Increasing gripping rate

Abstract

The invention relates to a method for gripping with 6 degrees of freedom, in particular for gripping, and setting down, using a gripper system, wherein the gripper system comprises at least one fixed depth camera and at least one gripper device, and wherein the method comprises: detecting image data of a scene using the at least one depth camera; determining a depth image based on the image data; determining a 6D pose of the at least one known object based on the depth image; determining a CAD object representation based on the determined 6D pose of the at least one known object and known CAD data of the at least one known object; and determining a gripping probability based on the depth image and the CAD object representation.

Description

[0001] Description

[0002] Increase the gripping rate

[0003] The present invention relates to a method for gripping with 6 degrees of freedom, a system, and a computer program or computer program product.

[0004] Grasping and placing known objects is one of the more well-known problems in robotics. Common solutions for selecting grasps usually consist of three components. In most cases, a vision system is used to determine the spatial pose of the object to be grasped. The vision system is typically supplied with the geometric properties of the object, for example, as a CAD (computer-aided design) model. Based on the vision system, a model can be calculated that outputs one or more spatial poses of instances of the object being sought. This is usually the case in systems where the vision system is attached to a grasping robot.

[0005] Furthermore, a grasp sampling method is typically used, which suggests one or more grasps per object for given gripper and robot properties. These usually involve custom-programmed heuristics that take the geometric properties of the objects into account. However, data-driven methods can also be used. Furthermore, a method for evaluating the suggested grasps can be used to select the grasp with the highest grasping probability.

[0006] It may happen that, especially with fixed cameras, especially with cameras that are not attached to the (movable) gripper, there is no or insufficient depth information that can be used to calculate a grip.

[0007] Furthermore, attaching a camera, especially a depth camera, to the robot is often not possible due to space or safety reasons. The object of the present invention is therefore to improve the gripping rate, preferably to remedy the above-mentioned problem.

[0008] This object is achieved by a method having the features of claim 1. Claim 11 provides a system for operating and / or monitoring a machine, in particular a multi-axis machine under protection and / or a robot. Claim 12 protects a computer program or a computer program product. The subclaims relate to advantageous developments.

[0009] According to one embodiment of the present invention, a method for gripping with 6 degrees of freedom, in particular with the aid of a gripping system, is provided. The (gripping) system comprises at least one fixed depth camera and at least one gripping device. In one embodiment, the method comprises capturing image data of a scene with the at least one depth camera. In one embodiment, the method further comprises determining a depth image based on the image data. In one embodiment, the method further comprises determining a 6D pose of at least one known object based on the depth image. In one embodiment, the at least one known object is an object to be gripped by the gripping system.Furthermore, in one embodiment, the method comprises determining a CAD object representation based on the determined 6D pose of the at least one known object and known CAD data of the at least one known object, in particular additionally based on the depth image and / or a color image, in particular if a color image is available, in particular if the depth camera is configured to capture a color image, in particular image data with color information, or if this is included in the image data. Furthermore, in one embodiment, the method comprises determining a grasp probability based on the depth image and the CAD object representation.

[0010] "Determining a CAD object representation" is to be understood herein as inserting, in particular replacing, a CAD object in place of a (known) object recognized on the basis of the depth image, in particular with the recognized pose. In one embodiment, the determination comprises generating a 3D geometric model of the at least one recognized object, in particular of all objects recognized (on the basis of the depth image), and is in particular additionally based on the depth image. In one embodiment, a CAD object representation can have edge models, surface models, volume models, or the like, in particular a combination of, in particular, identical and / or different model types (edge ​​models, surface models, volume models, or the like).

[0011] "Detected" is preferably understood herein as "if detected." In one embodiment, CAD object representation is preferably understood as (image-based) meshing. A "CAD object representation" as used herein is to be understood in particular as a (virtual) polygon mesh, a (virtual) edge model, a (virtual) surface and / or volume model, or the like. The CAD object representation should be understood in particular as a depth image with CAD object(s), in particular as a depth image with CAD mesh.

[0012] In one embodiment, a gripping device can be a robotic device. Furthermore, in one embodiment, a gripping device can preferably be understood herein as a robot with an end effector, in particular a gripper. In one embodiment, the gripper is designed for workpiece handling and operates, in particular, mechanically, vacuum-based, magnetically, and / or adhesively. A workpiece, in particular an object to be gripped, can be manipulated by the gripping device, in particular by the end effector, in particular via frictional connection, material connection, and / or positive connection.

[0013] "Fixed depth camera" should preferably be understood herein as a depth camera that is fixed or positionally fixed. In one embodiment, the depth camera is not attached to a robot or the movable gripping device. In one embodiment, the depth camera can be mounted at a position from which it can record or capture a scene, in particular a scene with a collection of objects to be grasped.

[0014] A "grasping probability" should preferably be understood herein as the probability of a successful (or unsuccessful) grasp. In one embodiment, the grasping probability can be a number between zero and one, which can in particular include zero and one.

[0015] The method does not apply to purely planar handles. In other words, the method does not apply to vertical handles with degrees of freedom, in particular three degrees of freedom, in two directions, for example, x and y, and a rotation, for example, around z.

[0016] Advantageously, in one embodiment with the rendered CAD object representation based on the CAD data, the method can in particular carry out a CAD comparison (English: “CAD matching”) or include it in the determination of the grasping probability. Thus, in one embodiment, at large angles, in particular at angles of a perspective of a grasping pose on the object to be grasped of more than 45° to the viewing axis of the depth camera, a grasping pose can be recognized (at all) and a grasping probability can be determined accordingly, even if only little depth information is available for the object to be grasped, or the depth image from the perspective of the grasping pose has only little information, in particular less than 30% of depth information compared to depth information where the depth camera is (at least essentially) congruent and in the same direction as the perspective axis of the grasping pose.In particular, this makes it possible to increase the cycle time when determining a grasping probability, and especially during grasping. Furthermore, in one embodiment, the method can select the most promising grasp or decide whether grasping is possible at all.

[0017] According to one embodiment, the method may further comprise determining a color image based on the CAD data. In particular, the method may further comprise determining textures and / or shadows. In one embodiment, the determination of the color image, or of the textures and / or shadows, may be performed using ray tracing. In one embodiment, a color image may preferably also refer to a gray image (with gradations of gray tones). The further description of embodiments or examples with color images should not be understood as limiting.

[0018] According to one embodiment, the depth camera is an RGB depth camera, in particular a depth camera that can capture or provide color information. "RGB" is not to be understood as limiting it to a color space comprising red, green, and blue (RGB). In other words, an RGB depth camera, in one embodiment, can also capture color information in addition to depth information. This color information can be used, in particular, to determine a color image. In one embodiment, the determined color image can include textures and / or shadows.

[0019] According to one embodiment, the method can further comprise: determining a color image, if the at least one depth camera is an RGB depth camera, based on the image data, in particular determining textures and / or shadows in the color image. In one embodiment, the determination of a depth image can additionally be based on the color image. As a result, in one embodiment, more information can advantageously be used to determine a depth image, in particular an improved and / or more detailed depth image can be determined. In one embodiment, a grasping probability can additionally be determined based on the color image. Advantageously, in one embodiment, the color image can provide additional information regarding the grasping pose or the success of the grasping position (grasping probability).This makes it possible, in one embodiment, to determine the success of a grip more accurately and / or better, in particular with greater precision. This makes it possible, in one embodiment, to make the gripping process to be determined and / or controlled by the method more robust.

[0020] According to one embodiment, determining the grasping probability can include stacking the depth image and the depth image with the CAD object representation, in particular also the color image, into a tensor, in particular in such a way that all images have the same height and width. Furthermore, one embodiment includes five channels, in particular two depth channels and three color channels. Alternatively, the depth image and the depth image with the CAD object representation can be rendered in a single image with only one depth channel.

[0021] According to one embodiment, the determination of the depth image, the determination of the CAD object representation, and / or the determination of the color image can be performed from a perspective of at least one predetermined gripping pose of the gripping device on the object to be grasped. In particular, predetermined gripping poses can typically be determined using known methods, preferably by grasp sampling, in a step preceding the method. Advantageously, in one embodiment, at least one predetermined gripping pose is available for an object to be grasped, or at least one predetermined gripping pose is stored for the object to be grasped.In one embodiment, this makes it possible to recognize a grasping pose in the scene, in particular in the depth image and / or the rendered CAD object representation and / or the color image, and to determine or ascertain a grasping probability for the grasping pose from the perspective of the at least one predetermined grasping pose. In other words, a representation of the grip can be determined using the depth image and / or the rendered CAD object representation and / or the color image, from which representation a grasping probability for the grip can be determined (by means of the method). Advantageously, this makes it possible, in particular, to select the most promising grip or to decide whether grasping is possible at all.

[0022] A “perspective of the gripping pose” should preferably be understood herein as a direction that can relate in particular to a longitudinal axis of the gripping device and / or in particular to a longitudinal axis of the end effector of the gripping device or to a central axis or longitudinal axis of a component of the system for gripping, in particular of the system for gripping and placing.

[0023] When grasping with six degrees of freedom, the perspective of the at least one predetermined gripping pose of the gripping device typically does not correspond to the camera perspective of the at least one depth camera. Therefore, for a perspective deviating from the depth camera's line of sight, there is usually less depth information in the image data—in other words, fewer points with depth information in a point cloud—than if a perspective of the gripping pose (at least substantially) coincides with the depth camera's line of sight.Advantageously, in one embodiment, by determining the CAD object representation and / or determining the color image, in particular determining the color image based on CAD data, depth information can be "obtained", which can enable a grasping probability to be determined, even if only little depth information is available for the object to be grasped in the depth image from the perspective of the grasping pose.

[0024] According to one embodiment, the method may further comprise a randomized selection of objects that, in particular, do not relate to the at least one predetermined gripping pose. In one embodiment, the method may comprise a randomized selection of (heavily) hidden objects that, in particular, do not relate to the at least one predetermined gripping pose. For the randomly selected objects, the method may, in one embodiment, comprise that no CAD object representation is determined for these randomly selected objects. In one embodiment, this can be used for training a machine learning algorithm that executes or can execute (at least partially) a method according to the invention.In one embodiment, this advantageously enables the generation of training data for a machine learning algorithm, in particular for an artificial neural network. In one embodiment, this allows for better training of an algorithm that (at least partially) executes the method. In particular, in one embodiment, this advantageously allows for the acceleration of the method.

[0025] According to one embodiment, the method can further comprise gripping the object to be gripped, in particular gripping the object to be gripped with the highest gripping probability, further in particular gripping and placing the object, in particular the object with the highest gripping probability. Advantageously, in one embodiment, this can enable the object to be gripped, in particular the object with the highest gripping probability, to be selected more efficiently and / or more efficiently and, in particular, a process of the gripping system can be accelerated and / or improved. Advantageously, in one embodiment, the method can accelerate a decision for an object to be gripped and, in particular, optimize the efficiency of a gripping sequence for the objects to be gripped.

[0026] According to one embodiment, the acquisition of the image data can be carried out (temporally) between the gripping of a first object (to be gripped) and the gripping of a second object (to be gripped), in particular within the scene. Advantageously, in one embodiment, this makes it possible for an object to be gripped, in particular within the scene, to be determined, in particular in a (temporal) method section in which the scene is not obscured by a gripping system or its gripping device and / or for a process of the gripping system to run more efficiently and / or faster. According to one embodiment, the acquisition of the image data can be carried out while the gripped object is being placed down. In particular, the method can be carried out while the gripped object is being placed down. Advantageously, in one embodiment, this makes it possible for cycle time (of the gripping system) to be and / or to be saved.

[0027] According to one embodiment, the determination of a gripping probability can further be based on known, in particular tabular, data of the gripping system, the gripping device and / or the object to be gripped. In one embodiment, this makes it possible for data, in particular data that is not captured by the at least one depth camera, to be and / or to be incorporated into the determination of a gripping probability. In one embodiment, this tabular data can comprise information about at least one gripping pose, a gripping ellipse, in particular its size, and in particular other available information. Advantageously, this makes it possible for a gripping probability to be determined more accurately and / or better than, in particular, without (known) data.Furthermore, in one embodiment, this can make it possible for a gripping process of the gripping system to be carried out more quickly and / or more accurately.

[0028] In one embodiment, the method can comprise determining images from the perspective of the at least one gripper component. The images from the perspective of the at least one gripper component can be determined or rendered from the point cloud of the depth information (image data), in particular from the depth image, and / or the CAD object representation. This can, in particular, enable a surface structure at the gripping position to be assessed more accurately and / or better, particularly in combination with the depth image and / or the CAD object representation and / or the color image.

[0029] A gripper component should preferably be understood herein as one of the, in particular mechanical, parts of the gripping system that are or can be used for gripping objects. In other words, the gripper component can preferably be understood as one of the active elements or as the at least one active element of the end effector of the gripping device. According to one embodiment, the determination of a gripping probability can further be based on a depth image and / or a CAD object representation from the perspective of (at least) one gripper component. This can, in particular, enable obstacles that could hinder, in particular, the gripping ellipse of the gripping device of the gripping system, in particular the (gripping ellipse of the) gripper component during a (potential) gripping process, to be detected and, in particular, to be incorporated into the determination of a gripping probability.

[0030] According to one embodiment, the method, in particular the determination of a grasping probability, can be carried out by a machine learning algorithm, in particular by an artificial neural network. This can enable the method and / or in particular the determination of a grasping probability to evaluate proposed, in particular predetermined, grips (for a grasping pose) with greater precision.

[0031] According to one embodiment, the method may further comprise evaluating a grasping probability for an object to be grasped lower if little depth information is available for the grasping pose than a grasping probability for an object to be grasped for which a lot of depth information is available. "Little depth information" is to be understood herein as less than 40% depth information from the perspective of the grasping pose, in particular less than 30% depth information, less than 25% depth information, less than 20% depth information, less than 15% depth information, or less than 10% depth information from the perspective of the grasping pose, compared to a grasping pose whose perspective coincides with a viewing axis of the depth camera."A lot of depth information" should be understood herein as more than 25% depth information, in particular more than 40% depth information, more than 50% depth information, more than 75% depth information, or more than 80% depth information compared to a grasping pose whose perspective coincides with a viewing axis of the depth camera, or in particular compared to an object to be grasped from the perspective of the grasping pose with little depth information. In particular, objects to be grasped with little depth information can be excluded from a determination of the grasping probability. This can advantageously enable the method to be carried out more efficiently because, in particular, when determining the grasping probability, a grasping probability needs to be determined for fewer candidates.

[0032] In one embodiment, the method is carried out by a fully convolutional neural network. This makes it possible, in particular, to use larger input images (depth image, CAD object representation image, color image) than with neural networks that are not constructed as a fully convolutional neural network. In one embodiment, a fully convolutional network makes it possible to evaluate images more quickly than, in particular, without a neural network. Furthermore, an embodiment with a fully convolutional neural network can advantageously evaluate as many handles as possible while simultaneously ensuring a high cycle time.

[0033] According to one embodiment of the present invention, a system is provided for operating and / or monitoring a gripping, in particular a gripping and placing, operation of a multi-axis machine, in particular a robot. In one embodiment, the gripping system comprises a gripping device, in particular a multi-axis gripping device. Furthermore, in one embodiment, the gripping system comprises at least one fixed depth camera. In one embodiment, the system is configured to carry out a method as described in the above embodiments and / or comprises:

[0034] means for determining a depth image from the image data;

[0035] means for determining a 6D pose of the at least one known object based on the depth image;

[0036] Means for determining a CAD object representation based on the determined 6D pose of the at least one known object and known CAD data of the at least one known object; means for determining a grasp probability based on the depth image and the CAD object representation.

[0037] In one embodiment, the system or its means comprises: means for determining a color image based on the image data, in particular means for determining textures and / or shadows in the color image, wherein means for determining a grasping probability are additionally configured to determine a grasping probability based on the color image.

[0038] In one embodiment, the system or its means comprises: means for randomly selecting objects, in particular selecting hidden objects that do not relate to the at least one predetermined grasping pose, wherein no CAD object representation is determined for the randomly selected objects

[0039] In one embodiment, the system or its means comprises: means for gripping the object with the highest gripping probability, in particular for gripping and depositing the object with the highest gripping probability. In particular, the means for gripping and, in particular, depositing can be a gripping device.

[0040] In one embodiment, the system or its means comprises: means for determining a gripping probability, wherein the gripping probability is further based on a depth image and / or a CAD object representation from the perspective of a gripper component.

[0041] A system and / or means within the meaning of the present invention can be designed in hardware and / or software, in particular at least one, in particular digital, processing unit, in particular a microprocessor unit (CPU), graphics card (GPU) or the like, preferably connected to a memory and / or bus system for data or signals, and / or one or more programs or program modules. The processing unit can be designed to execute instructions implemented as a program stored in a memory system, to detect input signals from a data bus, and / or to output signals to a data bus. A memory system can have one or more, in particular different, storage media, in particular optical, magnetic, solid-state, and / or other non-volatile media. The program can be designed in such a way that it embodies the methods described here orcapable of carrying out such a method, so that the processing unit can carry out the steps of such methods and thus in particular operate or monitor the machine.

[0042] In one embodiment, a computer program product can comprise, in particular be, a storage medium, in particular a computer-readable and / or non-volatile one, for storing a program or instructions or with a program or instructions stored thereon. In one embodiment, execution of this program or these instructions by a system or a controller, in particular a computer or an arrangement of multiple computers, causes the system or the controller, in particular the computer(s), to carry out a method described here or one or more of its steps, or the program or the instructions are configured to do so.

[0043] In one embodiment, one or more, in particular all, steps of the method are carried out completely or partially automatically, in particular by the controller or its means.

[0044] In one embodiment, the system comprises the robot.

[0045] Further advantages and features emerge from the subclaims and the exemplary embodiments. The following shows, partly schematically:

[0046] Fig. 1 : a system for operating and / or monitoring a gripping, in particular a gripping and depositing, of a multi-axis machine according to an embodiment of the present invention;

[0047] Fig. 2: a scene with schematically represented depth information from the perspective of the grasping pose according to an embodiment of the present invention; Fig. 3: a scene with schematically represented CAD object representation from the perspective of the grasping pose according to an embodiment of the present invention;

[0048] Fig. 4 shows a scene with a schematically illustrated color image from the perspective of the grasping pose according to an embodiment of the present invention;

[0049] Fig. 5 schematically shows a machine learning algorithm according to an embodiment of the present invention; and

[0050] Fig. 6 schematically shows a method according to an embodiment of the present invention.

[0051] Fig. 1 shows a system 1 for operating and / or monitoring a gripping operation, in particular a gripping and placing operation, of a multi-axis machine 2, which is shown as a robot 2 in the schematic representation of Fig. 1. The robot 2 comprises a gripping device 3. Furthermore, a fixed depth camera 4 is shown schematically. The depth camera 4 "observes" a scene that includes several objects 5 to be gripped. The system 1 is designed to determine a gripping probability for at least one of the objects 5 to be gripped based on the image data of the scene 10 acquired by the depth camera 4.

[0052] The system 1 can then, with the aid of the gripping device 3, approach the object 5 to be grasped with a grasping pose and, in particular, grasp the object 5 to be grasped for which a high (compared to the other objects 5 to be grasped), in particular the highest, grasping probability was determined on the basis of the depth image and the CAD object representation (and the color image) using a grip evaluation based on the image data captured by the depth camera 4.

[0053] Additionally or alternatively, if an object 5 to be grasped has been selected, in particular based on its grasping probability, the robot 2 can approach a grasping pose and grasp the object 5 to be grasped. The robot 2 can then manipulate the grasped object and, in particular, place it at a predetermined location or randomly select it from predetermined locations and place the grasped object. Figure 2 schematically shows a scene 30 with schematically illustrated depth information from the perspective of the grasping pose of an object 5 to be grasped. Depth information that can be determined from image data of a depth camera 4 is, as schematically illustrated from the perspective of a grasping pose, usually incomplete and / or (at least partially) not present.For the object 5 to be grasped, it is therefore difficult or impossible to determine from the depth image 30 whether a grasping pose can be approached by the gripping system 1 and whether a gripping device 3 can execute the grip. The gripping position 20 for which a predetermined grasping pose was determined, in particular by means of grasp sampling, is shown in Figure 2 as conditionally accessible. When determining the grasping probability, the (previously known) data on the gripping device 3 are taken into account, so that it can be determined whether a grip at the gripping position 20 is possible, conditionally possible, or not possible.

[0054] Figure 3 schematically shows a scene with CAD object representation, or CAD object representations, for the objects 5 to be grasped from the perspective of the grasping pose. For the corresponding representation, the depth image 30 was used to determine a 6D pose of the objects 5 to be grasped. With this information on the pose of the objects 5 to be grasped, CAD object representations for or with the existing CAD data of the objects 5 to be grasped can then be rendered in the scene 31. In particular, the grasping position 20 can be represented more precisely and the determination of a grasping probability can be carried out on the basis of the depth image and the CAD object representation(s). In this way, a (better) decision can be made as to whether a grasp at the grasping position 20 is possible, conditionally possible, or not possible.

[0055] Figure 4 schematically shows a scene with a color image from the perspective of the gripping pose of the object 5 to be grasped. From the depth image 30, which is shown as an example in Figure 2, a 6D pose can be determined and, in particular, as shown in Figure 3, a scene with CAD object representation(s) 31 can be rendered. With the additional information, which can be captured in particular by an RGB depth camera, a color image 32 can be determined, in particular based on the CAD object representation 31, which includes (additional) information on the colors of the object 5 to be grasped. Based on the first color information 41, the second color information 42, the third color information 43 and / or the fourth color information 44, a surface texture of the object 5 to be grasped, in particular at the gripping position 20, can be determined. This information can be determined on the one hand from the color information 41, 42, 43, 44.The color information 41, 42, 43, 44 of the color image can also be used to represent shadows and / or to draw conclusions about (various) materials of the object 5 to be grasped. Alternatively or additionally, a surface texture of the object 5 to be grasped can be determined from additional information (data) that, in particular, cannot be captured by the RGB depth camera 4. Based on the (additional) information of the depth image 32, in combination with the depth image and the CAD object representation, a grasping probability can then be determined.

[0056] Figure 5 shows a schematic representation of a machine learning algorithm 50. The network here shows three inputs, for example: a depth image, an image rendered with CAD object representation(s), and a color image. These images are processed by the algorithm, which can be an artificial neural network, as shown in the example, and a result is determined in the form of a grasping probability for the object 5 to be grasped.

[0057] Figure 6 schematically shows a method for gripping 60, which comprises steps S100 to S108. The method 60 comprises a step of capturing image data of a scene S100, and determining a depth image based on the image data S102. The method 60 further comprises determining a 6D pose of the at least one known object 5 based on the depth image S104. Furthermore, the method 60 comprises determining a CAD object representation S106 based on the determined 6D pose of the object 5 to be grasped and known CAD data of the known object 5. In step S108, a gripping probability for gripping the object 5 to be grasped is determined based on the depth image determined in step S102 and the CAD object representation determined in step S106.Based on the grasping probability determined in S108, a decision can be made as to whether the object to be grasped should be grasped by the gripping system, in particular the gripping device of the gripping system, or whether another object to be grasped exists that has a higher grasping probability. Subsequently, the system can be prompted, in particular, to grasp the object to be grasped with, in particular, the highest grasping probability. The grasped object can then be placed in a designated or randomized position, in particular in a deposit area.

[0058] Although exemplary embodiments have been explained in the preceding description, it should be noted that a large number of modifications are possible.

[0059] Furthermore, it should be noted that the exemplary embodiments are merely examples and are not intended to limit the scope of protection, applications, or structure in any way. Rather, the preceding description provides the skilled person with a guide for implementing at least one exemplary embodiment. Various modifications, particularly with regard to the function and arrangement of the described components, may be made without departing from the scope of protection as defined by the claims and equivalent combinations of features.

[0060] List of reference symbols

[0061] 1 system

[0062] 2 multi-axis machine, especially robot

[0063] 3 gripping device

[0064] 4 fixed depth cameras

[0065] 5 object to be grasped

[0066] 10 Scene

[0067] 20 gripping position

[0068] 30 Depth image from the perspective of the grasping pose

[0069] 31 CAD object representation from the perspective of the grasping pose

[0070] 32 Color image from the perspective of the grasping pose

[0071] 41 first color information

[0072] 42 second color information

[0073] 43 third color information

[0074] 44 fourth color information

[0075] 50 Machine Learning Algorithms

[0076] 60 methods for gripping

[0077] S100 Capturing image data of a scene

[0078] S102 Determining a depth image S104 Determining a 6D pose

[0079] S106 Determining a CAD object representation

[0080] S108 Determining a grasping probability

Claims

Patent claims Method (60) for gripping with 6 degrees of freedom, in particular for gripping and Depositing, using a gripping system (1), wherein the gripping system (1) comprises at least one fixed depth camera (4) and at least one gripping device (3), wherein the method (60) comprises: - Capture (S100) image data of a scene with at least one depth camera (4); - Determining (S102) a depth image (30) from the image data; - Determine (S104) a 6D pose of at least one known object using the depth image (30); - Determining (S106) a CAD object representation (31) based on the determined 6D pose of the at least one known object (5) and known CAD data of the at least one known object (5); - Determining (S108) a grasping probability based on the depth image (30) and the CAD object representation (31). Method (60) according to the preceding claim, wherein the determination of the CAD object representation (31) is additionally based on the depth image (30). Method (60) according to any of the preceding claims, characterized in that the method (60) further comprises: Determine, if the at least one depth camera (4) is an RGB depth camera, a color image (32) based on the image data, in particular determine textures and / or shadows in the color image (32), wherein the determination (S102) of a depth image is additionally based on the color image (32), or wherein a gripping probability is additionally determined based on the color image (32). Method (60) according to the preceding claim, characterized in that the determination of the depth image (30), the determination of the CAD object representation (31) and / or the determination of the color image (32) is carried out from a perspective of at least one predetermined gripping pose of the gripping device on the object to be gripped.

5. Method (60) according to one of the preceding claims characterized in that the method (60) further comprises: grasping the object with the highest probability of being grasped, in particular grasping and placing the object with the highest probability of being grasped.

6. Method (60) according to one of the preceding claims, characterized in that the acquisition of the image data is carried out between the grasping of a first object and the grasping of a second object.

7. Method (60) according to one of the preceding claims, characterized in that the acquisition of the image data is carried out during a deposit.

8. Method (60) according to one of the preceding claims characterized in that the determination of a grasping probability is further based on known, in particular tabular, data of the grasping system, the grasping device and / or the object to be grasped.

9. Method (60) according to one of the preceding claims, characterized in that the determination of a gripping probability is further based on a depth image (30) and / or a CAD object representation (31) from the perspective of a gripper component.

10. Method (60) according to one of the preceding claims, characterized in that the determination of a grasping probability is carried out by a machine learning algorithm, in particular by an artificial neural network.

11. System (1) for operating and / or monitoring a gripping, in particular a gripping and placing, of a multi-axis machine (2), in particular a robot (2), comprising a gripping device (3), in particular a multi-axis gripping device (3), and at least one fixed depth camera (4), wherein the gripping system (1) is configured to carry out a method (60) according to one of the preceding claims. Computer program or computer program product, wherein the computer program or computer program product, in particular stored on a computer-readable and / or non-volatile storage medium, contains instructions which, when executed by one or more computers or a system (1) according to claim 11, cause the computer(s) or system (1) to perform a method (60) according to any one of claims 1 to 10.

Images