DE102019213676B3 - Robot control - Google Patents

Abstract

The present invention relates to a method for controlling a robot (10) which has a number DoF of controllable degrees of freedom (x, y, z, α, β, γ), wherein: changes in one or more first of several steps (S10) in the degrees of freedom of a first subset of the degrees of freedom, which has at least one and at most DoF-1 of the degrees of freedom, are each limited to a maximum value, this maximum value in each case at most 10% of a maximum value of a change in the respective degree of freedom in at least one other of the several steps is; and in one or more second of the plurality of steps (S20) changes in the degrees of freedom of a second subset of the degrees of freedom, which has at least one and at most DoF-1 of the degrees of freedom and is at least partially different from the first subset, are each limited to a maximum value , this maximum value being at most 10% of a maximum value of a change in the respective degree of freedom in at least one other of the plurality of steps; where changes in the degrees of freedom of a first remainder of the degrees of freedom complementary to the first subset in the first step or the first steps and changes in the degrees of freedom of a second remainder of the degrees of freedom complementary to the second subset in the second step or the second steps by a controller ( 20) can be commanded by means of machine learning.

Description

The present invention relates to a method and system for controlling a robot and a computer program product for carrying out the method.

From in-house practice, controllers are known that use machine learning to command changes in the robot's degrees of freedom and thus control the robot.

The DE 10 2019 001 207 A1 relates to a collaborative robot for processing a workpiece, which has a robot tool for processing the workpiece and a mode selection device for transferring the collaborative robot into a learning mode and in a working mode and is configured to learn a working movement of the robot tool for processing the workpiece in the learning mode and to perform the work movement learned in the learning mode in the work mode, the learning mode being configured so that the collaborative robot learns the work movement of the robot tool on the workpiece to be processed in the learning mode.

The DE 10 2018 009 008 A1 relates to a control device and a machine learning device that enable control for gripping an object with a low reaction force, the machine learning device included in the control device having a status observation unit that observes the shape of the gripped object shape data on a shape of the gripped object as a status variable representing a current status of an environment, an identification data acquisition unit that acquires grasping width data representing a width of the hand of the robot when grasping the grasped object as identification data, and comprises a learning unit that performs learning using the status variable and the identification data so that the grasped object shape data is associated with the grasping width data.

The DE 10 2018 001 026 A1 relates to a robot system provided with a robot control device, the robot control device comprising an operation control unit and a learning control unit that executes learning control in which a vibration correction amount for correcting a vibration generated at a control target portion of a robot is calculated and the vibration correction amount the next time in the operation command is used.

The DE 10 2017 011 361 A1 relates to a machine learning device for learning the operation of a robot and a laser scanner that includes a state observation unit that observes a state of a tip end of the robot on which the laser scanner is mounted and a state of an optical component in the laser scanner as state data, a unit for achieving of determination data including at least one of a processing time of the robot on which the laser scanner is mounted, a drive current that drives the robot, a guide path of the laser scanner, a passage time over a processable surface on which the laser scanner 13 performs processing, and a A distance between the robot and a part on which the laser scanner performs processing receives as determination data, and comprises a learning unit that explains the operation of the robot and the laser scanner based on an output of the state observation unit and an output of the unit for obtaining v on determination data learned.

The DE 10 2017 105 598 A1 relates to a power supply system for an electric vehicle, comprising: a rail, a first battery pack, a second battery pack and a drive device.

The DE 10 2016 012 227 A1 relates to a method for automatically correcting the position of a robot arm of a robot with the following steps: approaching a target position by a robot arm of a robot; Determining an actual position actually reached by the robot arm; Learning of a position correction by the robot based on a deviation of the actual position from the target position using a deep learning algorithm.

The object of the present invention is to improve the control of a robot.

This object is achieved by a method with the features of claim 1. Claims 12, 13 provide a system or computer program product for performing a method described here under protection. The subclaims relate to advantageous developments.

According to one embodiment of the present invention, a robot has a number of DoF of, in particular, controllable degrees of freedom.

In one embodiment, the robot has a robot arm with at least three, in particular at least six, in one embodiment at least seven, (movement) axes or joints, in one embodiment rotary axes or joints, and drives, in particular electric motors, for adjusting these joints .

In one embodiment, an end effector of the robot arm can advantageously be shifted in three spatial directions by at least three (rotary) joints, by at least six (rotary) joints advantageously also be oriented as desired, by at least seven (rotary) joints, advantageously with different robot poses, in particular while avoiding collisions and / or with each optimized poses.

The present invention is particularly suitable for such robots, in particular because of its mostly complex, non-linear kinematics, but without being restricted thereto. Thus, in one embodiment, the robot can additionally or alternatively have a mobile base, in particular a chassis, and in one embodiment it can be a driverless transport vehicle (“Automated Guided Vehicle”, AGV). The present invention is also particularly suitable for this, in particular because of the complex, dynamic environment.

• - in einem oder mehreren Schritten, die vorliegend ohne Beschränkung der Allgemeinheit als erste Schritte einer Mehrzahl von Schritten bezeichnet werden, (die bzw. eventuelle) Änderungen in den, insbesondere allen, Freiheitsgraden einer ersten, insbesondere vorgegebenen, (echten) Teil- bzw. Untermenge der DoF Freiheitsgrade, die wenigstens einen und höchstens (DoF-1) der Freiheitsgrade aufweist, jeweils auf einen Maximalwert begrenzt, wobei dieser Maximalwert (des bzw. in dem jeweiligen Freiheitsgrad(s)) jeweils (betragsmäßig) höchstens 10%, insbesondere höchstens 1%, in einer Ausführung höchstens 0,1%, eines, insbesondere kommandierten oder möglichen bzw. zulässigen, Maximalwerts einer Änderung in d(ies)em jeweiligen Freiheitsgrad in wenigstens einem anderen der Schritte der Mehrzahl von Schritten beträgt, und
• - in einem oder mehreren Schritten der Mehrzahl von Schritten, die vorliegend ohne Beschränkung der Allgemeinheit als zweite Schritte bezeichnet werden, (die bzw. eventuelle) Änderungen in den, insbesondere allen, Freiheitsgraden einer zweiten, insbesondere vorgegebenen, (echten) Teil- bzw. Untermenge der DoF Freiheitsgrade, die wenigstens einen und höchstens (DoF-1) der Freiheitsgrade aufweist und von der ersten Teilmenge wenigstens teilweise verschieden ist, jeweils auf einen Maximalwert begrenzt, wobei dieser Maximalwert (des bzw. in dem jeweiligen Freiheitsgrad(s)) jeweils (betragsmäßig) höchstens 10%, insbesondere höchstens 1%, in einer Ausführung höchstens 0,1%, eines, insbesondere kommandierten oder möglichen bzw. zulässigen, Maximalwerts einer Änderung in d(ies)em jeweiligen Freiheitsgrad in wenigstens einem anderen der Schritte der Mehrzahl von Schritten beträgt,

wobei

• - (die bzw. eventuelle) Änderungen in den Freiheitsgraden eines zur ersten Teilmenge komplementären ersten Rests der Freiheitsgrade und in einer Ausführung (die bzw. eventuelle) auf den jeweiligen Maximalwert begrenzten Änderungen in den Freiheitsgraden der ersten Teilmenge in dem bzw. den ersten Schritt(en) und
• - (die bzw. eventuelle) Änderungen in den Freiheitsgraden eines zur zweiten Teilmenge komplementären zweiten Rests der Freiheitsgrade und in einer Ausführung (die bzw. eventuelle) auf den jeweiligen Maximalwert begrenzten Änderungen in den Freiheitsgraden der zweiten Teilmenge in dem bzw. den zweiten Schritt(en),

von einem Regler mittels maschinellen Lernens kommandiert werden.According to one embodiment of the present invention, in order to control the robot, in particular to carry out a given task,

• - in one or more steps, which in the present case are referred to as the first steps of a plurality of steps without loss of generality, (the or possible) changes in the, in particular all, degrees of freedom of a first, in particular predetermined, (real) partial or Subset of the DoF degrees of freedom, which has at least one and at most (DoF-1) of the degrees of freedom, each limited to a maximum value, this maximum value (of or in the respective degree of freedom (s)) in each case (in terms of amount) at most 10%, in particular at most 1%, in one embodiment at most 0.1%, of a, in particular commanded or possible or permissible, maximum value of a change in the respective degree of freedom in at least one other of the steps of the plurality of steps, and
• - in one or more steps of the plurality of steps, which in the present case are referred to as second steps without loss of generality, (the or possible) changes in the, in particular all, degrees of freedom of a second, in particular predetermined, (real) partial or Subset of the DoF degrees of freedom, which has at least one and at most (DoF-1) of the degrees of freedom and is at least partially different from the first subset, each limited to a maximum value, this maximum value (of or in the respective degree of freedom (s)) (in terms of amount) no more than 10%, in particular no more than 1%, in one embodiment no more than 0.1%, a, in particular commanded or possible or permissible, maximum value of a change in d (ies) em respective degree of freedom in at least one other of the steps of the plurality of steps,

in which

• - (the or possible) changes in the degrees of freedom of a first remainder of the degrees of freedom which are complementary to the first subset and in one embodiment (the or possible) changes in the degrees of freedom of the first subset in the first step (s) limited to the respective maximum value ( en) and
• - (the or possible) changes in the degrees of freedom of a second remainder of the degrees of freedom, which is complementary to the second subset, and in one embodiment (the or possible) changes in the degrees of freedom of the second subset in the second step or steps limited to the respective maximum value ( en),

can be commanded by a controller using machine learning.

• - in einem oder mehreren Schritten der Mehrzahl von Schritten, die vorliegend ohne Beschränkung der Allgemeinheit als dritte Schritte bezeichnet werden, (die bzw. eventuelle) Änderungen in den, insbesondere allen, Freiheitsgraden einer dritten, insbesondere vorgegebenen, (echten) Teil- bzw. Untermenge der DoF Freiheitsgrade, die wenigstens einen und höchstens (DoF-1) der Freiheitsgrade aufweist und von der ersten und zweiten Teilmenge wenigstens teilweise verschieden ist, jeweils auf einen Maximalwert begrenzt, wobei dieser Maximalwert (des bzw. in dem jeweiligen Freiheitsgrad(s)) jeweils (betragsmäßig) höchstens 10%, insbesondere höchstens 1%, in einer Ausführung höchstens 0,1%, eines, insbesondere kommandierten oder möglichen bzw. zulässigen, Maximalwerts einer Änderung in d(ies)em jeweiligen Freiheitsgrad in wenigstens einem anderen der Schritte der Mehrzahl von Schritten beträgt, wobei (die bzw. eventuelle) Änderungen in den Freiheitsgraden eines zur dritten Teilmenge komplementären dritten Rests der Freiheitsgrade und in einer Ausführung (die bzw. eventuelle) auf den jeweiligen Maximalwert begrenzten Änderungen in den Freiheitsgraden der dritten Teilmenge in dem bzw. den dritten Schritt(en) von dem Regler mittels maschinellen Lernens kommandiert werden.

In one embodiment are, in particular,

• - in one or more steps of the plurality of steps, which in the present case are referred to as third steps without loss of generality, (the or possible) changes in the, in particular all, degrees of freedom of a third, in particular predetermined, (real) partial or Subset of the DoF degrees of freedom, which has at least one and at most (DoF-1) of the degrees of freedom and is at least partially different from the first and second subset, each limited to a maximum value, this maximum value (of or in the respective degree of freedom (s)) ) in each case (in terms of amount) a maximum of 10%, in particular a maximum of 1%, in one embodiment a maximum of 0.1%, a, in particular commanded or possible or permissible, maximum value of a change in d (ies) em respective degree of freedom in at least one other of the steps of the plurality of steps, with (the or possible) changes in the degrees of freedom of a complementary to the third subset n third remainder of the degrees of freedom and in one embodiment (the or possible) changes in the degrees of freedom of the third subset limited to the respective maximum value in the third step (s) are commanded by the controller by means of machine learning.

For example, if the robot has DoF = 6 degrees of freedom {x, y, z, α, β, γ}, then a first subset can be {α, β, γ}, the first remainder complementary to this can be {x, y, z} , a second subset {x, y, z, α, β}, the complementary second residue corresponding to {γ}, and a third subset {x, y, z, γ}, the complementary third residue corresponding to {α, β }.

• - Änderungen im Freiheitsgrad α auf einen Maximalwert α_max begrenzt, der (betragsmäßig) höchstens 10%, insbesondere höchstens 1%, in einer Ausführung höchstens 0,1%, eines, insbesondere kommandierten oder möglichen bzw. zulässigen, Maximalwerts einer Änderung in diesem Freiheitsgrad α in einem oder mehreren dritten Schritt(en) beträgt;
• - Änderungen im Freiheitsgrad β auf einen Maximalwert β_max begrenzt, der (betragsmäßig) höchstens 10%, insbesondere höchstens 1%, in einer Ausführung höchstens 0,1%, eines, insbesondere kommandierten oder möglichen bzw. zulässigen, Maximalwerts einer Änderung in diesem Freiheitsgrad β in dem bzw. den dritten Schritt(en) beträgt; und
• - Änderungen im Freiheitsgrad γ auf einen Maximalwert γ_max begrenzt, der (betragsmäßig) höchstens 10%, insbesondere höchstens 1%, in einer Ausführung höchstens 0,1%, eines, insbesondere kommandierten oder möglichen bzw. zulässigen, Maximalwerts einer Änderung in diesem Freiheitsgrad γ in einem oder mehreren zweiten Schritt(en) beträgt; beispielsweise also Änderungen in allen rotatorischen Freiheitsgraden auf einen (verglichen mit den zweiten bzw. dritten Schritten) niedrige(re)n Maximalwert begrenzt, in dem bzw. den zweiten Schritt(en)
• - Änderungen im Freiheitsgrad x auf einen Maximalwert x_max begrenzt, der (betragsmäßig) höchstens 10%, insbesondere höchstens 1%, in einer Ausführung höchstens 0,1%, eines, insbesondere kommandierten oder möglichen bzw. zulässigen, Maximalwerts einer Änderung in diesem Freiheitsgrad x in dem bzw. den ersten Schritt(en) beträgt;
• - Änderungen im Freiheitsgrad y auf einen Maximalwert y_max begrenzt, der (betragsmäßig) höchstens 10%, insbesondere höchstens 1%, in einer Ausführung höchstens 0,1%, eines, insbesondere kommandierten oder möglichen bzw. zulässigen, Maximalwerts einer Änderung in diesem Freiheitsgrad y in dem bzw. den ersten Schritt(en) beträgt;
• - Änderungen im Freiheitsgrad z auf einen Maximalwert z_max begrenzt, der (betragsmäßig) höchstens 10%, insbesondere höchstens 1%, in einer Ausführung höchstens 0,1%, eines, insbesondere kommandierten oder möglichen bzw. zulässigen, Maximalwerts einer Änderung in diesem Freiheitsgrad z in dem bzw. den ersten Schritt(en) beträgt;
• - Änderungen im Freiheitsgrad α auf einen Maximalwert α_max begrenzt, der (betragsmäßig) höchstens 10%, insbesondere höchstens 1%, in einer Ausführung höchstens 0,1%, eines, insbesondere kommandierten oder möglichen bzw. zulässigen, Maximalwerts einer Änderung in diesem Freiheitsgrad α in dem bzw. den dritten Schritt(en) beträgt; und
• - Änderungen im Freiheitsgrad β auf einen Maximalwert β_max begrenzt, der (betragsmäßig) höchstens 10%, insbesondere höchstens 1%, in einer Ausführung höchstens 0,1%, eines, insbesondere kommandierten oder möglichen bzw. zulässigen, Maximalwerts einer Änderung in diesem Freiheitsgrad β in dem bzw. den dritten Schritt(en) beträgt; beispielsweise also Änderungen in allen translatorischen und zwei rotatorischen Freiheitsgraden auf einen (verglichen mit den ersten bzw. dritten Schritten) niedrige(re)n Maximalwert begrenzt, und in dem bzw. den dritten Schritt(en)
• - Änderungen im Freiheitsgrad x auf einen Maximalwert x_max begrenzt, der (betragsmäßig) höchstens 10%, insbesondere höchstens 1%, in einer Ausführung höchstens 0,1%, eines, insbesondere kommandierten oder möglichen bzw. zulässigen, Maximalwerts einer Änderung in diesem Freiheitsgrad x in dem bzw. den ersten Schritt(en) beträgt;
• - Änderungen im Freiheitsgrad y auf einen Maximalwert y_max begrenzt, der (betragsmäßig) höchstens 10%, insbesondere höchstens 1%, in einer Ausführung höchstens 0,1%, eines, insbesondere kommandierten oder möglichen bzw. zulässigen, Maximalwerts einer Änderung in diesem Freiheitsgrad y in dem bzw. den ersten Schritt(en) beträgt;
• - Änderungen im Freiheitsgrad z auf einen Maximalwert z_max begrenzt, der (betragsmäßig) höchstens 10%, insbesondere höchstens 1%, in einer Ausführung höchstens 0,1%, eines, insbesondere kommandierten oder möglichen bzw. zulässigen, Maximalwerts einer Änderung in diesem Freiheitsgrad z in dem bzw. den ersten Schritt(en) beträgt; und
• - Änderungen im Freiheitsgrad γ auf einen Maximalwert γ_max begrenzt, der (betragsmäßig) höchstens 10%, insbesondere höchstens 1%, in einer Ausführung höchstens 0,1%, eines, insbesondere kommandierten oder möglichen bzw. zulässigen, Maximalwerts einer Änderung in diesem Freiheitsgrad γ in dem bzw. den zweiten Schritt(en) beträgt

beispielsweise also Änderungen in allen translatorischen und dem anderen der rotatorischen Freiheitsgrade auf einen (verglichen mit den ersten bzw. zweiten Schritten) niedrige(re)n Maximalwert begrenzt.Then are or will be in one or more first step (s)

• - Changes in the degree of freedom α limited to a maximum value α _max , which (in terms of amount) does not exceed 10%, in particular a maximum of 1%, in an execution at most 0.1%, in particular a commanded or possible or permissible, maximum value of a change in this degree of freedom α in one or more third step (s);
• - Changes in the degree of freedom β limited to a maximum value β _max which (in terms of amount) does not exceed 10%, in particular a maximum of 1%, in one embodiment a maximum of 0.1%, of a, in particular commanded or possible or permissible, maximum value of a change in this degree of freedom β in the third step (s); and
• - Changes in the degree of freedom γ limited to a maximum value γ _max which (in terms of amount) at most 10%, in particular at most 1%, in one embodiment at most 0.1%, of a, in particular commanded or possible or permissible, maximum value of a change in this degree of freedom γ in one or more second step (s); For example, changes in all rotational degrees of freedom are limited to a (compared to the second or third steps) low (re) n maximum value, in the second step (s)
• - Changes in the degree of freedom x limited to a maximum value x _max which (in terms of amount) does not exceed 10%, in particular a maximum of 1%, in one embodiment a maximum of 0.1%, of a, in particular commanded or possible or permissible, maximum value of a change in this degree of freedom x in the first step (s);
• - Changes in the degree of freedom y limited to a maximum value y _max which (in terms of amount) does not exceed 10%, in particular a maximum of 1%, in one embodiment a maximum of 0.1%, of a, in particular commanded or possible or permissible, maximum value of a change in this degree of freedom y in the first step (s);
• - Changes in the degree of freedom z limited to a maximum value z _max , which (in terms of amount) at most 10%, in particular at most 1%, in one embodiment at most 0.1%, of a, in particular commanded or possible or permissible, maximum value of a change in this degree of freedom z in the first step (s);
• - Changes in the degree of freedom α limited to a maximum value α _max which (in terms of amount) at most 10%, in particular at most 1%, in one embodiment at most 0.1%, of a, in particular commanded or possible or permissible, maximum value of a change in this degree of freedom α in the third step (s); and
• - Changes in the degree of freedom β limited to a maximum value β _max which (in terms of amount) does not exceed 10%, in particular a maximum of 1%, in one embodiment a maximum of 0.1%, of a, in particular commanded or possible or permissible, maximum value of a change in this degree of freedom β in the third step (s); For example, changes in all translational and two rotational degrees of freedom are limited to a (compared to the first or third steps) low (re) n maximum value, and in the third step (s)
• - Changes in the degree of freedom x limited to a maximum value x _max which (in terms of amount) does not exceed 10%, in particular a maximum of 1%, in one embodiment a maximum of 0.1%, of a, in particular commanded or possible or permissible, maximum value of a change in this degree of freedom x in the first step (s);
• - Changes in the degree of freedom y limited to a maximum value y _max which (in terms of amount) does not exceed 10%, in particular a maximum of 1%, in one embodiment a maximum of 0.1%, of a, in particular commanded or possible or permissible, maximum value of a change in this degree of freedom y in the first step (s);
• - Changes in the degree of freedom z limited to a maximum value z _max , which (in terms of amount) at most 10%, in particular at most 1%, in one embodiment at most 0.1%, of a, in particular commanded or possible or permissible, maximum value of a change in this degree of freedom z in the first step (s); and
• - Changes in the degree of freedom γ limited to a maximum value γ _max which (in terms of amount) at most 10%, in particular at most 1%, in one embodiment at most 0.1%, of a, in particular commanded or possible or permissible, maximum value of a change in this degree of freedom γ in the second step (s)

For example, changes in all translational and the other of the rotational degrees of freedom are limited to a (compared to the first or second steps) low (re) n maximum value.

As mentioned above, the terms “first”, “second” and “third” do not necessarily imply a chronological order.

Due to this (strong) limitation of the changes in the degrees of freedom of the respective subsets, machine learning can be improved in one execution, in particular its speed, convergence, reliability and / or the (learned) behavior of the robot, in particular the execution of a given task by the Robot. By using different subsets, all DoF degrees of freedom can advantageously be used or implemented in one embodiment.

In one embodiment, the steps are, in particular, successive steps, in particular time steps, of the controller or control steps, in particular control clock steps. In the present case, controlling is also understood to mean regulating or commanding on the basis of (specified) setpoint and (recorded) actual values.

The 1 until (Dof - 1) degrees of freedom of a subset Ω = {φ _i , ..., φ _n } of the DoF degrees of freedom {φ ₁ , ..., φ _DoF } and the degrees of freedom {φ ₁ , ..., φ _{(i -1)} , φ _{(n + 1)} , ..., φ _DoF } of the remainder π of the DoF degrees of freedom complementary to this subset result in one embodiment together the DoF degrees of freedom (DoF = number of degrees of freedom of the subset + number of degrees of freedom of the complementary Rest).

In one embodiment, each of the DoF degrees of freedom of the robot is an element of at least one of the residues, so that in each of the DoF degrees of freedom of the robot, a (less limited) change can be commanded in at least one of the steps. In one embodiment, two or more of the subsets each have at least one common degree of freedom in pairs and differ by at least one degree of freedom, which is only an element of one subset but not an element of the other subset.

In one embodiment, the DoF degrees of freedom have, in particular, Cartesian degrees of freedom, in particular an end effector of the robot, in one embodiment a Cartesian position in one, two or three spatial directions and / or an orientation or rotation by one, two or three Axes, for example Euler or cardan angles or the like, in one embodiment all, in particular six and / or Cartesian, degrees of freedom, in particular of an end effector, of the robot or those), in particular Cartesian, degrees of freedom, in particular of an end effector, of the robot that are required to carry out a or the specified task are required, for the execution of which the robot is controlled by means of the method or system. In one embodiment, the Cartesian degrees of freedom are in an environment-proof or robot-based or endeffector-proof reference system. In one embodiment, the DoF degrees of freedom have one or more redundancy parameters, for example an elbow position of a seven-axis robot arm or the like.

If, for example, the robot with its end effector is to place a workpiece at an arbitrary point without changing its orientation, the DoF degrees of freedom can have, in particular, the three translational Cartesian degrees of freedom of the end effector. On the other hand, if the six-axis robot, for example with its end effector, is to insert a workpiece into a receptacle, possibly with a change in orientation, the DoF degrees of freedom can have, in particular, the three translational and three rotational degrees of freedom of the end effector.

By using these degrees of freedom, in one embodiment the robot can be controlled particularly advantageously, in particular reliably (more), quickly (more) and / or intuitively (more).

In one embodiment, the controller commands the changes using (an) operational space control (framework).

As a result, in one embodiment with Cartesian degrees of freedom, the robot can be controlled particularly advantageously, in particular reliably (more), quickly (more) and / or intuitively (more).

In one embodiment, one or more of the remainder of the degrees of freedom (each) have only translational degrees of freedom on. In a further development, one or more of the remainder of the degrees of freedom (each) have only one (in particular different) translational degree of freedom and / or one or more of the remainder of the degrees of freedom (each) only have two or more translational degrees of freedom.

Additionally or alternatively, in one embodiment, one or more of the remainder of the degrees of freedom (each) only have rotational degrees of freedom. In a further development, one or more of the remainder of the degrees of freedom (each) have only one (in particular different) rotational degree of freedom and / or one or more of the remainder of the degrees of freedom (each) only have two or more rotational degrees of freedom.

This is based on the knowledge that, in one embodiment, the division into translational and rotational degrees of freedom can improve machine learning, in particular its speed, convergence, reliability and / or the (learned) behavior of the robot, in particular the execution of a given task the robot.

In one embodiment, the machine learning can be improved, in particular its speed, convergence and / or reliability, by restricting it to only one translational or rotational degree of freedom in at least one of the steps. By combining two or three translational or rotational degrees of freedom in at least one of the steps, the (learned) behavior of the robot can be improved in one embodiment, in particular the execution of a predetermined task by the robot. The combination of three translational degrees of freedom in a remainder and the division of the individual rotational degrees of freedom into a remainder or a combination of two rotatory degrees of freedom has proven to be particularly advantageous.

Additionally or alternatively, in one embodiment, one or more of the remainder of the degrees of freedom have at least, in one development only, one translational and at least, in one development only, one rotational degree of freedom.

By combining, in particular only, one translational and, in particular, only one rotational degree of freedom in at least one of the steps, the (learned) behavior of the robot can be improved in one embodiment, in particular the execution of a predetermined task by the robot.

In one embodiment, one or more of the maximum values to which (in each case) a change in a degree of freedom of one of the subsets is limited to zero, in a further development the corresponding change is multiplied by zero (for this purpose). As a result, the (learned) behavior of the robot can be improved in one embodiment, in particular the execution of a predetermined task by the robot.

Additionally or alternatively, one or more of the maximum values to which a change in a degree of freedom of one of the subsets is limited are not equal to zero, in a further development the corresponding change (for this purpose) is measured with a corresponding sigmoid function or multiplied by a correspondingly small, in particular constant, factor. This can result in particular numerical and / or analytical advantages in machine learning in one embodiment.

A maximum value to which a change in a degree of freedom of one of the subsets in one of the steps is limited is or is not changed in one embodiment as a function of the maximum value of the change in this degree of freedom in the at least one other of the (plurality of) steps or chosen or determined or determined. For example, by multiplying determined changes in the degrees of freedom of the subset by 0.1 or 0.01 or 0.001, if necessary by means of a sigmoid function (s), in one (the) step (s) and multiplying determined changes of these degrees of freedom with (approximately) 1 in the other (the) step (s) the limitation according to the invention to maximum values can be implemented. Correspondingly, in one embodiment, one or more of the maximum values, to which a change in a degree of freedom of one of the subsets is or will be limited, are constant and / or specified before the robot is operated or controlled. In this respect, the reference to maximum values of changes in at least one other of the (majority of) step (s) is understood to be merely a (quantitative) reference variable, without the maximum values being limited to the changes in degrees of freedom of a subset in a step Having to change in other steps with the maximum values of changes in these degrees of freedom.

In one embodiment, changes other than zero are (also) initially determined in the degrees of freedom of the subsets and (only) then limited to the respective maximum value, in one embodiment multiplied by zero or a corresponding sigmoid function or the like. As a result, machine learning, in particular its speed, convergence and / or reliability, can be improved in one embodiment.

In one embodiment, in particular, a sequence of the steps is fixedly predetermined. For example, one or more successive remainders can cyclically initially only have translational degrees of freedom individually or in combination and one or more subsequent remainders each only have rotational degrees of freedom individually or in combination and / or in each case combinations of one or more translational and one or more rotational degrees of freedom.

In a further development, in particular, the sequence of the steps is fixedly specified by or as a function of a user input (“user input dependent”), in one execution before the task is carried out. In one embodiment, the user is shown a selection of the possible steps for this purpose.

As a result, machine learning, in particular its speed, convergence, reliability and / or the (learned) behavior of the robot can be improved in one embodiment.

However, this can mean that in some steps no change is commanded or that this step is unnecessarily just waited for, so to speak.

In one embodiment, a sequence of the steps, in one embodiment dependent on user input, is variably predefined by the controller, in one embodiment while the task is being carried out.

As a result, machine learning, in particular its speed, convergence, reliability and / or the (learned) behavior of the robot can be improved in one embodiment.

In one embodiment, a user input can be used to select or be dependent on whether the sequence of the steps, in particular depending on user input, is to be predefined or variably predefined by the controller. Additionally or alternatively, in one embodiment, the variable specification by the controller can be conditioned by or as a function of a user input or, in turn, be dependent on user input, for example by the user specifying which step sequences are permitted or can be specified by the controller or like that. In one embodiment, the user is shown a corresponding selection or input option for this purpose.

As a result, machine learning, in particular its speed, convergence, reliability and / or the (learned) behavior of the robot can be improved in one embodiment.

In one embodiment, the controller specifies the sequence of the steps by means of one or more selection parameters, in particular cascaded, in a further development by means of a, in particular primary, selection parameter, whether the respective remainder has only translational or only rotational degrees of freedom or not, and / or by means of at least one, in particular secondary, selection parameter, which (group) of the translational or rotational degrees of freedom the respective remainder has.

As a result, machine learning, in particular its speed, convergence, reliability and / or the (learned) behavior of the robot can be improved in one embodiment.

In one embodiment, in particular, the degrees of freedom of one or more of the remainders or the corresponding complementary subset (s) are fixedly predetermined, in one development by or as a function of a user input, in one embodiment before the execution of the Task. For example, a user can specify through appropriate input that a remainder or a subset should have only three translational degrees of freedom or only one or which translational degree of freedom or only one or two or which rotational degree (s) of freedom (s). In one embodiment, the user is shown a corresponding selection or input option for this purpose.

As a result, machine learning, in particular its speed, convergence, reliability and / or the (learned) behavior of the robot can be improved in one embodiment.

Additionally or alternatively, in one embodiment the degrees of freedom of one or more of the remainders or the corresponding complementary subset (s) are variably specified by the controller, in one embodiment depending on user input, in one embodiment while the task is being carried out.

As a result, the (learned) behavior of the robot can be improved in one embodiment.

In one embodiment, a user input can be used to select or be dependent on whether, in particular as a function of user input, fixed predetermined remainders or subsets and / or residues or subsets variably predefined by the controller are used. Additionally or alternatively, in one embodiment, the variable specification of one or more of the remainders or subsets by the controller can be conditioned by or as a function of a user input or, in turn, be dependent on user input, for example by the user specifying which combinations of Degrees of freedom are permissible or the like. In one embodiment, the user is shown a corresponding selection or input option for this purpose.

As a result, machine learning, in particular its Speed, convergence, reliability and / or the (learned) behavior of the robot can be improved.

In one embodiment, reinforcement learning, preferably deep reinforcement learning, in particular a (deep) reinforcement learning algorithm, is used to command the changes by means of the controller and / or to specify the sequence of the steps and / or the degrees of freedom of at least one of the subsets. In particular for this purpose, the controller has in one embodiment at least one artificial neural network, preferably a deep neural network, in particular with several dense layers in one embodiment a convolutional neural network.

For this machine learning, the present idea of a limitation, in particular zeroing or, at least essentially, suppression, of commanded changes in a part of the DoF is controllable, in particular Cartesian, degrees of freedom in different (controller) steps, in particular in (one) operational space Control (framework), particularly advantageous, it being possible through this machine learning, in particular in combination with one or more of the features mentioned here, to improve the (learned) behavior of the robot in a particularly advantageous manner.

In one embodiment, the artificial neural network of the controller has a first sub-network that determines target changes in the DoF degrees of freedom on the basis of input, in particular sensor data, and a second sub-network that uses these input data to determine these target changes in the degrees of freedom the respective subset is limited to the respective maximum value, in particular for this purpose the sequence of the steps and / or the degrees of freedom of at least one of the subsets specifies.

As a result, machine learning can be implemented particularly advantageously, in particular differentiable, in one embodiment, in particular therefore have or use a backpropagation algorithm in one embodiment.

• - Mittel zum Begrenzen von (den bzw. eventuellen) Änderungen in den, insbesondere allen, Freiheitsgraden einer ersten, insbesondere vorgegebenen, Teil- bzw. Untermenge der DoF Freiheitsgrade, die wenigstens einen und höchstens (DoF-1) der Freiheitsgrade aufweist, in einem oder mehreren ersten Schritt(en) einer Mehrzahl von Schritten jeweils auf einen Maximalwert, wobei dieser Maximalwert jeweils (betragsmäßig) höchstens 10%, insbesondere höchstens 1%, in einer Ausführung höchstens 0,1%, eines, insbesondere kommandierten oder möglichen bzw. zulässigen, Maximalwerts einer Änderung in dem jeweiligen Freiheitsgrad in wenigstens einem anderen der (Mehrzahl der) Schritte beträgt, und
• - Mittel zum Begrenzen von (den bzw. eventuellen) Änderungen in den, insbesondere allen, Freiheitsgraden einer zweiten, insbesondere vorgegebenen, Teil- bzw. Untermenge der DoF Freiheitsgrade, die wenigstens einen und höchstens (DoF-1) der Freiheitsgrade aufweist und von der ersten Teilmenge wenigstens teilweise verschieden ist, in einem oder mehreren zweiten Schritt(en) der Mehrzahl der Schritte jeweils auf einen Maximalwert, wobei dieser Maximalwert jeweils (betragsmäßig) höchstens 10%, insbesondere höchstens 1%, in einer Ausführung höchstens 0,1%, eines, insbesondere kommandierten oder möglichen bzw. zulässigen, Maximalwerts einer Änderung in dem jeweiligen Freiheitsgrad in dem jeweiligen Freiheitsgrad in wenigstens einem anderen der (Mehrzahl der) Schritte beträgt,
• - und in einer Weiterbildung Mittel zum Begrenzen von (den bzw. eventuellen) Änderungen in den, insbesondere allen, Freiheitsgraden einer dritten, insbesondere vorgegebenen, Teil- bzw. Untermenge der DoF Freiheitsgrade, die wenigstens einen und höchstens (DoF-1) der Freiheitsgrade aufweist und von der ersten und zweiten Teilmenge wenigstens teilweise verschieden ist, in einem oder mehreren dritten Schritt(en) der Mehrzahl der Schritte jeweils auf einen Maximalwert, wobei dieser Maximalwert jeweils (betragsmäßig) höchstens 10%, insbesondere höchstens 1%, in einer Ausführung höchstens 0,1%, eines, insbesondere kommandierten oder möglichen bzw. zulässigen, Maximalwerts einer Änderung in dem jeweiligen Freiheitsgrad in wenigstens einem anderen der (Mehrzahl der) Schritte beträgt, und
• - einen Regler zum Kommandieren von
  • - (den bzw. eventuellen) Änderungen in den Freiheitsgraden eines zur ersten Teilmenge komplementären ersten Rests der Freiheitsgrade und in einer Ausführung (den bzw. eventuellen) auf den jeweiligen Maximalwert begrenzten Änderungen in den Freiheitsgraden der ersten Teilmenge in dem bzw. den ersten Schritt(en) und
  • - (den bzw. eventuellen) Änderungen in den Freiheitsgraden eines zur zweiten Teilmenge komplementären zweiten Rests der Freiheitsgrade und in einer Ausführung (den bzw. eventuellen) auf den jeweiligen Maximalwert begrenzten Änderungen in den Freiheitsgraden der zweiten Teilmenge in dem bzw. den zweiten Schritt(en),
  • - und in einer Weiterbildung (den bzw. eventuellen) Änderungen in den Freiheitsgraden eines zur dritten Teilmenge komplementären dritten Rests der Freiheitsgrade und in einer Ausführung (den bzw. eventuellen) auf den jeweiligen Maximalwert begrenzten Änderungen in den Freiheitsgraden der dritten Teilmenge in dem bzw. den dritten Schritt(en),
  mittels maschinellen Lernens.

According to one embodiment of the present invention, a system, in particular hardware and / or software, in particular in terms of programming, is set up to carry out a method described here and / or has:

• Means for limiting (the or possible) changes in the, in particular all, degrees of freedom of a first, in particular predetermined, subset or subset of the DoF degrees of freedom, which has at least one and at most (DoF-1) of the degrees of freedom in one or several first step (s) of a plurality of steps each to a maximum value, this maximum value in each case (in terms of amount) at most 10%, in particular at most 1%, in one implementation at most 0.1%, one, in particular commanded or possible or permissible , Maximum value of a change in the respective degree of freedom in at least one other of the (plurality of) steps, and
• - Means for limiting (the or possible) changes in the, in particular all, degrees of freedom of a second, in particular predetermined, subset or subset of the DoF degrees of freedom, which has at least one and at most (DoF-1) of the degrees of freedom and from the first subset is at least partially different, in one or more second step (s) of the plurality of steps each to a maximum value, this maximum value in each case (in terms of amount) at most 10%, in particular at most 1%, in one embodiment at most 0.1%, a, in particular commanded or possible or permissible, maximum value of a change in the respective degree of freedom in the respective degree of freedom in at least one other of the (plurality of) steps,
• - and in a further development means for limiting (the or possible) changes in the, in particular all, degrees of freedom of a third, in particular predetermined, subset or subset of the DoF degrees of freedom, the at least one and at most (DoF-1) of the degrees of freedom and is at least partially different from the first and second subset, in one or more third step (s) of the plurality of steps each to a maximum value, this maximum value in each case (in terms of amount) at most 10%, in particular at most 1%, in one embodiment at most 0.1%, in particular a commanded or possible or permissible, maximum value of a change in the respective degree of freedom in at least one other of the (majority of) steps, and
• - a controller for commanding
  • - (the or possible) changes in the degrees of freedom of a first remainder of the degrees of freedom that are complementary to the first subset and in one embodiment (the or possible) changes in the degrees of freedom of the first subset in the first step (s) limited to the respective maximum value ( en) and
  • - (the or possible) changes in the degrees of freedom of a second remainder of the degrees of freedom complementary to the second subset and limited in one embodiment (the or possible) to the respective maximum value Changes in the degrees of freedom of the second subset in the second step (s),
  • - and in a further development (the or possible) changes in the degrees of freedom of a third remainder of the degrees of freedom, which is complementary to the third subset, and in an embodiment (the or possible) changes in the degrees of freedom of the third subset limited to the respective maximum value in the or the third step (s),
  using machine learning.

• - Mittel zum Kommandieren der Änderung mittels Operational Space Control;
• - Mittel zur, insbesondere benutzereingabenabhängigen, Vorgabe der Reihenfolge der Schritte;
• - Mittel zur Vorgabe der Reihenfolge der Schritte mittels wenigstens eines Auswahlparameters; und/oder
• - Mittel zur, insbesondere benutzereingabenabhängigen, Vorgabe der Freiheitsgrade wenigstens eines der Reste.

In one embodiment, the system or its means or its controller has:

• - Means for commanding the change by means of operational space control;
• - Means for specifying the sequence of the steps, in particular depending on user input;
• - Means for specifying the sequence of the steps by means of at least one selection parameter; and or
• - Means for, in particular user input-dependent, specification of the degrees of freedom of at least one of the remainders.

A means within the meaning of the present invention can be designed in terms of hardware and / or software, in particular a processing, in particular microprocessor unit (CPU), graphics card (GPU), preferably a data or signal connected to a memory and / or bus system, in particular digital processing unit ) or the like, and / or one or more programs or program modules. The processing unit can be designed to process commands that are implemented as a program stored in a memory system, to acquire input signals from a data bus and / or to output output signals to a data bus. A storage 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 or is capable of executing the methods described here, so that the processing unit can execute the steps of such methods and can thus in particular control the robot. In one embodiment, a computer program product can have, in particular, a non-volatile storage medium for storing a program or with a program stored on it, with the execution of this program causing a system or a controller, in particular a computer, to generate a program described here Process or one or more of its steps.

In one embodiment, one or more, in particular all, steps of the method are carried out completely or partially automatically, in particular by the system.

In one embodiment, the system has the robot.

• 1: ein System zum Steuern eines Roboters nach einer Ausführung der vorliegenden Erfindung;
• 2: ein Verfahren zum Steuern des Roboters nach einer Ausführung der vorliegenden Erfindung; und
• 3: einen Teil eines Systems zum Steuern eines Roboters nach einer weiteren Ausführung der vorliegenden Erfindung.

Further advantages and features emerge from the subclaims and the exemplary embodiments. This shows, partly schematically:

• 1 : a system for controlling a robot according to an embodiment of the present invention;
• 2 : a method of controlling the robot according to an embodiment of the present invention; and
• 3 Fig. 3 shows part of a system for controlling a robot according to another embodiment of the present invention.

1 shows a system for controlling a seven-axis robot (arm) 10 with an end effector 11 according to an embodiment of the present invention.

In a modification not shown, the robot can also be another robot, in particular a six-axis robot arm.

A robot controller communicates with the robot (arm) 10 , controls in particular its axle drives and receives current actual values of the joint angle from them q ₁ - q ₇ as well as joint moments T ₁ - T ₇ .

The controller has a regulator for this purpose 20th with an artificial neural network 21st and an operational space control framework 22nd on.

The neural network 21st uses reinforcement learning to provide target values for changes in three translational degrees of freedom x , y , z and target values for changes in three rotational degrees of freedom α , β , γ of the end effector 11 in an endeffector-proof reference system. In a modification that is not shown, a reference system that is robot-based or fixed to the environment can also be used.

In the exemplary embodiment, a fixed, predetermined first subset comprises the three rotational degrees of freedom α , β , γ , the complementary first remainder corresponding to the three translational degrees of freedom x , y , z .

A fixed, predetermined second subset comprises the three translational degrees of freedom x , y , z and the rotational degree of freedom γ , the complementary second remainder corresponding to the two rotational degrees of freedom α , β .

A fixed, predetermined third subset includes the three translational degrees of freedom x , y , z and the two rotational degrees of freedom α , β , the complementary third remainder corresponding to only the rotational degree of freedom γ .

The neural network 21st also provides values for two cascaded selection parameters P ₁ ∈ [0, 1], P ₂ ∈ [0, 1] by means of reinforcement learning.

If the first selection parameter value is in a controller step P ₁ below 0.5, in this control step, for example by (multiplication by) corresponding (n) sigmoid functions generated by the neural network 21st Delivered or determined target values for changes in the three rotational degrees of freedom α , β , γ set to zero, for example multiplied by zero, or limited to a very small maximum value, for example multiplied by 0.1 or 0.01 or 0.001, the target values for changes in three translational degrees of freedom x , y , z however, forwarded unchanged, for example by a factor> 0.9, in particular 1, multiplied, and by means of the Operational Space Control Framework 22nd mapped to corresponding target values for the axes or axle drives ( 2 : Step S10 ).

Is the first selection parameter value in a controller step P ₁ however, at least 0.5 and the second selection parameter value P ₂ less than 0.5 are used in this control step, for example by (multiplication by) the corresponding sigmoid functions generated by the neural network 21st Delivered or determined target values for changes in the three translational degrees of freedom x , y , z and target values for changes in the two rotational degrees of freedom α , β set to zero, for example multiplied by zero, or limited to a very small maximum value, for example multiplied by 0.1 or by 0.01 or by 0.001, the target value for a change in the rotational degree of freedom γ however, forwarded unchanged, for example by a factor> 0.9, in particular 1, multiplied, and by means of the Operational Space Control Framework 22nd mapped to corresponding target values for the axes or axle drives ( 2 : Step S20 ).

Is the first selection parameter value in a controller step P ₁ at least 0.5 and the second selection parameter value P ₂ at least 0.5 are determined in this control step, for example by (multiplication by) the corresponding sigmoid functions generated by the neural network 21st Delivered or determined target values for changes in three translational degrees of freedom x , y , z and the target value for a change in the rotational degree of freedom γ set to zero, for example multiplied by zero, or limited to a very small maximum value, for example multiplied by 0.1 or by 0.01 or by 0.001, the target values for changes in the two rotational degrees of freedom α , β however, forwarded unchanged, for example by a factor> 0.9, in particular 1, multiplied, and by means of the Operational Space Control Framework 22nd mapped to corresponding target values for the axes or axle drives ( 2 : Step S30 ).

In this way, primarily (cf. P ₁ ) Translation and rotation and in a second cascade (cf. P ₂ ) Rotation around the z-axis and rotations around the x- and y-axis separately, whereby the reinforcement learning algorithm itself learns when it should perform which of these types of movement.

• - in einem ersten Schritt eines Zyklus nur die Soll-Werte für Änderungen in drei translatorischen Freiheitsgraden x, y, z unverändert weitergeleitet und die Soll-Werte für Änderungen in den drei rotatorischen Freiheitsgraden α, β, γ auf Null gesetzt bzw. auf einen sehr kleinen Maximalwert begrenzt werden;
• - in einem darauffolgenden zweiten Schritt des Zyklus nur der Soll-Wert für eine Änderung im rotatorischen Freiheitsgrad γ unverändert weitergeleitet und die Soll-Werte für Änderungen in drei translatorischen Freiheitsgraden x, y, z und Soll-Werte für Änderungen in den zwei rotatorischen Freiheitsgraden α, β auf Null gesetzt bzw. auf einen sehr kleinen Maximalwert begrenzt werden; und
• - in einem darauffolgenden dritten, letzten Schritt des Zyklus die Soll-Werte für Änderungen in den zwei rotatorischen Freiheitsgraden α, β unverändert weitergeleitet und die Soll-Werte für Änderungen in drei translatorischen Freiheitsgraden x, y, z und der Soll-Wert für eine Änderung im rotatorischen Freiheitsgrad γ auf Null gesetzt bzw. auf einen sehr kleinen Maximalwert begrenzt werden;

oder zyklisch jeweils

• - in einem ersten Schritt eines Zyklus nur der Soll-Wert für eine Änderungen in einem ersten der drei translatorischen Freiheitsgraden x, y, z unverändert weitergeleitet und die Soll-Werte für Änderungen in den anderen beiden der drei translatorischen Freiheitsgraden x, y, z sowie die Soll-Werte für Änderungen in den drei rotatorischen Freiheitsgraden α, β, γ auf Null gesetzt bzw. auf einen sehr kleinen Maximalwert begrenzt werden;
• - in einem darauffolgenden zweiten Schritt des Zyklus nur der Soll-Wert für eine Änderungen in einem zweiten der drei translatorischen Freiheitsgraden x, y, z unverändert weitergeleitet und die Soll-Werte für Änderungen in den anderen beiden der drei translatorischen Freiheitsgraden x, y, z sowie die Soll-Werte für Änderungen in den drei rotatorischen Freiheitsgraden α, β, γ auf Null gesetzt bzw. auf einen sehr kleinen Maximalwert begrenzt werden;
• - in einem darauffolgenden dritten Schritt des Zyklus nur der Soll-Wert für eine Änderungen in einem dritten, letzten der drei translatorischen Freiheitsgraden x, y, z unverändert weitergeleitet und die Soll-Werte für Änderungen in den anderen beiden der drei translatorischen Freiheitsgraden x, y, z sowie die Soll-Werte für Änderungen in den drei rotatorischen Freiheitsgraden α, β, γ auf Null gesetzt bzw. auf einen sehr kleinen Maximalwert begrenzt werden;
• - in einem darauffolgenden vierten Schritt des Zyklus nur der Soll-Wert für eine Änderungen in einem ersten der drei rotatorischen Freiheitsgraden α, β, γ unverändert weitergeleitet und die Soll-Werte für Änderungen in den anderen beiden der drei drei rotatorischen Freiheitsgraden α, β, γ sowie die Soll-Werte für Änderungen in den drei translatorischen Freiheitsgraden x, y, z auf Null gesetzt bzw. auf einen sehr kleinen Maximalwert begrenzt werden;
• - in einem darauffolgenden fünften Schritt des Zyklus nur der Soll-Wert für eine Änderungen in einem zweiten der drei rotatorischen Freiheitsgraden α, β, γ unverändert weitergeleitet und die Soll-Werte für Änderungen in den anderen beiden der drei drei rotatorischen Freiheitsgraden α, β, γ sowie die Soll-Werte für Änderungen in den drei translatorischen Freiheitsgraden x, y, z auf Null gesetzt bzw. auf einen sehr kleinen Maximalwert begrenzt werden; und
• - in einem darauffolgenden sechsten, letzten Schritt des Zyklus nur der Soll-Wert für eine Änderungen in einem dritten, letzten der drei rotatorischen Freiheitsgraden α, β, γ unverändert weitergeleitet und die Soll-Werte für Änderungen in den anderen beiden der drei rotatorischen Freiheitsgraden α, β, γ sowie die Soll-Werte für Änderungen in den drei translatorischen Freiheitsgraden x, y, z auf Null gesetzt bzw. auf einen sehr kleinen Maximalwert begrenzt werden.

In a modification that is not shown, the sequence of the steps can also be fixed, that is, for example, cyclically in each case

• - In a first step of a cycle, only the target values for changes in three translational degrees of freedom x , y , z forwarded unchanged and the target values for changes in the three rotational degrees of freedom α , β , γ set to zero or limited to a very small maximum value;
• - In a subsequent second step of the cycle, only the target value for a change in the rotational degree of freedom γ forwarded unchanged and the target values for changes in three translational degrees of freedom x , y , z and target values for changes in the two rotational degrees of freedom α , β set to zero or limited to a very small maximum value; and
• - In a subsequent third, last step of the cycle, the target values for changes in the two rotational degrees of freedom α , β forwarded unchanged and the target values for changes in three translational degrees of freedom x , y , z and the target value for a change in the rotational degree of freedom γ set to zero or limited to a very small maximum value;

or cyclically each

• - In a first step of a cycle, only the target value for a change in a first of the three translational degrees of freedom x , y , z forwarded unchanged and the target values for changes in the other two of the three translational degrees of freedom x , y , z as well as the target values for changes in the three rotational degrees of freedom α , β , γ set to zero or limited to a very small maximum value;
• - In a subsequent second step of the cycle, only the target value for a change in a second of the three translational degrees of freedom x , y , z forwarded unchanged and the target values for changes in the other two of the three translational degrees of freedom x , y , z as well as the target values for changes in the three rotational degrees of freedom α , β , γ set to zero or limited to a very small maximum value;
• - In a subsequent third step of the cycle, only the target value for a change in a third, last of the three translational degrees of freedom x , y , z forwarded unchanged and the target values for changes in the other two of the three translational degrees of freedom x , y , z as well as the target values for changes in the three rotational degrees of freedom α , β , γ set to zero or limited to a very small maximum value;
• - In a subsequent fourth step of the cycle, only the target value for a change in a first of the three rotational degrees of freedom α , β , γ forwarded unchanged and the target values for changes in the other two of the three rotational degrees of freedom α , β , γ as well as the target values for changes in the three translational degrees of freedom x , y , z set to zero or limited to a very small maximum value;
• - In a subsequent fifth step of the cycle, only the target value for a change in a second of the three rotational degrees of freedom α , β , γ forwarded unchanged and the target values for changes in the other two of the three rotational degrees of freedom α , β , γ as well as the target values for changes in the three translational degrees of freedom x , y , z set to zero or limited to a very small maximum value; and
• - In a subsequent sixth, last step of the cycle, only the setpoint value for a change in a third, last of the three rotational degrees of freedom α , β , γ forwarded unchanged and the target values for changes in the other two of the three rotational degrees of freedom α , β , γ as well as the target values for changes in the three translational degrees of freedom x , y , z set to zero or limited to a very small maximum value.

3 shows an artificial neural network 100 a controller one to 1 analog system for controlling a robot according to a further embodiment of the present invention, so that reference is made to the above description and only differences are discussed below.

The neural network executing the 3 has an input (s) layer 101 for input, in particular sensor data, for example the joint angles q ₁ - q ₇ as well as joint moments T ₁ - T ₇ , and several layers 102 which these input or sensor data to outputs 103 , for example the changes explained above in the three translational degrees of freedom x , y , z and in the three rotational degrees of freedom α , β , γ , and thus a first sub-network of the artificial neural network 100 form.

In addition, the neural network has further layers in the manner of a side branch 104 that the input or sensor data on so many outputs 105 map the maximum number of individual actions to be carried out, and thus a second sub-network of the artificial neural network 100 form.

This allows the reinforcement learning algorithm to learn flexibly, in particular selectively the values of the outputs via the second subnetwork 103 of the first subnetwork approximately or absolutely unchanged to an output 106 for the Operational Space Control Framework 22nd forward or set approximately or absolutely to zero.

List of reference symbols

10

Robot (arm)

11

End effector

20th

Regulator

21st

artificial neural network

22nd

Operational Space Control

100

artificial neural network

101

Entrance (shift)

102

Layers of the first subnet

103

Outputs of the first subnet

104

Layers of the second subnet

105

Outputs of the second subnet

106

output

q ₁ - q ₇

Joint angle

T ₁ - T ₇

Joint moment

x, y, z

(Changes in) translational Cartesian degrees of freedom

α, β, γ

(Changes in) rotatory Cartesian degrees of freedom

P ₁ , P ₂

Selection parameters

Claims (13)

A method for controlling a robot (10) having a number DoF of controllable degrees of freedom (x, y, z, α, β, γ), the method comprising a plurality of steps, the plurality being greater than or equal to two: - In one or more first of steps (S10) of the plurality of steps, changes in the degrees of freedom of a first subset of the degrees of freedom, which has at least one and at most DoF-1 of the degrees of freedom, are each limited to a maximum value, this maximum value in each case at most 10 % of a maximum value of a change in the respective degree of freedom in at least one other of the plurality of steps; and - In one or more second steps (S20) of the plurality of steps, changes in the degrees of freedom of a second subset of the degrees of freedom, which has at least one and at most DoF-1 of the degrees of freedom and is at least partially different from the first subset, each limited to a maximum value are, this maximum value being at most 10% of a maximum value of a change in the respective degree of freedom in at least one other of the plurality of steps; in which at least changes in the degrees of freedom of a first remainder of the degrees of freedom complementary to the first subset in the first step or the first steps and changes in the degrees of freedom of a second remainder of the degrees of freedom complementary to the second subset in the second step or the second steps by a controller ( 20) can be commanded by means of machine learning. Procedure according to Claim 1 , characterized in that: - in one or more third steps (S30) of the plurality of steps, changes in the degrees of freedom of a third subset of the degrees of freedom which has at least one and at most DoF-1 of the degrees of freedom and of the first and second subset at least partially is different, are each limited to a maximum value, this maximum value being at most 10% of a maximum value of a change in the respective degree of freedom in at least one other of the plurality of steps; changes in the degrees of freedom of a third remainder of the degrees of freedom, which is complementary to the third subset, in the third step or the third steps are commanded by the controller (20) by means of machine learning. Method according to one of the preceding claims, characterized in that the DoF degrees of freedom have Cartesian degrees of freedom (x, y, z, α, β, γ) of the robot. Method according to the preceding claim, characterized in that at least one of the remainder of the degrees of freedom has only one or more translational degrees of freedom (x, y, z) and / or at least one of the remainder of the degrees of freedom has only one or more rotational degrees of freedom (α, β, γ) and / or at least one of the remainder of the degrees of freedom has at least one translational and at least one rotational degree of freedom. Method according to one of the preceding claims, characterized in that at least one of the maximum values to which a change in a degree of freedom of one of the subsets is limited is equal to zero and / or at least one of the maximum values to which a change in a degree of freedom of one of the subsets is is limited, is different from zero. Method according to one of the preceding claims, characterized in that the controller (20) commands the changes by means of operational space control (22). Method according to one of the preceding claims, characterized in that a sequence of the steps is fixed or is variably predetermined by the controller. Method according to the preceding claim, characterized in that the controller specifies the sequence of the steps by means of at least one selection parameter (P ₁ , P ₂ ). Method according to one of the preceding claims, characterized in that the degrees of freedom of at least one of the subsets or at least one of the remainders are fixed and / or the degrees of freedom of at least one of the subsets or at least one of the remainders are variably predefined by the controller. Method according to one of the preceding claims, characterized in that the controller commands the changes by means of reinforcement learning and / or specifies the sequence of the steps and / or the degrees of freedom of at least one of the subsets or at least one of the remainder by means of reinforcement learning and / or at least one artificial neural network (21; 100). Method according to the preceding claim, characterized in that the artificial neural network (100) is a first sub-network (102, 103), which on the basis of input data (q ₁ - q ₇ , T ₁ - T ₇ ) determines target changes in the DoF degrees of freedom, and has a second subnetwork (104, 105) which, based on the input data, has the Target changes in the degrees of freedom of the respective subset limited to the respective maximum value. System for controlling a robot) which has a number DoF of controllable degrees of freedom (x, y, z, α, β, γ), the system being set up to carry out a method according to one of the preceding claims. Computer program product with a program code, which is stored on a medium readable by a computer, for carrying out a method according to one of the preceding Claims 1 to 11 .

Images