JP2009125920A - Robot work operation optimization device - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a robot work operation optimization device automatically determining the best operation command rapidly with the small number of operational trials. <P>SOLUTION: An operation result evaluation section 3 observes the operational result of a robot section 1, and evaluates whether or not the operational result is appropriate. A possible improved operation command generation section 4 generates a possible operation command to be estimated to optimize the operation of the robot section 1, generates a new operation command based on the evaluation result by the operation result evaluation section 3 with respect to the operation state of the robot section 1 by the generation of the possible operation command, and finds out the optimized operation command by repeating the generation of the possible operation command and the generation of the new operation command. <P>COPYRIGHT: (C)2009,JPO&INPIT

Description

  The present invention relates to a robot work operation optimizing apparatus that achieves the shortest tact time without interfering with the work environment in a work trajectory plan for an industrial robot.

An industrial robot is an industrial device mainly for the purpose of automating various factory operations. In such industrial robot work trajectory planning, it is important not to make work mistakes, to interfere with the work environment, and to achieve the shortest tact time, but it is suitable for offline programming that uses a simulator to plan work. For this reason, it was difficult to completely model the physical shape and operation timing of the work environment, and a compromise was made with an appropriate margin.
In order to improve this, work trajectory planning is done by online programming that had been done before the advent of offline programming, but this needs to be performed by the worker, and results obtained according to the skill of the worker, There was a problem that the work time required to obtain the results varied.

  For example, Patent Document 1 exemplifies a robot controller that estimates a robot hand position error using a system model based on a neural network and applies a correction to improve the accuracy. A means for learning by using measured values of errors at some predetermined positions as teacher signals is shown.

  Further, for example, in Patent Document 2, there is a means for performing interference check between a laser welding robot and a welding point using a robot simulation apparatus, exemplifying the interfering portion as a programming operator, and supporting adjustment of the moving speed of the robot. It is shown.

JP-A-9-319420 JP 2006-344052 A

  However, in the above conventional technique, an operator needs to select a teacher signal used for learning a system model (a plurality of input / output pairs for the system), but the teacher improves the learning speed and the estimation accuracy. The selection of signals is difficult, and in order to improve the estimation accuracy, it is necessary to prepare a large number of teacher signals, and accordingly, the learning speed tends to become slower. In addition, the operator's intervention is indispensable for the correction of the robot moving speed, and the system model includes a margin that does not match the actual system, so there is a problem in the optimality of the solution.

  The present invention has been made to solve the above-described problems, and in a work trajectory plan of an industrial robot, a robot capable of automatically determining the best work command quickly and with a small number of work trials. The object is to obtain a work motion optimization device.

  The robot work operation optimizing device according to the present invention observes the operation result of the robot unit and evaluates whether the operation result is appropriate, and the operation of the robot unit is most preferable. An improved work command candidate generation unit that generates estimated work command candidates, and the improved work command candidate generation unit evaluates the work result evaluation unit for the operation state of the robot unit caused by generating the work command candidates. Based on the result, a new work command is generated and repeated to obtain an optimized work command.

  The robot work operation optimizing device according to the present invention generates work command candidates that are estimated to be the most preferable operation of the robot unit, and the evaluation result of the work result evaluation unit for the operation state of the robot unit by this is generated. Based on this, a new work command is generated and repeated, so that the best work command can be automatically determined quickly with a small number of work trials.

Embodiment 1 FIG.
The illustrated apparatus includes a robot unit 1, a robot controller unit 2, a work result evaluation unit 3, an improved work command candidate generation unit 4, and a work instruction unit 5, and the improved work command candidate generation unit 4 includes a system model unit. 6, a system model reconstruction unit 7 and a work command candidate optimization unit 8 are provided.
The robot unit 1 is an industrial device that includes a robot and a surrounding work environment, and executes various automatic tasks that are targets of the present invention. The robot unit 1 is used, for example, for automating various operations in a factory such as bin picking work for picking up individual parts from piled parts, picking up parts, assembling work, painting work, and welding work.

  The robot controller unit 2 supplies electric energy to a motor or the like attached to the robot in order to realize the work content indicated by the inputted work command by actually moving the robot. The supplied energy is controlled so that the robot moves appropriately. For this purpose, the operation state of the robot is constantly monitored by using a feedback signal from a sensor attached to the robot or a sensor attached to the outside, and a control operation is executed.

The work result evaluation unit 3 receives a work command for the robot controller unit 2 and an actual movement of the robot unit 1 and inputs an evaluation value such as a combination data thereof and a magnitude of an error thereof to an improved work command candidate generation unit. 4 is output.
Based on the data from the work result evaluation unit 3, the improved work command candidate generation unit 4 outputs an improved work command to be next instructed to the robot. That is, the improved work command candidate generating unit 4 generates a work command candidate that is estimated to be the most preferable operation of the robot unit 1 and the operation state of the robot unit 1 due to the generation of the work command candidate. A new work command is generated based on the evaluation result of the work result evaluation unit 3, and an optimized work command is obtained by repeating this.
The work instruction unit 5 is configured to output a work command to the robot controller unit 2 based on the improved work command output from the work instruction unit 5.

  Based on the data from the work result evaluation unit 3, the system model unit 6 in the improved work command candidate generation unit 4 models information such as the robot itself, the shape information of the surrounding environment where the robot is installed, and the operation timing of the peripheral device. Yes. Since it is a model, it is configured to be able to estimate the output for a certain input. For example, when a robot tact time is modeled based on interference check of surrounding environment objects of the robot or a certain work instruction, the estimated tact time is estimated when a work command that does not actually measure the work result is given. Can be output.

  The system model reconfiguration unit 7 updates the internal state (system model) of the system model unit 6 to a new one based on the data from the work result evaluation unit 3. By this operation, the prediction accuracy of the system model unit 6 is improved for a wider range of inputs.

  The work command candidate optimizing unit 8 is configured to use the system model unit 6 to select one or more work candidates to be performed next by the robot and to output the selected work candidates to the work instruction unit 5. As this selection method, for a large number of work command candidates, the work command candidates use the system model unit 6 to quantify the degree to which the estimation accuracy of the system model unit 6 is expected to be more accurate, or The degree to which the time is expected to be shortened is quantified, and one or a plurality of work command candidates whose numerical values indicate the most preferable values are selected. A shorter tact time is preferable, and even a small estimation error is better, so the minimum value is selected. At this time, since there are a plurality of quantified degrees, a method is used in which, for example, several tact times are selected in the order of decreasing values, and then those having a small estimation error are selected. Alternatively, a method is used in which numerical values of two degrees are multiplied, and the result of the multiplication is selected in ascending order.

  The robot work operation optimizing device of the present invention is realized by using a computer, and the robot controller unit 2 to the operation command candidate optimizing unit 8 execute software and software corresponding to each function. It consists of hardware such as CPU and memory.

Next, the operation of the first embodiment will be described.
FIG. 2 is an explanatory diagram showing the operation content related to the candidate selection.
In the model of FIG. 2, a region (current interference identification boundary) f where the robot is estimated to interfere with surrounding objects is drawn. The values on the horizontal and vertical axes indicate work command parameters. The tact time estimation model is represented by F (S _{n + 1} ). S _n is a set of observation results of n times of operations performed n times. F (S _n ) and f (S _n ) represent models that give the most appropriate estimation result from a set of observation results of n operations.

F (S _{n + 1} ) and f (S _{n + 1} ) are estimation models that are updated when the candidate S _{n + 1} to be worked on next is added to the set of observation results. Since it is unclear when interference occurs due to the candidate S _{n + 1} and when interference does not occur, the difference A _fh , the current f (S _n ) resulting from each case is assumed, assuming both cases. A _fg is numerically expressed by the following equation, and a candidate S _{n + 1} that _maximizes the numerical value is finally selected. In FIG. 2, the identification boundary h is a new identification boundary that is observed when interference occurs when the work is performed under a certain condition based on the identification boundary f, and the identification boundary g is It is a new identification boundary that is observed if it does not occur.

を計算しても良い。尚、１００点というのは単なる例であり５０点でも１０００点でも効果は変わらない。また、候補は最終的に一つに絞らなくても、複数選んで、複数を作業結果評価部３で評価しても同等の効果が得られる。 In the above formula, multiplication is used, but F (S _{n + 1} ) is first calculated and about 100 are selected from the smallest ones.

May be calculated. Note that 100 points is merely an example, and the effect does not change at 50 points or 1000 points. Even if the candidates are not finally limited to one, even if a plurality of candidates are selected and evaluated by the work result evaluation unit 3, the same effect can be obtained.

  Although it is a candidate of a work command, for example, it is given as an enumeration of waypoints in the operation of the robot. However, since there are an infinite number of points in space, they are discretized at appropriate intervals. As a result, selectable candidates are a finite set of points, and each candidate is one of them.

F (S _{n + 1} ) and f (S _{n + 1} ) include, for example, multiple regression analysis, support vector machine, support vector regression, kernel regression, Bayes point machine, Bayesian estimation, maximum likelihood estimation, response surface, radius basis A combination of a function model and a parameter determination algorithm such as a function network and a neural network is used.

FIG. 3 is a flowchart showing the operation of the robot work movement optimizing apparatus shown in FIG.
First, in step ST11, the robot unit 1 performs an initial operation experiment under several work command conditions. Next, in step ST12, the system model reconfiguration unit 7 determines the interference identification boundary of the system model unit 6 from the operation result obtained at that time (operation result for each of a plurality of experimental points). Next, in step ST13, the work command candidate optimization unit 8 sets a point in the vicinity of the identification boundary determined in step ST12 as the next experimental candidate point. Then, it is determined whether or not all candidate points have already been experimented (step ST14). If so, the process ends. If not, the process proceeds to step ST15. In step ST15, the system model unit 6 calculates the above formula (1) or formula (2) for all generated candidate points. Next, in step ST16, the work command candidate optimization unit 8 selects a candidate point at which this value is maximized as the next experimental point. Next, in step ST17, an operation experiment is actually performed using the experimental point selected in step ST16, and an operation result is obtained. And it returns to step ST12 and repeats these processes below.

By the way, the large value of the sum of A _fh and A _fg calculated in this embodiment / (difference between A _fh and A _fg +1.0) means that interference occurs as a result of the work by the candidate points. That is, the amount of change in the identification boundary is almost the same when the interference occurs and when the interference does not occur, and the amount of change is large. That is, the difference value (denominator) becomes smaller if the amount of change becomes equal, and the sum value (numerator) becomes larger if the amount of change is large. From these points, as the candidate of the next experimental point, it is possible to select a point that is likely to be predicted that the identification boundary will correctly pass therethrough.

  In the above embodiment, the value of 1.0 in the sum / (difference + 1.0) is avoided in order to prevent the sum / difference value from becoming infinite when the difference is 0. It may be a numerical value other than 1.0. However, if the value is decreased, the effect of the difference is greatly evaluated, and if the value is increased, the effect of the difference is evaluated small. Therefore, a value is appropriately selected according to conditions and the like.

Also, when constructing the system model unit 6 first, it is necessary to make the S _n. This is an experiment using an orthogonal table. Not only L8 but also L4, L16, L32, L64, L9, L27, L81, L243, L12, L18, L36, L72, etc. may be used, or multi-way layout, Latin square, Greco Latin square, A uniform plan, an optimum plan such as D-optimum, G-optimum, and A-optimal, a composite plan in response surface plan, a Box and Behnken plan, or a completely random plan may be used as appropriate. Alternatively, select the work order to consider the operator experience to a promising, it may have been made a S _n.

  When the next work command candidate is generated, it is checked whether the previous experimental points are uniformly distributed. If not, the candidate points that have not been tested yet are found. The experiment may be performed by randomly selecting one of the points. For example, whether or not the experimental points so far are uniformly distributed in the space where the experiment can be performed is tested by the chi-square test. If the test points are unevenly distributed as a result of the test, one of the lattice points not yet tested is selected at random. In this case, in order to search the searchable space evenly, even if the correct interference identification boundary is scattered like a small island, the speed at which the identification boundary converges to the correct position should be increased. There is an expected effect.

When using a support vector machine as a model, several parameters of the support vector machine can be adjusted. Simply put, the expressive power of the identification boundary can be adjusted. For example, when it is desired to draw a complicated identification boundary, a certain parameter value may be increased. However, as a disadvantage, the time required to process the support vector machine formula increases. If the value of the parameter is reduced, the calculation time is shortened, but it may not be possible to create an identification boundary that can correctly identify the results of all experimental points.
Therefore, the value of the corresponding parameter is initially made small to reduce the calculation time. And in the step of determining the identification boundary, the value of the corresponding parameter can be completely separated when it becomes impossible to obtain the identification boundary that completely separates the result of the experimental points obtained up to that point. Configure to increase until At the same time, it is configured to reduce until just before it can not be completely separated. Accordingly, there is an effect that it is possible to obtain a robot work motion optimization device that requires a short calculation time for selecting an experimental point candidate.

  As described above, according to the robot operation optimization device of the first embodiment, the robot unit including the robot and the surrounding work environment of the robot, the robot controller unit that operates the robot unit according to the input operation command, An operation result evaluation unit that observes the operation result of the robot unit and evaluates whether the operation result is appropriate, and an improvement operation that generates candidate operation commands that are estimated to be the most preferable operation of the robot unit A command candidate generation unit and a work instruction unit that outputs a work command to the robot controller unit based on the work command candidate generated by the improved work command candidate generation unit. A new work command is generated based on the evaluation result of the work result evaluation unit for the operation state of the robot unit due to the generation of the command candidate, By repeating Les, since to obtain the optimized work order, it is possible to automatically determine quickly best work command with less work number of trials.

  In addition, according to the robot work operation optimizing apparatus of the first embodiment, the improved work command candidate generation unit performs the interference check between the robot itself and the surrounding environment object of the robot and the calculation of the work operation time in the robot unit. Using a system model unit that has a system model, a system model reconfiguration unit that updates the system model using an evaluation result based on observation of the operation result of the robot unit from the work result evaluation unit, and a system model at a certain point in time It has a work command candidate optimization unit that generates candidate work commands that are estimated to be the most favorable motion of the robot part, so it selects experiment point candidates based on the system model that reflects the actual motion results Therefore, the best work command can be quickly determined. Even if the robot is operated in a so-called robot simulator and the interference check is performed by performing an interference check operation between the environment model built in the simulator and the robot model, A distance sensor is attached to each hand or arm, or a joint torque sensor or joint torque estimation calculation is used to detect that the actual robot is in contact with the environment and check for interference. Even if configured so as to achieve the same effect, the same effect can be obtained.

  Furthermore, while the above has been described with respect to an example where the robot avoids interference with obstacles, other forms may be employed. In a system in which a robot and a vision sensor are combined, when a vision sensor installed in a fixed manner captures an object, the robot needs to move to a position where the body of the robot does not enter the field of vision of the vision sensor. In other words, the robot needs to move out of the field of view of the vision sensor and be on standby so as not to interfere with the imaging of the vision sensor. When the imaging of the grasped object by the vision sensor is completed, the robot moves again within the field of view of the vision sensor and performs a grasping operation in order to grasp the work object. At this time, it is considered that it is advantageous for shortening the entire working time that the robot is waiting outside the visual field of the vision sensor. That is, it is possible to operate in a short tact time while avoiding the interference of the vision sensor by processing, assuming that it is the same judgment as the interference check described above whether it is in the vision sensor field of view or not. There is an effect that an orbit is obtained.

Embodiment 2. FIG.
Since the configuration of the second embodiment on the drawing is the same as that of the first embodiment, FIG.
A robot work command is usually described in a kind of programming language called a robot language. The grammar often follows that of the BASIC language. Robot work is assigned to each line of those languages. Information other than the robot operation such as numerical calculation is also described, but here, only the description regarding the robot operation and the command regarding the communication with the peripheral device are focused.

In the present embodiment, each time the robot unit 1 performs work, the work result evaluation unit 3 records the number of times of execution and the time required for execution of each work. This recording is performed with the actual robot operating in the actual environment. Based on the record, the improvement work command candidate generation unit 4 calculates and calculates the product value of the time required for execution × the number of executions, and arranges them in descending order.
Note that this ranking can be presented to the operator and is out of the viewpoint of automatic optimization. However, it is better to replace the work at the top of the ranking from robot work to dedicated tool work or dedicated processing equipment. You can also consider whether.

  In the present embodiment, the improvement work command candidate generation unit 4 is based on information on the execution time of each operation procedure observed by the work result evaluation unit 3. As long as there is no hindrance to work execution, the operation procedure is changed so as to reduce the tact time in the entire operation, and a work command is generated. Specifically, it is sequentially determined whether or not the following items are applicable, and when they are applicable, the work operation is actually improved.

(1) Optimization of operation with high execution frequency Using the method described in the first embodiment, work operation with high execution frequency is improved toward higher-speed operation. This corresponds to, for example, realizing an operation that passes through an obstacle.

(2) Optimization of compound branch For example, in the situation where the work arrival waiting position from the upstream process is programmed and the actual position arriving from the upstream process varies, it is statistically Probability distribution of arrival positions is obtained. The workpiece arrival waiting position is moved to the center of this distribution.
Alternatively, when there are a plurality of types of workpieces that arrive and the hands that can handle them are different, it is necessary to change the hands. As the work operations are repeated, the probability distribution of the actual arrival type and the conditional probability regarding the arrival order are obtained statistically. According to this distribution, the robot program is rewritten so that the hand is attached in advance.

(3) Optimization of conditional branch decision processing It may take some time to judge the situation, such as inspection after machining. The probability is statistically determined whether the observed branching probability is true or false. In accordance with this probability, the operation is switched to an inspection operation that can be quickly determined. Alternatively, when there are a plurality of inspection processes, an inspection sequence with a high probability of a comprehensive final determination result is generated, and a work operation for realizing this is reprogrammed.

(4) Execution order exchange Based on the observed work result information, the execution order is exchanged in order that finishes earlier. The order is changed so that the work that can be done first is completed during the work completion waiting time from the peripheral device. Whether it is meaningful to exchange can be judged from each observed work time.

(5) Work in-line The work that needs to be changed can be done in advance by placing multiple workpieces in the buffer area with the same hand in advance, and assembling and reusing jigs that are used many times. Rewrite the robot program.

(6) Resource allocation optimization Optimization of temporary table use, optimization of finger assignment when using double hands, and optimization of dedicated machine processes such as pressing.

The improved work command candidate generation unit 4 outputs the result of performing the above items to the work instruction unit 5 as an improved work command. A work command is output from the work instruction unit 5 to the robot controller unit 2, and the robot unit 1 operates.
Although it is an end condition for repetition, when the tact time is not updated for a long time, when the magnitudes of A _fh and A _fg become smaller than a predetermined value in the operation in the work command candidate optimization unit 8 I will end it.

  As described above, according to the robot work motion optimizing device of the second embodiment, the improved work command candidate generation unit creates a work command by enumerating a series of motion procedures in the robot programming language, Based on the information on the execution time of each operation procedure observed by the result evaluation unit, unless there is a hindrance to work execution, the operation procedure is changed so as to shorten the tact time in the entire operation so that a work command is generated. As a result, there is an effect of obtaining a robot work motion improvement device capable of automatically determining the best work command quickly and with a small number of work trials.

BRIEF DESCRIPTION OF THE DRAWINGS It is a block diagram which shows the robot operation | movement optimization apparatus by Embodiment 1 of this invention. It is explanatory drawing which shows the operation | movement of the interference identification boundary and candidate selection of the robot operation | movement optimization apparatus by Embodiment 1 of this invention. It is a flowchart which shows operation | movement of the robot operation | movement optimization apparatus by Embodiment 1 of this invention.

Explanation of symbols

  DESCRIPTION OF SYMBOLS 1 Robot part, 2 Robot controller part, 3 Work result evaluation part, 4 Improvement work command candidate generation part, 5 Work instruction part, 6 System model part, 7 System model reconstruction part, 8 Work command candidate optimization part.

Claims (3)

A robot section composed of a robot and a surrounding work environment of the robot;
A robot controller for operating the robot in accordance with the input work command;
An operation result evaluation unit that observes the operation result of the robot unit and evaluates whether the operation result is appropriate;
An improved work command candidate generating unit for generating a work command candidate that is estimated to be the most preferable operation of the robot unit;
A work instruction unit that outputs a work command to the robot controller unit based on the work command candidate generated by the improved work command candidate generation unit;
The improved work command candidate generation unit generates a new work command based on the evaluation result of the work result evaluation unit for the operation state of the robot unit caused by generating the work command candidate, and repeats this, A robot work motion optimizing device characterized by obtaining an optimized work command. The improvement work command candidate generator is
A system model unit having a system model for performing interference check and calculation of work operation time between the robot itself and the surrounding environment object of the robot in the robot unit;
A system model reconfiguration unit that updates the system model using an evaluation result based on observation of an operation result of the robot unit from a work result evaluation unit;
The work command candidate optimizing unit for generating a work command candidate that is estimated to be the most preferable operation of the robot unit using a system model at a certain point in time. Robot work motion optimization device.   The improved work command candidate generation unit creates a work command by enumerating a series of operation procedures in the robot programming language, and based on the execution time information of each operation procedure observed by the work result evaluation unit. 2. The robot work motion optimizing device according to claim 1, wherein a work command is generated by replacing the operation procedure so as to shorten a tact time in the entire motion, as long as the execution is not hindered.

Images