Collision-free robot paths

Abstract

The invention relates to a method for planning a path of a robot, comprising the following steps: providing (S100) alternative target poses of the robot; providing (S100) environmental data, which specify a geometry of an environment of the robot; evaluating (S110) the alternative target poses with the help of a machine-learned forecast for a collision-free path for the respective target pose on the basis of this environmental data; selecting (S110) one of the alternative target poses on the basis of this evaluation as a target pose candidate; and carrying out (S120, S130) path planning on the basis of this target pose candidate. A method for machine learning of the forecast comprises the following steps: specifying (S10) alternative learning poses of the robot; providing (S10), in particular alternative, environmental learning data, which specify a geometry of an environment of the robot; carrying out (S20, S30) path planning on the basis of these learning poses and environment learning data; and machine learning (S60) of the forecast on the basis of the learning poses, environmental learning data and results of the path planning. The invention also relates to a system and to a computer program (product).

Description

[0001] Description

[0002] Collision-free robot paths

[0003] The present invention relates to a method for planning a path of a robot using a machine-learned prediction for a collision-free path, a method for machine learning the prediction, and a system and computer program or computer program product for carrying out a method described here.

[0004] A common requirement for robots is to assume one or more target poses, for example to grasp or place an object, to process a workpiece, or the like.

[0005] Often, several alternative target poses are available, for example alternative grips, alternative objects, alternative storage or handling positions.

[0006] Editing poses or the like.

[0007] If it is determined (only) during path planning for an (already) selected alternative target pose that the planned path for this target pose (is likely) to lead to a collision of the robot, or if no collision-free path is found, the entire path planning process must be repeated for another of the alternative target poses. Trying and discarding multiple target poses in this way significantly increases the time required.

[0008] An object of an embodiment of the present invention is to improve the planning or execution of robot paths, preferably to reduce, preferably to avoid, one or more of the aforementioned disadvantages or problems.

[0009] This object is achieved by a method having the features of claim 1 and 8. Claims 12 and 13 represent a system or computer program or

[0010] A computer program product for implementing a method described herein is protected. The subclaims relate to advantageous developments. According to one embodiment of the present invention, a method for planning a robot path comprises the steps:

[0011] - Providing alternative target poses of the robot;

[0012] - Providing environmental data that specifies a geometry of an environment of the robot;

[0013] - Evaluating these alternative target poses using a machine-learned prediction for a, in a further development for finding a, collision-free trajectory for the respective target pose on the basis of these environmental data, in particular on the basis of these environmental data of the respective target pose;

[0014] - selecting one of the alternative target poses based on this evaluation as a candidate target pose; and

[0015] - Perform path planning based on this target pose candidate.

[0016] One embodiment of the present invention is based on the idea of ​​first determining or selecting, in a multi-stage method, a target pose candidate that is promising, in particular more promising, preferably the or one of the most promising, with regard to, in particular the finding of a collision-free path, with the aid of or on the basis of an (evaluation with the aid of a) machine-learned forecast and then carrying out a path planning on the basis of this or for this most promising target pose candidate(s).

[0017] Compared to testing target pose candidates without ab initio consideration of collision probabilities, this approach can save statistically significant time. Compared to a single-stage determination of possible collision-free paths using an artificial neural network, the machine-learned prediction for (finding) a collision-free path can be better trained and / or work more reliably and / or faster. Additionally or alternatively, this approach can increase the reliability that no collision will occur on the planned path, particularly through path planning in a second stage.

[0018] In one embodiment, the robot has at least one robot arm. Additionally or alternatively, the robot, in particular the robot arm, has at least three, in particular at least six, and in one embodiment at least seven, joints or (motion) axes, in particular pivot joints or axes actuated or adjustable by the robot's drives and / or pivot joints or axes. For such robots, the present invention is particularly advantageous due to the kinematics, operating conditions, and collision potential.

[0019] In one embodiment, the provision of alternative target poses comprises specifying alternative target poses, in a further development based on user input and / or at least partially automated, in one embodiment based on environmental data of the robot, in particular environmental detection and / or processing data, for example (detected) objects to be grasped, possible storage locations, workpieces to be processed, robot-guided tools, or the like. Additionally or alternatively, the provision can in particular comprise retrieving or loading stored alternative target poses. In one embodiment, this allows more promising or more expedient target poses to be used in advance and / or the process to be accelerated.

[0020] A pose in the sense of the present invention comprises, in one embodiment, a position of the robot, in particular its joints, and / or a position, in particular one-, two- or three-dimensional, and / or an orientation, in particular one-, two- or three-dimensional, in particular of at least one robot-fixed reference, preferably of an end effector, TCP or the like.

[0021] The provision of environmental data that specifies a geometry of the robot's environment comprises, in one embodiment, sensory detection of the environment, in a further development using at least one camera, in particular a stationary or robot-guided camera, and / or data processing, in particular filtering, transformation, or the like, of environmental data and / or retrieving or loading stored environmental data. In one embodiment, the provided environmental data is based on sensory detection of the robot's environment, in particular using at least one camera, in particular a stationary or robot-guided camera, and / or a theoretical model of the environment, for example a CAD model of a robot cell or the like. In one embodiment, this can improve the reliability of collision avoidance.A machine-learned prediction for a collision-free path for a target pose, in particular for finding a collision-free path for the target pose, determines or delivers in one embodiment on the basis of environmental data (and the target pose) a value, preferably a binary value, a probability value or the like, which depends on whether a collision-free path for the target pose can be (successfully) planned or whether a collision is likely to occur or with what probability on a path planned for the target pose.In one embodiment, the machine-learned prediction F maps various input data X, each comprising (the) environmental data c and one of the alternative target poses x, to output data Y, which in one embodiment indicate in binary form whether a collision-free trajectory for this target pose x can probably be (successfully) planned or whether a collision is likely to occur on a trajectory planned for this target pose x, or indicate a probability for this, for example between zero and one or the like (F(X) = Y).

[0022] As already explained, in one embodiment, the or one of the alternative target poses is selected as the target pose candidate for which the machine-learned prediction predicts that a collision-free trajectory can probably be (successfully) planned, or that no collision is likely to occur on a planned trajectory, or that the probability of this is higher than for one or more of the other alternative target poses, preferably the highest. If two or more of the alternative target poses are evaluated equally, in a further development, one of these target poses is selected randomly or according to further criteria, for example, a predefined prioritization or the like.

[0023] Preferably, the planning of a robot path, which is subsequently checked for collisions, is performed in a similar manner during the machine learning prediction and the execution of path planning based on the selected target pose candidate. This can reduce the probability of failed attempts in one embodiment.

[0024] The implementation of path planning comprises, in one embodiment, in particular in a first stage, planning at least one path of the robot and, in particular in a second stage, checking this planned path, in a further development of the target pose and / or one or more (other) poses of the robot along the path, for collisions of the robot, in particular with itself and / or the environment. This planning of a path can in particular comprise specifying one or more path points, in particular (further) poses of the robot along the path, and, in one embodiment, connecting these path points, for example by means of interpolation. This allows the path planning to be carried out more quickly in one embodiment.

[0025] In one embodiment, when performing path planning, a (first) robot path is first planned and checked for robot collisions. If a collision is detected, another robot path is planned one or more times and checked for robot collisions, preferably until a termination criterion, such as a predefined maximum number of attempts or the like, is met. As a result, in one embodiment, several (different) attempts are made to plan a path for a selected target pose candidate, if necessary, and these paths are then checked for collisions. A new target pose candidate is selected only if a collision is detected (also) in several of these (differently) planned paths for this target pose candidate. As a result, in one embodiment, prioritization within the target poses that is not based on predicted collisions can advantageously be taken into account.For example, for a first gripping position for which the machine-learned prediction has predicted a collision-free path and the execution of a path planning has resulted in a first path for which a collision was detected during the collision check (yet or contrary to the prediction), one or more other paths can first be planned and also checked for collisions before a different gripping pose is selected as the new target pose candidate instead of this first gripping pose, for example if the first gripping pose can be reached more quickly by the robot or is closer to a storage location or the like.

[0026] In another embodiment, when performing path planning, only one robot path is planned at a time and checked for collisions. If a collision is detected, a new target pose candidate is selected instead of this target pose candidate, and a new path planning is performed for this new path. Thus, in one embodiment, when a collision is detected for a selected target pose candidate, a new target pose candidate is selected next. This can advantageously reduce the probability of failed attempts in one embodiment.

[0027] In one embodiment, performing path planning comprises an attempt to plan a robot path (ab initio) while avoiding collisions, in particular with the help of a collision-free path planner. For example, starting from the target pose, the path planner can attempt to plan a path in a space determined to be collision-free, in particular, to determine such a path, to successively determine collision-free path points on the way to a desired path end, or the like. In one embodiment, this can reduce the probability of failed attempts.

[0028] In one embodiment, path planning is carried out on the basis of the provided environmental data; in a further development, the planned path(s) are checked for collisions of the robot on the basis of the environmental data or an attempt is made to plan a path of the robot (ab initio) on the basis of the environmental data while avoiding collisions of the robot, in particular with the help of a collision-free path planner.

[0029] In one embodiment, the procedure comprises the following steps, which may be repeated several times in a further training:

[0030] - selecting another of the alternative target poses based on the evaluation as the new target pose candidate if a termination condition is met during the previous execution of the path planning; and

[0031] - Perform path planning based on this new target pose candidate.

[0032] The termination condition may in particular include that no path could be planned while avoiding collisions, in particular within a predetermined time or maximum number of attempts or the like, or that a collision was detected during the testing of a planned path, or that a termination criterion for this path planning, for example a predetermined maximum number of attempts or the like, is met. Thus, in one embodiment, if path planning for a target pose candidate fails, in particular if no path could be planned for this while avoiding collisions, or that a collision was detected during the testing of a planned path or several planned paths (in each case), a new target pose candidate is selected instead, and path planning is carried out again on the basis of this new target pose candidate. In one embodiment, this can reduce the probability of failed attempts.

[0033] In one embodiment, the alternative target poses include starting poses of the robot, end poses of the robot, and / or working poses of the robot; in a further development, picking poses, in particular gripping poses of the robot, and / or release poses of the robot. Additionally or alternatively, in one embodiment, one or the (respective) path of the robot includes the (respective) target pose, in particular the (respective) target pose candidate.

[0034] Thus, a planned path can be, in particular, a path for gripping one of several objects, for gripping an object in one of several gripping positions, a path for moving to one of several storage positions, a path for moving to or from one or more processing positions, or the like. The present invention is particularly advantageous for this purpose, particularly due to the kinematics, operating conditions, and collision potential.

[0035] In one embodiment, the environmental data comprises surface points and / or images, in a further development depth images and / or from a predetermined perspective, preferably from a perspective of a camera with which the environment is recorded or with the aid of which the environmental data is generated, and / or an end effector, preferably a gripper, of the robot or a perspective shifted relative to this end effector perspective, in particular in the gripping direction.

[0036] Surface points are particularly suitable for machine-learned predictions, while images can advantageously represent the environment and / or can be processed more directly.

[0037] In one embodiment, the robot follows the (successfully) planned path, in particular the path (successfully) planned while avoiding collisions of the robot, or the planned path for which no collision was detected during testing. Accordingly, in one embodiment, the present invention relates to a method for (planning and) following a (planned) path with the robot, or to a corresponding system.

[0038] According to one embodiment of the present invention, a method for machine learning a prediction based on environmental data (and a target pose of a robot) for a collision-free trajectory for the target pose comprises the steps:

[0039] - Specifying alternative learning poses for the robot;

[0040] - Providing, in one embodiment, alternative environment learning data that (each) specifies a geometry of a (different) environment of the robot;

[0041] - Performing path planning based on these learning poses and environment learning data; and

[0042] - machine learning of the forecast based on the learning poses, environmental learning data and results of the path planning.

[0043] The machine-learned forecast can be carried out, in particular, using at least one artificial neural network or at least one other model or principle of artificial intelligence or machine learning, in particular for classification or regression. As already explained above, in one embodiment, the machine-learned forecast F maps input data X, each of which comprises environmental data c and a target pose x, to output data Yj (F(X) = Y) and, accordingly, in one embodiment, based on input learning data X', the respective environmental learning data c' orc' (alternative environment learning data) and a learning pose x, and machine-learned, in particular trained, data Y', each based on a result of trajectory planning performed for this learning pose and environment learning data, in a further development depend on whether the trajectory planning was carried out successfully or not, in a further development each specify a corresponding probability (if, for example, a stochastic trajectory planner is used for data collection, several results can arise for a pose, for example, the planner could find a trajectory in 6 out of 10 cases, in which case the corresponding data value would then be set to 0.6). By taking different environment learning data c' into account, the machine-learned prediction can advantageously be used more successfully in different environments in one embodiment.

[0044] In one embodiment, the environment learning data is, preferably, created using one or more, in particular numerical, simulations, in particular alternative environment learning data is created using simulations of different environments. Additionally or alternatively, in one embodiment, the path planning, on the basis of which the forecast is machine-learned, is carried out using one or more, in particular numerical, simulations; in a further development, path planning is carried out using simulations of different environments or on the basis of alternative environment learning data or alternative simulated or virtual environments. Particularly preferably, the same simulated or virtual environments are used as a basis for carrying out path planning and the machine learning of the forecast based on the results of this path planning.For one or more virtual environments, environment learning data is created using a simulation, a path planning is performed based on this environment learning data, and the result is linked to this environment learning data, or this result and this environment learning data are used for machine learning of the forecast. In other words, the forecast for target poses and various virtual environments learns machine learning to predict the success of planning a collision-free path for this target pose in this virtual environment.

[0045] This allows the forecast to be learned more quickly by machines and / or its forecast quality to be improved.

[0046] The features explained above with reference to path planning, in particular that the implementation of path planning comprises planning at least one path of the robot and checking this planned path for collisions of the robot or an attempt to plan a path of the robot while avoiding collisions of the robot, can additionally or alternatively also be implemented in the machine learning of the prediction, so that reference is made to the above explanations in this regard. Accordingly, in one embodiment, the environment learning data additionally or alternatively comprise surface points and / or images, in particular depth images and / or from a predetermined perspective.In a further development, surface points are first generated, preferably with the aid of at least one simulation, and from these at least one image, in particular a depth image and / or from a predetermined perspective, preferably from a perspective of a camera with which, in one embodiment (for planning the path), the environment is recorded or with whose aid, in one embodiment (for planning the path), the environment data is generated, and / or an end effector, preferably a gripper, of the robot or a perspective shifted relative to this end effector perspective, in particular in the gripping direction.It is particularly advantageous if environmental data and environmental learning data are of the same type, in particular both comprise surface points or both comprise images, in particular depth images and / or from the same perspective, and / or if the implementation of path planning on the basis of a target pose candidate and the implementation of path planning on the basis of a learning pose are of the same type, in particular both comprise a preferably similar planning of at least one path of the robot and a preferably similar checking of this planned path for collisions of the robot or a preferably similar attempt to plan a path of the robot while avoiding collisions of the robot, since this enables the machine-learned forecast to better predict the conditions when planning the path.

[0047] According to one embodiment of the present invention, a system for planning, in a further development for traveling, a path of a robot and / or for machine learning a prediction (based on environmental data and a target pose of the robot) for a collision-free path (for the target pose) of a robot, in particular hardware and / or software, in particular program technology, is set up to carry out a method described here.

[0048] According to one embodiment of the present invention, a system comprises:

[0049] - means for providing alternative target poses of the robot;

[0050] - means for providing environmental data specifying a geometry of an environment of the robot;

[0051] - means for evaluating the alternative target poses using a machine-learned prediction for a collision-free trajectory for the respective target pose based on these environmental data; means for selecting one of the alternative target poses as a candidate target pose based on this evaluation; and

[0052] Means for performing path planning based on this target pose candidate.

[0053] In one embodiment, the system or its means comprises:

[0054] - means for selecting, in particular repeatedly selecting, another of the alternative target poses on the basis of the evaluation as a new target pose candidate if a termination condition is met during the previous execution of the path planning; and

[0055] - Means for performing path planning based on this new target pose candidate.

[0056] In one embodiment, the system or its means comprises:

[0057] - Means for traversing the planned path with the robot.

[0058] According to one embodiment of the present invention, a system comprises:

[0059] - Means for specifying alternative learning poses of the robot;

[0060] - means for providing, in particular alternative, environment learning data specifying a geometry of an environment of the robot;

[0061] - means for performing path planning based on these learning poses and environmental learning data; and

[0062] - Means for machine learning prediction based on the learning poses, environment learning data and trajectory planning results.

[0063] In one embodiment, the system or its means comprises:

[0064] - means for creating the environmental learning data using at least one simulation; and / or

[0065] - Means for performing the path planning for machine learning using at least one simulation.

[0066] In one embodiment, the or at least one of the system(s) or its means for performing path planning comprises:

[0067] - means for planning at least one path of the robot; and

[0068] - Means for checking this planned path for collisions of the robot. In one embodiment, the system or at least one of the systems or its means for performing path planning comprises:

[0069] - Means for attempting to plan a path of the robot while avoiding collisions.

[0070] 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 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 oris capable of executing such a method, so that the processing unit can carry out the steps of such methods and thus in particular plan a path of a robot, in a further development control the robot to follow the path, and / or can machine-learn a prediction (based on environmental data and a target pose of the robot) for a collision-free path (for the target pose) of a robot. In one embodiment, a computer program product can have, 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 several computers, causes the system or controller, in particular the computer orthe computer, to carry out a procedure described here or one or more of its steps, or the program or instructions are configured to do so.

[0071] In one embodiment, one or more, in particular all, steps of a method described here are fully or partially computer-implemented, or one or more, in particular all, steps of the method are fully or partially automated, in particular by a system described here or its means. In one embodiment, a system comprises the robot.

[0072] In one embodiment, at least one of the path planning steps mentioned here is performed using or by a stochastic path planner. This can improve the machine learning of the forecast and / or the execution of path planning based on target pose candidates, in particular increasing convergence and / or speed.

[0073] Further advantages and features emerge from the subclaims and the exemplary embodiments. The following shows, partly schematically:

[0074] Fig. 1: a system for performing the method of Fig. 2 according to an embodiment of the present invention; and

[0075] Fig. 2: a method for machine learning a prediction, planning a path using the machine learned prediction and following the planned path with a robot according to an embodiment of the present invention.

[0076] Fig. 1 shows a robot 1, a robot controller 2, a camera 3 and an environment indicated by a container 4, from which the robot 1 is to transport objects 5A-5C.

[0077] For this purpose, according to a method for machine learning a prediction based on environmental data and a target pose of a robot for a collision-free trajectory for the target pose according to one embodiment of the present invention, in a step S10, various alternative environment learning data are provided with the aid of simulations of various environments, of which two simulated or virtual environments 4', 4" are indicated by dashed or dash-dotted lines in Fig. 1, and one of the simulated or virtual environments or the corresponding environment learning data is selected. This environment learning data can, for example, be surface point clouds or image data rendered from these and correspond to a depth image of the camera 3. For or in the selected simulated or virtual environment, alternative learning poses of the robot for gripping objects are specified in step S10, of which two are shown in Fig.1 three learning poses 1A-1C are indicated by dotted gripper poses, and one of these is selected.

[0078] In steps S20, S30, a path planning is carried out for the selected learning pose and the selected environment learning data or simulated or virtual environment based on this selected learning pose and environment learning data and its result is saved.

[0079] In one embodiment, in step S20, a path is first planned based on the learning pose, and in step S30, this planned path is checked for collisions of the robot using a simulation based on the environmental learning data, wherein the result of this path planning in step S30 is stored as to whether or not a collision was detected in the planned path.

[0080] In another embodiment, in step S20, an attempt is made to plan a path of the robot while avoiding collisions of the robot using a simulation based on the environment learning data or in the selected simulated or virtual environment, and in step S30, as a result of this path planning, it is stored whether or not a path could be planned while avoiding collisions.

[0081] In step S40, a check is made to determine whether trajectory planning has already been performed for all learning poses. If this is not the case (S40: "N"), in step S45, one of the learning poses specified in step S10 for which trajectory planning has not yet been performed is selected, and steps S20 and S30 are performed again.

[0082] Otherwise (S40: "Y"), a step S50 checks whether trajectory planning has been performed for all simulated or virtual environments or alternative environment learning data for the learning poses. If this is not the case (S50: "N"), a simulated or virtual environment or the corresponding environment learning data for which trajectory planning has not yet been performed is selected in a step S55, and steps S20 and S30 are repeated. After processing all environment learning data (S50: "Y"), a prediction is machine-learned in a step S60 based on the learning poses, environment learning data, and trajectory planning results. Of course, this can also be performed while repeating steps S20–S50.

[0083] According to a method for planning a robot path, alternative target poses of the robot and environmental data specifying a geometry of the robot's environment are provided in a step S100. For example, the camera 3 can capture the real environment 4 and generate the environmental data therefrom, and possible target poses for grasping the various objects 5A-5C can be specified.

[0084] In a step S110, the alternative target poses are evaluated using the machine-learned prediction, and on the basis of this evaluation, the one or more target poses are selected as the target pose candidate for which the machine-learned prediction predicts successful planning of a trajectory without collisions or the highest probability of successful planning of a trajectory without collisions.

[0085] In steps S120, S130, path planning is then carried out analogously to the steps S20, S30 described above, wherein the path planning is carried out on the basis of this target pose candidate instead of a learning pose and on the basis of the environment data instead of the environment learning data.

[0086] Accordingly, in one embodiment, in step S120, a path is first planned based on the target pose candidate, and this planned path is checked for robot collisions in step S130 based on the environmental data. In another embodiment, in step S120, an attempt is made to plan a robot path based on the environmental data while avoiding robot collisions, and step S130 in Fig. 2 is omitted or step S140 follows step S120.

[0087] In step S140, a check is made to determine whether the path planning was successful. If this is the case (S140: "Y"), the planned path is followed by robot 1 in step S145, and the process is terminated. Step S145 can also be omitted or (only) involve saving the planned path for further, particularly later, use.

[0088] If the path planning was unsuccessful (S140: "N"), a check is performed to determine whether further target pose candidates are available. If this is the case (S150: "Y"), in a step S155, the target pose(s) for which the machine-learned prediction (also) predicts successful planning of a path without collisions or the highest probability of successful planning of a path without collisions among the remaining target poses for which steps S120-S140 have not yet been performed is selected as the new target pose candidate based on the evaluation.

[0089] For this new target pose candidate, steps S120-S140 are performed again.

[0090] If all target poses have been tried without successfully planning a path without collisions (S150: “N”), an error message is output and the process is also terminated or appropriate countermeasures are taken, for example changing the arrangement of the objects 5A-5C using the robot 1 or the like (Fig. 2: step S160).

[0091] Fig. 1 clearly illustrates a basic idea of ​​the present invention: based on the simulated or virtual environments 4', 4" or the corresponding alternative environment learning data, a prediction is machine-learned that predicts that successful path planning is not possible or less likely for edge positions. Accordingly, when planning the path for the detected environment 4, the pose for grasping the object 5B is evaluated as the most promising candidate, and path planning is carried out for this.

[0092] Although exemplary embodiments have been explained in the preceding description, it should be noted that numerous modifications are possible. 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, whereby various modifications, particularly with regard to the function and arrangement of the described components, can be made without departing from the scope of protection as it results from the claims and equivalent combinations of features.

[0093] List of reference symbols

[0094] 1 robot

[0095] 1A-1C Target / learning poses for grasping an object

[0096] 2 Robot control 3 Camera

[0097] 4 containers (environment)

[0098] 4'; 4" virtual / simulated environment

[0099] 5A-5C Item

Claims

Patent claims 1. Method for planning a path of a robot (1), comprising the steps: - Providing (S100) alternative target poses of the robot; - Providing (S100) environmental data specifying a geometry of an environment of the robot; - Evaluating (S110) the alternative target poses using a machine-learned prediction for a collision-free trajectory for the respective target pose based on this environmental data; - selecting (S110) one of the alternative target poses based on this evaluation as a candidate target pose; and - Carrying out (S120, S130) a path planning based on this target pose candidate.

2. Method according to claim 1, characterized in that the implementation of a path planning on the basis of the target pose candidate comprises planning (S120) at least one path of the robot and checking (S130) this planned path for collisions of the robot.

3. The method according to claim 1, characterized in that the execution of a path planning on the basis of the target pose candidate comprises an attempt (S120) to plan a path of the robot while avoiding collisions of the robot.

4. Method according to one of the preceding claims, characterized by the steps, in particular repeated several times: - selecting (S155) another of the alternative target poses on the basis of the evaluation as a new target pose candidate if a termination condition is met during the previous execution of the path planning; and - Perform path planning based on this new target pose candidate.

5. Method according to one of the preceding claims, characterized in that the alternative target poses comprise start poses, end poses and / or working poses, in particular pick-up and / or delivery poses, of the robot.

6. Method according to one of the preceding claims, characterized in that the environmental data comprise surface points and / or images, in particular depth images and / or from a predetermined perspective.

7. Method according to one of the preceding claims, characterized in that the forecast is machine-learned according to a method according to one of the following claims and / or the robot follows the planned path (S145).

8. Method for machine learning a prediction for a collision-free path for the target pose based on environmental data and a target pose of a robot (1), comprising the steps: - Specifying (S10) alternative learning poses of the robot; - Providing (S10) of, in particular alternative, environment learning data that specify a geometry of an environment of the robot; - Carrying out (S20, S30) path planning based on these learning poses and environment learning data; and - machine learning (S60) of the prediction based on the learning poses, environmental learning data and results of the path planning.

9. Method according to the preceding claim, characterized in that the environment learning data are created with the aid of at least one simulation and / or comprise surface points and / or images, in particular depth images and / or from a predetermined perspective, and / or the path planning is carried out with the aid of at least one simulation.

10. Method according to one of claims 8-9, characterized in that the implementation of a path planning based on a learning pose comprises planning (S20) at least one path of the robot and checking (S30) this planned path for collisions of the robot.

11. Method according to claims 8-9, characterized in that the execution of a path planning based on a learning pose comprises an attempt (S20) to plan a path of the robot while avoiding collisions of the robot.

12. System for planning, in particular following, a path of a robot (1) and / or for machine learning a prediction for a collision-free path of a robot, wherein the system is set up to carry out a method according to one of the preceding claims and / or: - means for providing alternative target poses of the robot; - means for providing environmental data specifying a geometry of an environment of the robot; - means for evaluating the alternative target poses using a machine-learned prediction for a collision-free trajectory for the respective target pose based on this environmental data; - means for selecting one of the alternative target poses based on this evaluation as a candidate target pose; and - has means for carrying out path planning on the basis of this target pose candidate and / or - Means for specifying alternative learning poses of the robot; - means for providing, in particular alternative, environment learning data specifying a geometry of an environment of the robot; - means for performing path planning based on these learning poses and environmental learning data; and - Means for machine learning of the prediction based on the learning poses, environmental learning data and results of the trajectory planning.

13. A computer program or computer program product, wherein the computer program or computer program product contains instructions, in particular stored on a computer-readable and / or non-volatile storage medium, which, when executed by one or more computers or a system according to claim 12, cause the computer(s) or the system to carry out a method according to one of claims 1 to 11.

Images