in the art. The terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the scope of this specification. The term "and / or" as used in this specification includes any and all combinations of one or more of the associated listed items. The singular forms "a," "the," and "the" as used in the embodiments of this specification and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise.

[0155] It should be noted that the logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device. More specific examples (a non-exhaustive list) of computer-readable media include: an electrical connection having one or more wires (electronic device), a portable computer disk drive (magnetic device), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Alternatively, the computer-readable medium may be paper or other suitable media on which the program can be printed, since the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in a computer memory.

[0156] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0157] In the description of this specification, references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0158] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this invention, "a plurality of" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0159] In this invention, unless otherwise explicitly specified and limited, the terms "installation," "connection," "linking," and "fixing," etc., should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral part; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication of two components or the interaction between two components, unless otherwise explicitly limited. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0160] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A method for generating a gripping posture for a robotic hand, characterized in that, The method includes: Obtain point cloud data of a set of objects to be grasped; wherein, the set of objects to be grasped includes multiple objects to be grasped; The point cloud data of the set of objects to be grasped is input into the target grasping posture generation model to obtain the corresponding target grasping posture data, so that the robotic arm can grasp the set of objects to be grasped. Obtaining the target grip posture generation model includes: obtaining sample grip posture data of a sample object set; The process of obtaining the sample grasping posture data includes: obtaining the force closure summary error of the robotic arm on the sample object set and the penetration depth data corresponding to the sample object set; The penetration depth data is determined in the following manner: Determine the first penetration depth between two adjacent sample objects in the sample object set, the second penetration depth between the robot and the sample object, and the third penetration depth between the robot and the bearing surface of the sample object set; The molding depth data is determined based on the first molding depth, the second molding depth, and the third molding depth.

2. The method according to claim 1, characterized in that, The target grip posture generation model is obtained through the following methods: Build an initial grip posture generation model; Acquire sample point cloud data of a sample object set, and sample grasping posture data of the robotic arm grasping the sample object set; wherein, the sample object set includes multiple sample objects; Using the sample gripping posture data as labels, the sample point cloud data is input into the initial gripping posture generation model for model training to obtain the target gripping posture generation model.

3. The method according to claim 2, characterized in that, The sample grip posture data was obtained using the following methods: Obtain the summation error of the force closure of the robotic arm on the sample object set and the penetration depth data corresponding to the sample object set; The initial sample gripping energy of the robotic arm on the sample object set is determined based on the penetration depth data and the force closure summation error. The sample gripping posture data is generated based on the initial sample gripping energy.

4. The method according to claim 3, characterized in that, The surface of the robotic arm has contact points with the sample object; The force closure summary error is determined in the following manner: Based on the force closure estimation error of the contact point on the contacted object, the summative force closure error of the manipulator on the sample object set is obtained; wherein, the contacted object is the sample object where the contact point is located.

5. The method according to claim 4, characterized in that, The force closure estimation error based on the contact point of the contacted object is integrated to obtain the summative force closure error of the manipulator on the sample object set, including: The summation of the force closure estimation error of the contact point to the contacted object is performed to obtain the sum of the force closure errors of the manipulator to the sample object set, which is used as the summation error of the force closure.

6. The method according to claim 3, characterized in that, Before determining the initial sample gripping energy, the method further includes: Obtain the joint over-limit data of the robotic arm when the joint angle exceeds the joint limit; The step of determining the initial sample gripping energy of the robotic arm on the sample object set based on the penetration depth data and the force closure summary error includes: The initial sample gripping energy is obtained by performing a weighted summation operation on the joint exceeding the limit data, the force closure summary error, and the penetration depth data.

7. The method according to claim 4, characterized in that, The force closure estimation error is determined in the following manner: Obtain the initial gripping posture data of the robotic arm; The force closure estimation error is determined based on the sample point cloud data and the initial grip posture data.

8. The method according to claim 3, characterized in that, The process of generating the sample gripping posture data based on the initial sample gripping energy includes: The initial sample gripping energy is optimized to obtain a target gripping energy that meets the preset energy conditions; The gripping posture data corresponding to the target gripping energy is used as sample gripping posture data.

9. A method for generating a gripping posture for a robotic hand, characterized in that, The robotic arm is used to grasp a set of objects, the set of objects including multiple objects to be grasped, and the surface of the robotic arm has contact points with the objects to be grasped. The method includes: The force closure estimation error of the contact point to the contacted object is integrated to obtain the summation error of the force closure of the robot arm to the set of objects; wherein, the contacted object is the object to be grasped at the contact point. The initial gripping energy of the robot arm on the object set is determined based on the penetration depth data corresponding to the object set and the force closure summary error. Target gripping posture data is generated based on the initial gripping energy; The penetration depth data is determined in the following manner: Determine the first penetration depth between two adjacent objects to be grasped in the object set, the second penetration depth between the robot and the object to be grasped, and the third penetration depth between the robot and the bearing surface of the object set; The molding depth data is determined based on the first molding depth, the second molding depth, and the third molding depth.

10. The method according to claim 9, characterized in that, The force closure estimation error of the contact point on the contacted object is integrated to obtain the summative force closure error of the robot arm on the set of objects, including: The summation of the force closure estimation error of the contact point to the contacted object is performed to obtain the sum of the force closure errors of the manipulator to the set of objects, which is used as the summation error of the force closure.

11. The method according to claim 9, characterized in that, Before determining the initial gripping energy, the method further includes: Obtain the joint over-limit data of the robotic arm when the joint angle exceeds the joint limit; The step of determining the initial gripping energy of the robotic arm on the object set based on the penetration depth data corresponding to the object set and the force closure summary error includes: The initial gripping energy is obtained by performing a weighted summation operation on the joint exceeding the limit data, the force closure summary error, and the penetration depth data.

12. The method according to any one of claims 9-11, characterized in that, The generation of target grip posture data based on the initial grip energy includes: The initial gripping energy is optimized to obtain a target gripping energy that meets the preset energy conditions; The gripping posture data corresponding to the target gripping energy is used as the target gripping posture data.

13. The method according to claim 12, characterized in that, The initial gripping energy is optimized using the following formula: in, Used to represent a collection of objects. This data represents the gripping posture used to generate the set of objects being grasped. () is used to represent grasping posture data. Collection of objects to grasp The probability, It is a normalization constant. It is a natural constant. Used to represent grip posture data Collection of objects to grasp Grasp the energy.

14. The method according to claim 13, characterized in that, The grip posture data is determined using the following formula. Collection of objects to grasp Grasping energy: in, For a collection of objects The number of objects to be grasped. Used to represent the surface of the robotic arm. The surface of the robotic arm The above is aimed at the first an object The contact point, User-represented contact points For the an object Force closure estimation error, Used to represent a collection of objects The corresponding molding depth data, Joint over-limit data used to indicate that the robotic arm exceeds the joint limit angle. , These are the weighting coefficients.

15. A method for constructing object grasping data samples, characterized in that, The method includes: The method involves acquiring point cloud data of a set of objects and target grasping posture data of a robotic arm gripping the set of objects; wherein the target grasping posture data is obtained by the method of any one of claims 9-14. Using the point cloud data and the target grasping posture data, an object grasping data sample is constructed for training a grasping posture generation model; wherein, the point cloud data serves as the input to the grasping posture generation model, and the target grasping posture data serves as the label corresponding to the point cloud data.

16. A device for generating a gripping posture for a robotic hand, characterized in that, The grip posture generation device is used to implement the grip posture generation method as described in any one of claims 1-8, and the device includes: A point cloud data acquisition module is used to acquire point cloud data of a set of objects to be grasped; wherein, the set of objects to be grasped includes multiple objects to be grasped. The grasping posture generation module is used to input the point cloud data of the set of objects to be grasped into the target grasping posture generation model to obtain the corresponding target grasping posture data, so that the robotic arm can grasp the set of objects to be grasped.

17. A device for generating a gripping posture for a robotic hand, characterized in that, The gripping posture generation device is used to implement the gripping posture generation method as described in any one of claims 9-14, the robotic arm is used to grip a set of objects, the set of objects includes a plurality of objects to be gripped, and the surface of the robotic arm has contact points with the objects to be gripped. The device includes: The force closure error integration module is used to integrate the force closure estimation error of the contact point to the contacted object to obtain the summary force closure error of the robot arm to the set of objects; wherein, the contacted object is the object to be grasped where the contact point is located. The initial energy determination module is used to determine the initial gripping energy of the robot arm on the object set based on the penetration depth data corresponding to the object set and the force closure summary error. The target posture generation module is used to generate target grip posture data based on the initial grip energy.

18. A device for constructing object grasping data samples, characterized in that, The device includes: A point cloud pose acquisition module is used to acquire point cloud data of an object set and target grasping pose data of a robotic arm grasping the object set; wherein the target grasping pose data is obtained by the method of any one of claims 9-14; The grasping sample construction module is used to construct object grasping data samples for training a grasping posture generation model using the point cloud data and the target grasping posture data; wherein, the point cloud data serves as the input of the grasping posture generation model, and the target grasping posture data serves as the label corresponding to the point cloud data.

19. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1-15.

20. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1-15.

Images