JP6742040B1 - Robot controller - Google Patents

Abstract

PROBLEM TO BE SOLVED: To realize flexible control according to a wide range of environments and situations, and to separately manage and modify control of operations according to each environment and situation. SOLUTION: A range of environment/situation in which each control routine that executes an operation (hereinafter, referred to as operation control) can achieve control for achieving an operation purpose is described by a machine learning model and model output (for example, within the range). Whether or not), appropriate operation control in the current situation is selected and executed. [Selection diagram] Figure 1

Description

The present invention relates to control of a robot that has some kind of sensor, senses surrounding objects with the sensor, and includes interaction with other objects (including objects that can also be referred to as “environments”, such as walls and floors). The present invention also relates to the configuration of a control program for management, evaluation verification, and improvement of robot motion control.

An environment where robots (eg, autonomously controlled housework robots, nursing robots, etc.) that perform actions involving interaction with other objects in an environment (open environment) with a degree of freedom in which the environment and tasks are not fixed are placed environments -It is necessary to operate flexibly according to the situation. Also, the operation to be performed varies depending on the task.

Although similar at first glance, the robot arm in the factory workshop only needs to move on a single path (route) for a single object (the environment, situation, and tasks are fixed), Different requirements are required. Also, when considering practical application, aspects such as evaluation verification and improvement will appear.

As an example of the movement, an object placed in various places is moved to another various place, grab an object from the side, move while grasping, place in another place, change direction after grasping, There are things such as passing a stick attached to a hole that an object has, inserting the grasped object into the gap between things, and stacking things.

Also, examples of the environment/situation in which an operation is performed include various places where an object is placed, what kind of object exists in the surroundings, whether the object physically interferes with another object, etc. ..

As an example of considering surrounding objects, a system has been proposed that confirms the existence of surrounding objects (which space the object occupies) and confirms whether or not the robot physically interferes (for example, Patent Document 1). 1).

Further, a system has been proposed in which sensor information is input to a machine learning model, a target device is controlled by the output of the machine learning model, and flexible control is realized (for example, refer to Patent Document 2).

JP, 2005-81445, A JP, 2017-64910, A

The conventional system disclosed in Patent Document 1 described above determines whether or not a surrounding object occupies the space, and operates in an area where no other object exists. Therefore, for example, actions involving physical interactions such as pushing away other objects and placing them at that location, or changes in actions depending on the type of object are not considered, and such actions are included. It is not possible to deal with the control that is performed.

Further, the system disclosed in Patent Document 2 allows flexible control in a wide range of situations by control using a machine learning model, but management, evaluation verification, and correction of operation control for each individual situation. Is difficult.

Further, in the above-mentioned conventional technique, in the correction of the operation control regarding the operation error in one situation, the influence of the correction on the operation of the operation control in another situation is not considered. The modification may affect the behavior of the motion control in other situations. In the situation where the operation was originally established (the situation in which the operation achieving the purpose could be realized), the operation may be changed by the modification, and the operation may not be established. Therefore, it is necessary to evaluate and verify again whether or not the operation is established in all situations.

An object of the present invention is to provide a robot control device capable of individually managing and correcting the control of operation according to each situation while realizing flexible control according to a wide range of environments and situations.

In order to achieve the above object, the invention according to claim 1 has, for each operation purpose, the number of operation controls corresponding to the operation purpose, and the physical structure/shape of the surrounding object or the target robot itself. The physical status including the physical characteristics, the current status of the characteristics that can be changed over time (position, posture, temperature), their relative relationship, and the information (status information) of their combination (status information) are input and given. When the machine learning model that outputs the output value indicating whether the operation is established or not (binary value, continuous value, or multidimensional vector value) in the input operation of the input operation control and the operation purpose and the status information are input. The output value for each of the operation controls corresponding to the operation purpose is acquired by giving the status information to the machine learning models for the number of operation controls corresponding to the operation purpose, and the acquired output values are acquired. An operation control selecting unit that selects an operation control to be executed from the operation controls corresponding to the operation purpose based on an output value; and an operation control executing unit that executes the operation control selected by the operation control selecting unit. The operation control is a control routine for sensing a surrounding object with a sensor and achieving an operation purpose with interaction with another object .

Invention according to claim 2, wherein the machine learning model, for each given operation control, or are constructed individually, be configured of a single model, the operation control other than the operation control to be learned Such learning is performed by using a model capable of learning the model of the target motion control by a method that does not affect the output of the output value in 1.

According to a third aspect of the invention, the status information includes at least the position of the object (the absolute position, relative position), and wherein the posture, the shape, to be included any of the types.

Invention according to claim 4, design specification value of the operation control, the operation result of the actual machine, the human heuristics, or the operation results or we generate within physics simulator, status data - established feasibility data pairs and characterized by further comprising a learning data generating unit for generating as a learning data of the machine learning model.

Invention of claim 5, wherein the learning data generating unit, when generating, using the physical simulator, in addition to the operation control, what state in the operation control define whether the state is an operation established at that operating control given conditions, executes the operation control given to various changing environments (situation information) in the physical simulator, to test whether the operation satisfied, the result, the status data - generated as established whether data pairs characterized in that it.

Invention according to claim 6, when testing the operation established by the physical simulator, the environment (status information) given operational control is established operation, at least one example given, starting the point, test It is characterized in that the environment (situation information) is scanned .

According to the present invention, there is provided a robot control device capable of individually managing and correcting the control of the operation according to each situation while realizing the flexible control according to a wide range of environments and situations.

More specifically, according to the invention described in claim 1, it is possible to flexibly judge the situation in which the operation of the operation control is established, and appropriately execute the operation. In other words, the boundaries of the range of situations where motion control motions are established, which are highly dimensional and extremely complex due to a large variety of large amounts of information, and which are difficult for humans to image or describe manually It becomes easy to describe by using the learning model.

According to the invention described in claim 2, in the machine learning model for judging the situation, addition of a situation judgment concerning a new motion control or correction of a situation judgment concerning an existing motion control does not affect a situation judgment of another motion control. It can be done without any impact.

According to the third aspect of the present invention, the same operational effects as those of the first and second aspects of the invention are exhibited.

According to the invention described in claim 4, the same function and effect as those of the invention described in claims 1 and 2 can be obtained.

According to the invention described in claim 5, it is possible to automatically construct or modify (learn or relearn) a model for situation determination.

According to the invention described in claim 6, when automatically constructing or correcting (learning or relearning) a model for situation determination, the given situation information is used as a starting point for scanning the situation to be tested. As a result, it is possible to efficiently test the range of conditions that hold.

FIG. 1 is a diagram for explaining the basic concept of the present invention. FIG. 2 is a diagram showing details of the basic concept of the present invention. FIG. 3 is a block diagram showing the configuration of the robot controller according to the embodiment. FIG. 4 is a diagram for explaining preprocessing by the robot controller according to the embodiment. FIG. 5 is a diagram illustrating an operation (operation control) of the robot controller according to the embodiment. FIG. 6 is a diagram illustrating a process at the time of maintenance/evaluation verification of the robot control device according to the embodiment. FIG. 7 is a diagram showing a specific example of a machine learning model included in the robot control device according to the embodiment.

Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings.

FIG. 1 is a diagram for explaining the basic concept of the present invention. Here, the “environment/situation range” when the target object is processed by the robot arm is illustrated. In the figure, “NG” exemplifies the “environment/situation range” in which the operation by the robot arm is impossible.

As shown in the figure, the present invention uses the concept of "no failure, range of operable environment/situation". In other words, for each operation purpose and operation control, "the range of operable environment/situation that does not fail" is automatically calculated, and each operation control is managed individually.

Here, the environment/situation means the type and placement of the target object and surrounding objects. It is a conceptually multidimensional space because it consists of various object types, positional relationships, and situations. In addition, depending on the motion, the shape and posture of the robot itself are also included in the environment/situation. Further, when considering the cooperative operation of a plurality of robots, the positional relationship/posture relationship of each robot is also included in the environment/situation.

In the present embodiment, a robot arm will be described as an example, but the present invention can be applied to all types of robots regardless of the shape and structure of the robot.

FIG. 2 is a diagram showing details of the basic concept of the present invention. As shown in the figure, in the present invention, the operation purpose is realized by combining a plurality of operation controls for each operation purpose. Each operation control is managed individually and is independent.

FIG. 2A illustrates an environment/situation conceptual space for an operation purpose A (for example, moving an object). In this example, the operation control A1 is possible in the operable environment/situation range A1. The operation control A2 is possible in the operable environment/situation range A2. In the operable environment/situation range A3, operation control A3 is possible.

FIG. 2B illustrates the conceptual space of the environment/situation for the motion purpose B (for example, changing the posture of the object). In this example, the operation control B1 is possible in the operable environment/situation range B1. In the operable environment/situation range B2, operation control B2 is possible. In the operable environment/situation range B3, operation control B3 is possible.

In the present invention, as shown in FIG. 2, one operation purpose is realized by combining one or more operation controls in consideration of the range where no failure occurs.

FIG. 3 is a block diagram showing the configuration of the robot controller 10 according to the embodiment. The robot control device 10 includes a target motion input unit 12, a current condition detection unit 14, a motion control selection unit 16, a motion control execution unit 18, and an ML (machine learning) that describes the operable condition range of motion control of each target motion. ) Model 20, operation control set 22, various specifications information input section 24 such as target operation/design specification value of operation control/actual machine operation value/empirical rule/operation control/establishment criteria/operation basic situation, situation data/approval It is composed of a data pair generation unit 26, an operation physical simulation unit 28, and an operation status range confirmation unit 30. The robot controller 10 has a model (ML model 20) that describes the operable state range for each motion control, and uses a data sample for learning the model as a design specification value for motion control, an actual machine motion value, a rule of thumb, and physics. It is obtained from a simulator (physical simulation unit 28 of operation) or the like. This processing is automatically performed regardless of the content of operation control.

When the machine learning model is used for the control in the motion control given by the various information input unit 24, the ML model 20 may be integrated into the machine learning model. For example, one type of machine in which the type of input of the machine learning model for motion control and the type of input of the ML model 20 are common, and an output for success/failure is provided separately from the output for motion control. For example, integration into a learning model.

Further, in the case of an individual machine learning model for each operation control, the input and output of each machine learning model do not have to be in the same format.

Further, the following two points can be said regarding the correspondence of the model output in the addition or modification of the operation in the configuration of the single model, instead of the individual machine learning model for each operation control.

(1) "The model structure is fixed and only the output value is changed, and the operation value is added and the situation range is corrected. It does not affect the output value related to the operation control other than the operation control to be learned. In the case of using a model capable of learning a model related to target motion control with a method that has little influence,” the output of the model is, for example, a multidimensional vector, and the addition of motion control uses a new pattern. Learning is performed so that the output value of is output.

(2) “A model that adds motion control and modifies the range of conditions by modifying and expanding the model configuration and modifying the output value, and affects the output value related to motion control other than the learning target motion control. If a model that can learn a model related to target motion control is used with a method that has little or no effect,” for example, by adding an output calculation unit for new motion control as a model configuration, A new operation control is added by changing the calculation structure.

In the following, for simplification of description, an individual machine learning model for each operation control is assumed and described.

FIG. 4 is a diagram for explaining pre-processing by the robot controller 10 according to the embodiment. Here, of the block diagram shown in FIG. 3, only the blocks related to the preprocessing are shown.

In order to learn the ML model 20 in the operation control to be used, the situation data/satisfiability data pair generation unit 26 generates a large number of situation data/satisfiability data pairs as learning data. The generation is based on the design specification value of the operation control, the operation value on the actual machine, the rule of thumb of the operation control creator, and the like. Whether or not to generate in various situations, the operation control and the judgment criterion of whether or not the operation is established, and the basic (representative) situation when the operation is performed are put in the operation physical simulation unit 28 to perform the operation control. It is generated based on the result of operating in various situations.

FIG. 5 is a diagram illustrating an operation (operation control) of the robot control device 10 according to the embodiment. Here, of the block diagram shown in FIG. 3, only blocks related to the operation (operation control) of the robot controller 10 are shown.

The motion control selection unit 16 uses the abstract model of the situation range corresponding to the target motion input from the target motion input unit 12 (the ML model 20 existing for the number of motion controls corresponding to the target motion) of the current situation. The current situation data obtained from the detection unit 14 is entered and the output of each ML model 20 is confirmed (the ML model 20 outputs an inference value indicating whether or not the corresponding operation control is possible).

The operation control selection unit 16 determines from the output value of the ML model 20 which operation control is operable in the current situation (note that a plurality of operation controls may be operable. When the ranges overlap. ), and the operation control is selected from the operation control set 22. Then, the operation control execution unit 18 executes the selected operation control. That is, the current situation is put into the ML model 20, and it is inferred whether or not there is a failure, and the operation control that is not failed is adopted and executed.

FIG. 6 is a diagram illustrating a process of maintenance/evaluation verification of the robot controller 10 according to the embodiment. Here, of the block diagram shown in FIG. 3, only blocks related to maintenance/evaluation verification are shown.

The confirming unit 30 for the range of operable status examines the ML model 20 of each operation control (learned in the pre-processing), what kind of operation control is associated with a certain target operation, and what kind of operation control. It is checked whether the operation control is also associated, and if necessary, a new operation control is created, and the operation control is associated in the procedure of “preprocessing”.

Further, when it is desired to improve the operation control in a certain target operation in a certain situation, the confirmation unit 30 of the range of the operable situation determines (inspects) the operation control associated with the situation, and with respect to the operation control, Make corrections. Since the motion controls are completely separated according to the possible situation range (that is, each motion control is independent), the modification does not affect other motion controls.

In this way, when confirming whether or not the device properly operates in a certain situation (check 1), it is possible to confirm the output of the ML model 20 by the confirmation unit 30 of the range of the operable state. Further, when it is desired to improve the control under a certain situation (check 2), the operation status range confirmation unit 30 may look at the ML model 20 and correct the corresponding operation control.

FIG. 7 is a diagram showing a specific example of the ML model 20 included in the robot control device 10 according to the embodiment. In this embodiment, the ML model 20 uses a DNN (Deep Neural Network) called a Three Dimensional Convolutional Neural Network. The DNN inputs the presence/absence of an object at each position in the space, the type of the object, etc., and outputs whether or not it is operable (OK/NG).

As described above, the robot control device 10 according to the present embodiment satisfies all the following four requirements 1 to 4 for realizing robot control.

・Requirement 1: Flexible control that is suitable for a wide range of environments and situations is possible.

This is possible because the robot control device 10 according to the present embodiment can combine a plurality of operation controls and there is no restriction on the type of control method.

・Requirement 2: Control that considers a physical phenomenon (interference of the target object with other objects, eg, pushing, friction, along, center of gravity balance, etc.) is possible.

This is possible because the robot control device 10 according to the present embodiment makes a determination by a physical simulation.

-Requirement 3: It is possible to clarify the range of environment/situation in which each control routine that executes an operation (hereinafter referred to as operation control) is established (that is, operates so as to achieve the purpose).

This is possible because the robot controller 10 according to the present embodiment models the operable condition range.

-Requirement 4: Each operation control can be individually tuned.

This is possible because the robot control device 10 according to the present embodiment handles motion control individually for each possible situation range.

INDUSTRIAL APPLICABILITY The present invention can be used as a robot control device, in particular, as a robot control device that realizes flexible control according to a wide range of environments and situations, and can individually manage and correct control of operation according to each situation. ..

10 Robot Control Device 12 Target Motion Input Unit 14 Current State Detection Unit 16 Motion Control Selection Unit 18 Motion Control Execution Unit 20 ML (Machine Learning) Model 22 Motion Control Set 24 Various Information Input Unit 26 Status Data/Success/Failure Data Pair generation unit 28 Physical simulation unit of operation 30 Confirmation unit of range of operable status

Claims (6)

For each operation purpose, there are the same number of operation controls as the operation purpose, and the physical state including the physical structure/shape of the surrounding object or the target robot itself, the current state of the characteristic that can change with time (position , Posture, temperature), their relative relations, and combinations thereof (situation information), any situation information is input, and whether or not a given operation control has an action in the input situation ( A machine learning model that outputs an output value indicating a binary value, a continuous value, or a multidimensional vector value,
When the operation purpose and the situation information are input, the situation information is given to the machine learning models corresponding to the number of operation controls corresponding to the operation purpose, so that each of the operation controls corresponding to the operation purpose is given. An operation control selecting unit that acquires the output value and selects an operation control to be executed from among the operation controls corresponding to the operation purpose based on the acquired output value;
An operation control execution unit that executes the operation control selected by the operation control selection unit ;
The robot control device is characterized in that the motion control is a control routine for sensing a surrounding object by a sensor and achieving an operation purpose by interacting with another object . The machine learning model is constructed individually for each given motion control, or even if it has a single model configuration, it affects the output of the output value in the motion control other than the motion control of the learning target. The robot controller according to claim 1, wherein such learning is performed by using a model capable of learning a model of motion control of a target by a method not given. The robot control device according to claim 1, wherein the situation information includes at least one of the position (absolute position, relative position), posture, shape, and type of the object. Design specification value of the operation control, the operation result of the actual machine, the human heuristics, or the operation results or we generate within physics simulator, status data - the establishment whether the data pairs, the learning data of the machine learning model and further comprising a learning data generating unit which generates the robot control apparatus according to claim 1, 2 or 3 wherein. The learning data generating unit, when generating, using the physical simulator, in addition to the operation control, given the conditions under which what state in the operation control to define a is either state operation established at that operating control, in the physical simulator The robot control according to claim 4 , wherein the environment (situation information) is changed variously, and given operation control is executed to test whether or not the operation is established, and the result is generated as the situation data-satisfiability data pair. apparatus. When testing the operation established by the physical simulator, the environment (status information) given operational control is established operation, at least one example given, the point as a starting point, the scanning of the test environment (situation information) The robot controller according to claim 5, which is performed .