WO2018193130A1

WO2018193130A1 - Method for creating a database of gripper poses, method for controlling a robot, computer-readable storage medium and materials handling system

Info

Publication number: WO2018193130A1
Application number: PCT/EP2018/060264
Authority: WO
Inventors: Maximo Alejandro ROA GARZON; Monika FLOREK-JASINSKA; Christian Emmerich; Elena Francesca GAMBARO; Annett STELZER; Michael Suppa
Original assignee: Roboception Gmbh
Priority date: 2017-04-21
Filing date: 2018-04-23
Publication date: 2018-10-25
Also published as: EP3612355A1

Abstract

It is difficult for a robot to grip an object. More particularly, the problem of rapidly determining suitable gripper poses for gripping objects has not previously been adequately resolved. The problem addressed by the invention is therefore that of specifying a method which simplifies gripping of objects by a robot. The problem is solved by a method for creating a database (40) of gripper posses, comprising the following: a. Storing, particularly by recording, at least one training gripper pose (P, P2, P4) for gripping a training object (1), particularly with regard to a training object coordinate system (2) assigned to the training object (1); b. Generating at least one further gripper pose (P', P'') for the training object (1) using at least one object prototype (30, 30', 30'') and the training gripper pose (P, P2, P4); c. Storing the at least one further gripper pose (P') and the training gripper pose (P, P2, P4) with reference to the training object (1) in a database (40).

Description

Method for creating a database with gripper poses, method for controlling a robot, computer-readable storage medium and

handling system

description

The invention relates to a method for creating a database with

Gripper poses, a method for controlling a robot, a corresponding computer-readable storage medium and a handling system.

Modern handling systems, such as multi-axis robots, serve to transport objects from a first location to a second location. For this purpose it is necessary that the robot grips the object by means of a gripper and holds it securely so that it does not fall down during the transport. In structured work environments, such as As is common in automotive production, the objects to be gripped are always in the same place with the same orientation, so that a gripping needs to be programmed only once and then can be executed repeatedly.

However, in unstructured workspaces, the objects may be on

different locations with different orientation occur, so that a robot must be able to adapt the gripping of the objects. An example is the gripping of various parts from a box (so-called "bin-picking").

To enable gripping of objects in unstructured work environments, robotic systems have perception systems, e.g. B. on stereo camera systems, the detection of the objects to be gripped

enable. Furthermore, the dimensions and the orientation of the object to be gripped can be determined by means of these perception systems, so that the points of application for the gripper can be determined.

Before an object can be grasped, the robot can be moved so that only one gripper can be activated. The orientation and location of the gripper can be described by a pose. A pose is as one Representation comprising three spatial coordinates and three angles, which describes the position and orientation of a second coordinate system with respect to a reference coordinate system.

When gripping a variety of difficulties occur.

For example, the mass distribution of the object to be gripped by means of a visual perception system is usually not recognizable. It can

Occur that the gripping of the object in certain places is not possible or very difficult. In addition, it can not be determined by means of a purely visually operating perception system with which force an object must be held. In addition, calculating possible gripper poses requires a long computation time and computational resources of a computer.

In the publication

Porges 0., Stouraitis T., Borst C, Roa M.A. (2014) "Reachability and Capability Analysis for Manipulation Tasks". In: Armada M., Sanfeliu A., Ferre M. (eds) ROBOT2013: First Iberian Robotics Conference. Advances in Intelligent Systems and Computing, vol 253. Springer, Cham describes the analysis of the grasping abilities of a robot before performing an actual gripping. For this purpose, a skill map is created for the robot and the gripper used

Working space of the robot indicates how many directions a point in the

Working space can be approached by the robot. The gripper as well as the kinematics of the robot are taken into account. The skill card is created independently of the objects to be gripped. The named

Publication further describes that a database can be used which contains information on possible gripping actions of known objects. A disadvantage of the procedure described in the cited publication is that the creation of the database is very complicated. It is not sufficient to store only a single gripper pose for each object. Rather, a variety of gripper poses must be manually recorded and stored in the database so that the gripping of an object can be carried out safely. Based on the described prior art, it is the task of

Invention to provide a method for creating a database with gripper poses, which addresses the above-mentioned disadvantages. In particular, the efficient filling of a database is to be made possible. In addition, a method for controlling a robot is to be specified, which allows an efficient control of the robot. In addition, a corresponding computer-readable storage medium should be specified. In addition, a handling system is to be specified, which allows the efficient gripping of an object.

The object is achieved by a method for creating a database with gripper poses according to claim 1.

In particular, the object is achieved by a method for creating a database with gripper poses comprising: a. Save, in particular by recording, at least one

Trainingsgreiferpose for gripping a training object, in particular with respect to a training object associated

Training object coordinate system;

b. Generate at least one other gripper pose for the training object using at least one object prototype and the

Trainingsgreiferpose;

c. Saving the at least one further Greiferpose and the

Trainingsgreiferpose with a reference to the training object in the database, in particular in an object database.

One core of the invention is therefore that one or a plurality of further gripper poses are generated for a single training gripper pose. So only the training gripper pose has to be set manually. It is particularly advantageous that the further gripper poses are generated using object prototypes. Thus, the properties of the object prototypes can be used to determine the further gripper poses. The described method therefore allows the efficient storage of at least one further gripper pose in addition to the training gripper pose.

The object is further achieved by a method for controlling a robot according to claim 2. In particular, the object is achieved by a method for controlling a robot, comprising:

Performing the steps of the method of creating a database described above;

- querying at least one pose from the object database;

- controlling the robot using the at least one pose.

Querying and using a gripper pose to control a robot from a database as described above enables

overall particularly efficient gripping method.

In one embodiment, a pose, particularly the training gripper pose, may be represented as a six-tuple having three spatial coordinates and three angles relative to a training object associated with the training object

Training object coordinate system has.

A pose can be stored particularly efficiently in a data structure that indicates the position and orientation with respect to the training object. By storing the information relating to the training object, a gripper pose can be used, regardless of where the object to be gripped is in space or in a robot coordinate system.

In one embodiment, the method may include selecting the at least one object prototype from a set of candidate prototypes.

The set of candidate prototypes may include a cylinder, a sphere, a box, a cone, a truncated cone, and / or a pyramid.

The at least one object prototype can thus be a geometric basic form, whereby a very flexible handling of the described method is possible. The object prototypes may be described by geometric parameters such as length, depth, width, height, and / or radius. The known geometry of the object prototype can then be utilized for automatic generation of the at least one further gripper pose. In one embodiment, at least one symmetry axis may be associated with the at least one object prototype, wherein a plurality of gripper poses may be generated taking into account the at least one symmetry axis, in particular by rotation about the at least one symmetry axis, preferably using a transformation matrix, wherein the gripper poses are respectively stored can.

It is particularly advantageous if a known symmetry axis of the

Object prototype is used to generate the variety of gripper poses. In this case, it is utilized that, for example, a rotationally symmetrical body can be gripped equally well using a plurality of poses. As a result, a large number of possible gripper poses can be generated, which deliver similarly good gripping results. This greatly simplifies the construction of an object database. An object prototype may be in a CAD data format, such as. IGES or STL. The CAD data format can directly or indirectly specify an axis of symmetry and / or a plane of symmetry that can be used to generate the further gripper poses.

In one embodiment, the training gripper pose and the further at least one gripper pose may each be stored with an object identification number that may be associated with the training object.

The assignment of the training gripper pose and the further at least one gripper pose to an object identification number makes it possible to find a plurality of possible poses for gripping an object, if the

Object identification number is known. This makes it possible to get out of the

Object database to query a variety of possible gripper poses at once for a single object.

In one embodiment, a storing of gripping parameters can be provided together with the poses in an object database, in particular gripper information, a gripping mode, a quality indication, a force, a gripping strategy and / or a grip description.

It is therefore possible that, in addition to the mentioned gripper poses, further data are stored in the object database, which can preferably be assigned to individual poses. This can be data that specifies the grip closer. For example, may be indicated by a force with which force an object should be grasped. Furthermore, a gripping type can be stored. In a multi-finger gripper z. For example, a finger is used to grasp an object only with the fingertips of the multi-finger gripper.

Gripper information may include information indicating a gripper type or even a particular gripper product.

Gripper information can be stored in a gripper table or gripper database. So it can be a reference to the poses

corresponding entry in the gripper table or gripper database are stored. The corresponding entry may also include a CAD model of the gripper or performance parameters, such as a maximum application force. Furthermore, at least one rule may be provided which indicates that for each pose

at least one gripper must be assigned. The rule can z. B. implemented as part of the object database.

A gripping strategy can specify how an object should be gripped.

For example, it can be provided that initially a high force is accessed, and after a certain time with a lower force. A quality indication may include indications as to whether a reliable gripping is possible, i. whether in experiments it turned out that one

appropriate gripper pose is particularly well suited. The saving of the gripping parameters has the advantage that gripping an object is easier

can be reproduced or executed.

In one embodiment, the method may include identifying the

Training object comprise, in particular by means of a perception system, for. B. by means of a stereo camera.

It is also possible that the training object is automatically identified, so that information about the training object automatically in the

Object database can be written. This makes the process even more efficient.

In one embodiment, the method may include storing

Object features for recognizing the training object comprise, in particular geometric information, such as corners and / or edges of the Training object, preferably such that poses are attributable to the object features. Object features may also include visual information. These include image descriptors, such as color and color intensity gradients,

Color values or gray values. Also image excerpts and other

Texture information can be stored as object features.

In order to better locate an object to be gripped in the object database, a description of the training object can be stored in the database. This description may be implemented by object features. In this case, the position and orientation of corners and / or edges can be stored so that a perception system can find and / or identify an object on the basis of the stored object features. The object features may be stored so as to be associated with the stored gripper poses. This assignment can be implemented directly or indirectly. For example, all poses may be stored in a first database table. All objects can be stored in a second database table and all object features can be stored in a third database table. An assignment can then be executed using corresponding primary keys and references in the tables.

In one embodiment, a variety of detection methods

be provided, with a Perzeptionsverfahren for automatic

Identification of the training object using stored

Object features can be selected.

It is therefore possible that, based on the object features present in the object database, a method for identifying the training object is selected. Are z. B. stored in the object database object features that affect image sections, a Pezeptionsverfahren is selected, which uses image sections for the identification of objects.

An object can be identified in such a way that first of all a hypothesis is drawn as to which object is in the view of a perception system.

Subsequently, it can be checked whether the hypothesis is correct. This means that successively a variety of known objects can be tried.

In one embodiment, the method may include: an identification of a target object to be gripped, in particular of an object identification number assigned to the target object;

- determining a Zielgreiferpose that is attributable to the target object; and

gripping the target object using the target gripper pose,

in particular by an end effector, e.g. B. a parallel gripper, a robot.

The described embodiment has the advantage that the gripping of an object by the robot is simplified. After identifying the object to be gripped, using an associated

Object identification number a Zielgreiferpose be determined.

In an embodiment, the determining the reading may comprise a plurality of candidate poses, in particular from an object database, wherein the reading may be performed in consideration of a reachability map and / or a capability map of a robot.

The described embodiment has the advantage that from the plurality of candidate poses a Zielgreiferpose can be selected, which leads to particularly good gripping results for a particular robot. This can be ensured via the reachability map and / or the ability map. A reachability map indicates which points in the room can be reached by the robot taking into account its kinematics. A skill card indicates how many directions the robot can reach to a point within its working space. Overall, with the described

Embodiment, the gripping of a target object safer and easier.

In an embodiment, the determining may include sorting one or more candidate poses, wherein the sorting may be performed considering the skill card.

It is advantageous if a sorting of the plurality of candidate poses

is performed so that an order arises. This allows the

Candidate poses are tried one after the other. Represents, for example out that a first candidate pose is not appropriate, then proceed to a second candidate pose.

In an embodiment, the method may include motion planning to move a robot from an initial configuration to a robot

Target configuration, wherein the target configuration may be associated with a Zielgreiferpose.

Following the sorting can be done using the first

Candidate pose a movement planning to be performed. The first

Candidate pose corresponds to the pose that has been identified as the most suitable pose during sorting. In this case, it can be established, for example, that the most suitable pose, taking into account a current robot configuration, for example due to obstacles, can not be achieved or insufficiently well. Therefore, a loop of selecting a target gripper pose and motion planning can be executed. This ensures that exactly that Zielgreiferpose is used, which is most suitable for the robot at the current time.

The object is further achieved by a computer-readable storage medium according to claim 15.

In particular, the problem is solved by a computer-readable

A storage medium containing instructions that cause at least one processor to implement a method as claimed in any one of the preceding claims when the instructions are executed by the at least one processor.

This results in similar or identical advantages as already in the

In connection with the methods described above have been described.

The object is further achieved by a handling system according to claim 16.

In particular, the object is achieved by a handling system comprising:

an object database, in particular an object database, after

one of the previously described methods is created; a robot adapted to grip an object;

a perception system for detecting an object which is arranged in particular on the robot; wherein the robot is adapted

using the detected object, a plurality of

Query candidate poses from the object database;

to filter the multiplicity of candidate poses, in particular under

Use of a reachability map and / or ability map to determine a target grab pose

- Set a robot configuration according to the Zielgreiferpose.

This results in similar or identical advantages as already in the

Related to the method described above are described.

Further advantageous embodiments will be apparent from the

Dependent claims.

The invention will now be described by means of several embodiments

described in more detail with reference to figures. Here: NEN robot 10 in an initial configuration before grasping an object 1;

Fig. 2a object 1 and a gripper 20 in a Anfahrtspose;

Fig. 2b shows the object 1 of Fig. 2a and the gripper 20 of Fig. 2a in one

Greiferpose;

FIG. 3 shows a schematic illustration of an image of an object 1 on an object prototype 30; FIG.

4a shows an object prototype 30 with symmetry axis S; 4b shows the generation of a plurality of gripper poses PP "under

Using an axis of symmetry of an object prototype 30;

5 shows a schematic representation of a database table 41;

Fig. 6 is a schematic representation of a handling system

comprising an object, a robot 10 and a

Object database 40; and

7 is a flow chart illustrating a method of controlling a

Robot 10 indicates.

In the following description, the same reference numerals are used for identical and equivalent parts.

Fig. 1 shows a robot 10 shown in an initial configuration. The robot 10 has a gripper 20 which is shown in FIG

Embodiment designed as a parallel gripper 20. That is, the

Parallel gripper 20 has two jaws, which are parallel to each other and

can be moved away from each other. To measure a force, 20 force sensors are provided on the jaws of the parallel gripper, which are designed to measure how much force is applied to an object to be gripped. For detecting objects, a stereo camera 11 is attached to the robot 10

arranged. The stereo camera 11 is aligned so that it also perceives the gripper 20. So also gripped objects can be permanently recorded. Therefore, the handle of the object can also be monitored. For example, it is possible to determine the slip (so-called "slip detection") By means of the stereo camera 11, it is possible to produce a three-dimensional map of the surroundings by moving the robot 10. The three-dimensional map consists of a point cloud representing the position in a first step individual discrete points in a coordinate system of the robot 10 indicates.

In a second step, a world model can be created from the three-dimensional map. It can be determined using image processing, where z. B. surfaces are located. Also, individual objects in the environment of the robot 10 can be recognized and their pose stored within the world model. When the pose of an object in the

Saved world model, it is not necessary for gripping the object, a perception system is used. Rather, the saved pose of the object can be used. In the exemplary embodiment of FIG. 1, a bottle 1 is arranged in the immediate vicinity of the robot 10. The bottle 1 is located within the working space of the robot 10, ie in the area in which the robot 10 can theoretically grasp objects. The stereo camera 11 now detects the contours of the bottle 1. In order to allow the best possible perception of the bottle 1 by the stereo camera 11, the robot 10 can be moved so that the bottle 1 is taken from as many angles through the stereo camera 11.

To record a training gripper pose, the robot 10 can be moved such that the bottle 1 can be gripped by the gripper 20. To record a training gripper pose, the robot 10 may e.g. B. over

Keypad, so a kind of remote control, procedure. In another embodiment, the robot 10 is a

torque-controlled robot that can be hand-guided in a gravity compensation mode. The training gripper pose is recorded manually by an operator.

In FIGS. 2a and 2b it is briefly explained how the gripping of an object 1 is carried out. Fig. 2a shows the gripper 20 in a Anfahrtspose. The approach pose is usually stored in relation to an object coordinate system 2. The object coordinate system 2 is the one to be gripped

Object 1 assigned. The approaching pose comprises a three-dimensional vector VI as well as three solid angles, which rotate around that of the

Specify object coordinate system 2 given spatial axes. The vector VI points from the origin of the object coordinate system 2 to the so-called "tool center point" (TCP) of the gripper

Hand coordinate system. By using a Anfahrtspose is ensured that the approach of the gripper 20 is performed controlled before gripping and there is no collision with protruding parts of the object 1. Fig. 2b shows the gripper 20 of Fig. 2a in a gripper pose, i. in a pose in which the object 1 can be grasped. The gripper pose of the gripper 20 is also stored with respect to the object coordinate system 2. The

Gripper pose includes the vector V2 and three solid angles. The vector V2 also points from the origin of the object coordinate system 2 to the TCP of the gripper 20. In order to grasp the object 1, only the gripper jaws of the gripper 20 need be compressed in the embodiment of FIG. 2b. For example, the gripper jaws may be compressed until a threshold force value, e.g. B. 20 or 50 Newton, is exceeded. After accessing by the gripper 20, the object 1 can be changed by changing the

Robot configuration to be moved.

3, 4a and 4b show how a plurality of gripper poses P, PP "can be generated for an object 1. First of all, it is shown in Fig. 3 how an object prototype is determined for an object 1. For this purpose, a multiplicity provided by different object prototypes 30, 30 \ 30 ". in the

Embodiment of Fig. 3, a cylinder 30, a cuboid 30 ^λ and a sphere 30 "shown.

The exact configuration of the object prototypes 30, 30 \ 30 "can be determined via a large number of parameters, such that the height of the cylinder 30 can be determined by means of the parameter LI The radius of the cylinder can be defined by means of the parameter R1 ^λ can be defined in its dimensions by means of the parameters L2, L3 and L4 The sphere 30 "can be defined via the radius R2.

In order to select an object prototype 30, 30 \ 30 "for the bottle 1, the parameters LI to L4 and Rl, R2 are changed so that a maximum volume of the object prototypes 30, 30 \ 30" through the bottle 1 is filled. The bottle 1 is thus placed in the object prototypes 30, 30 \ 30 "This can be carried out by known optimization methods The object prototype, which fills the largest volume of the bottle 1 and at the same time completely encloses the bottle 1, becomes an object prototype for the bottle 1 In the illustrated embodiment of Fig. 3, therefore, the cylinder 30 for the bottle 1 is selected as an object prototype.

The object prototypes 30, 30 \ 30 "of Fig. 3 have symmetries which can be determined by means of symmetry axes S. In Fig. 4a, the

Symmetry axis S for the cylinder 30 shown. A cylinder is

constructed rotationally symmetrical about the symmetry axis S. The symmetry is used in the following to generate a multiplicity of possible gripper poses Ρ ^λ , P "for gripping the cylinder 30.

For this purpose, FIG. FIG. 4 b shows a gripper pose P of a gripper which has been manually created by an operator by means of the robot 10. Under use the axis of symmetry S can further gripper poses Ρ be generated ^λ, P "now. For this purpose the Greiferpose P is rotated by means of a transformation matrix to the center of the cylinder 30 in a plan view around. In this manner, as can in any angular intervals, e.g., every 1 degree, every 5 degrees or every 10 degrees, a gripper pose P \ P "be generated. Of course everyone can

Object prototypes 30, 30 \ 30 "a plurality of axes of symmetry S are assigned, so that a very large number of gripper poses Ρ ^λ , P" can be generated. It is thus possible for a recorded gripper pose P for gripping the bottle 1 to generate a multiplicity of further gripper poses Ρ ^λ , P ", which are likewise suitable for gripping the bottle 1.

The recorded gripper pose P and the generated gripper pose P \ P "are then stored in an object database 40. Fig. 5 shows a schematic representation of a table 41 of the object database 40. The

Database table 41 includes the columns "Object", "Gripper", "PoseStart", "PoseGr." and "F". The column "object" contains identification numbers associated with objects 1. These may be, for example, the

Identification number act in another database table, are stored in the object descriptions. Each row of the database table 41 corresponds to a gripper pose. In this case, the gripper used is stored for each gripper pose P2, P4. In the embodiment shown, a parallel gripper is stored as a gripper type in both rows of the table. In addition to the gripper poses P2, P4 each an associated Anfahrtspose PI, P3 is stored. The associated approach positions PI, P3 can like the

Gripper poses of Fig. 4b are generated by exploiting symmetry axes. The gripper poses P2, P4 correspond to either recorded

Gripper poses P or generated gripper poses P P ". In addition to the

Gripper poses P2, P4 can also optionally a force value F are stored, indicating the force with which the jaws of the parallel gripper used to press against the object to be lifted 1.

Fig. 6 shows an embodiment in which the object database 40 is used to grasp an object. 6 shows a robot 10 with a stereo camera 11 and a gripper 20. The robot 10 further comprises a control device 12, which executes the control of the robot. Arranged next to the robot 10 is a bottle. In a first step, the stereo camera 11 recognizes the object to be gripped as a bottle. The recognized bottle can be assigned to an object type. Subsequently, will from the object database 40 a plurality of candidate poses 52 are queried. For this purpose, the controller 12 communicates via a network or locally with the object database 40. For the query, the controller 12 sends a

Identification number associated with the recognized bottle. Using the transmitted identification number, those become

Candidate poses P2, P4 selected that are compatible with the detected object. After the candidate poses P2, P4 have been received by the controller 12, they are filtered using a skill card. In the process, those candidate poses are sorted out that can not be reached. In addition, a sorting takes place, which is also performed using the skill card. The order of

Gripper poses are determined by the number of directions from which the coordinates associated with the gripper poses can be reached.

On the control device 12, a path planner is also implemented which maps the path of the robot 10 from the current configuration into the

Target configuration corresponding to a first candidate pose. Can the path planner pick a path from the current robot configuration to the

Calculate target robot configuration, the first candidate pose becomes

Used to grab the object. If the path planner can not calculate a path because, for example, another object obstructs the path, then the next candidate pose is selected in the sorted list of candidate poses. This process is carried out until a candidate pose P2, P4 is found for which the path planner can calculate a path.

The described method of this application is summarized in the flowchart of FIG. 7 again.

In the first step Sl, video data 51 is analyzed to identify objects 1, Tin of the environment of the robot 10. Known objects 1, are stored together with object features in the object database 40. Using the object features, a perception algorithm is selected which is used to identify the object 1. It is now started for the first known object of the object database 40 to determine whether the object is recognizable in the video data 51. If the object is not recognizable, the second object from object database 40 is continued. Possibly. For the second object, a different perception algorithm is used than for the first object. It is determined that the second object of the object database 40 in the Video data 51 is recognizable, the step S1 is completed. The resulting object data 52 are used in step S2 to query for the detected object from the object database 40 a list of candidate poses P2, P4.

The list of candidate poses 53 is filtered in step S3 using a capability map of a robot 10. In the process, those candidate poses are removed which can not be reached with the robot 10 or gripper 20 used. In addition, the list of candidate poses 53 is sorted using the skill card 54. The list of candidate poses 53 is further sorted using user-defined criteria. For example, the gripping type stored for the candidate poses is taken into account in the sorting. Has a user indicated that when using a multi-finger gripper a

Fingergrip is to be preferred, the list of candidate poses 53 is sorted accordingly.

The sorted list of candidate poses 55 is combined with a world model 56, which includes the poses of objects in the environment of the robot 10, to generate a path for the robot 10 in step S4

used. For path planning we take into account the inverse kinematics of the robot 10 used, which indicates the robot configuration, ie the joint angles, when the position of the gripper 20 is known. In addition, at the

Path planning performed a collision check, with objects that are stored in the world model 56 are taken into account in the path planning. If it is determined in step S5 that no path for the first candidate pose of the list can be determined, a new path planning is started with the next candidate pose of the list. The process repeats until it is determined in step S5 that a robot path 57 can be calculated. In step S6, the robot 10 is then moved using the robot path 57.

It should be noted at this point that all parts described above, taken alone and in any combination, especially in the

Drawings shown details are claimed as essential to the invention. Variations thereof are familiar to the person skilled in the art.

LIST OF REFERENCES:

1 object, bottle

2 object coordinate system 10 robots

11 3D camera

12 control device

20 parallel grippers

30, 30 \ 30 "object prototype

40 object database

41 database table

51 video data

52 object data

53 List of candidate poses

54 skill card

55 sorted list of candidate poses

56 world model

57 robot path

LI, L2, L3, L4 length

T1, T2, T3 radius

P, Ρ ^λ candidate pose

PI, P3 start-up pose

P2, P4 gripper pose

VI, V2 vector

F force

S symmetry axis

S1-S6 process steps

Claims

claims

A method of creating a database (40) with gripper poses comprising:

a. Save, in particular by recording, at least one

Trainingsgreiferpose (P, P2, P4) for gripping a training object (1), in particular with respect to a training object (1) associated training object coordinate system (2);

b. Generate at least one further gripper pose (P \ P ") for the

Training object (1) using at least one

Object prototypes (30, 30 \ 30 ") and the training gripper pose (P, P2, P4);

c. Storing the at least one further gripper pose (Ρ ^λ ) and the

Trainingsgreiferpose (P, P2, P4) with respect to the

Training object (1) in the database (40), in particular in an object database (40).

A method of controlling a robot (10), comprising:

Performing the steps of the method of claim 1 for creating the database (40);

- Querying at least one pose (P, P \ P ", P2, P4) from the

Object database (40);

- Controlling the robot (10) using the at least one pose (P, P \ P ", P2, P4).

3. The method according to any one of the preceding claims,

d a d u rc h e e n c i n e s, d a s s

a pose (P, P \ P ", PI, P2, P3, P4), in particular the exercise gripper pose (P, P2, P4), is represented as a six-tuple, the three

Spatial coordinates and three angles with respect to a the training object (1) associated training object coordinate system (2).

4. The method according to any one of the preceding claims

marked by

selecting the at least one object prototype (30, 30 \ 30 ") from a set of candidate prototypes (30, 30 \ 30").

5. The method according to any one of the preceding claims, in particular according to claim 3,

characterized in that

the set of candidate prototypes (30, 30 \ 30 ")

a cylinder (30);

a sphere (30 ");

- a cuboid (30 ^λ );

- a cone;

- a truncated cone; and or

- a pyramid

includes.

6. The method according to any one of the preceding claims,

characterized in that

at least one symmetry axis (S) is assigned to the at least one object prototype (30, 30 \ 30 "), a multiplicity of gripper poses (Ρ ^λ , P") being generated taking into account the at least one symmetry axis (S), in particular by rotation around the axis at least one axis of symmetry, preferably using a

Transformation matrix, wherein the plurality of gripper poses (Ρ ^λ , P ") are stored respectively.

7. The method according to any one of the preceding claims,

characterized in that

the training gripper pose (P, P2, P4) and the further at least one gripper pose (P \ P ") are each stored with an object identification number associated with the training object (1).

8. The method according to any one of the preceding claims,

characterized in that

the storing comprises storing gripping parameters together with the poses in an object database (40), in particular

Gripper information, a gripping type, a quality indication, a force (F), a gripping strategy and / or a grip description.

9. The method according to any one of the preceding claims, characterized by

identifying the training object (1), in particular by means of a perception system (11), e.g. by means of a stereo camera (11).

10. The method according to any one of the preceding claims,

in particular according to claim 9,

marked by

storing object features for recognizing the training object (1), in particular corners and / or edges of the training object (1), preferably such that poses can be assigned to the object features.

11. The method according to any one of the preceding claims, in particular according to claim 9 or 10,

marked by

- Identifying a target object to be gripped (), in particular an object identification number associated with the target object (12);

- determining a Zielgreiferpose (P2, P4), which is the target object () assignable;

- gripping the target object (12) using the target gripper pose (P2, P4), in particular by an end effector (20), e.g. a parallel gripper (20), a robot (10).

12. The method according to any one of the preceding claims, in particular according to claim 11,

characterized in that

the determining comprises reading out a plurality of candidate poses (P2, P4), in particular from an object database (40), wherein the reading is performed in consideration of a reachability map and / or a capability map of a robot (10).

13. The method according to any one of the preceding claims, in particular according to claim 11

characterized in that

the determining comprises sorting one or more candidate poses (P2, P4), the sorting taking into account the

Skill card is performed.

14. The method according to any one of the preceding claims, in particular according to claim 10

marked by

a motion planning for moving a robot (10) from an initial configuration in a target configuration, wherein the

Target configuration of a Zielgreiferpose (P2, P4) is assigned.

A computer readable storage medium containing instructions that cause at least one processor to implement a method as claimed in any one of the preceding claims when the instructions are executed by the at least one processor.

16. Handling system comprising:

- An object database (40), in particular an object database (40), which is created according to one of claims 1 to 14;

- A robot (10) which is adapted to an object (1,) to

to grab;

- A perception system (11) for detecting the object (1,), the

is arranged in particular on the robot (10),

wherein the robot (10) is designed to

- using the detected object (1,) a plurality of

Query candidate poses (P2, P4) from the object database (40);

to filter the plurality of candidate poses (P2, P4), in particular using a reachability map and / or a

Skill card to determine a target gripper pose (P2, P4);

- Set a robot configuration according to the Zielgreiferpose (P2, P4).