WO2023068224A1 - Robot simulation device, robot simulation system, and robot simulation method - Google Patents

Abstract

A robot simulation device (100) comprises: an input generation unit (12A) that generates a set of input parameters according to an input distribution (11D); a simulation execution unit (14A) that evaluates a corresponding output parameter for each generated input parameter; an output distribution estimation unit (13) that estimates output distribution data (14) obtained by accumulating sets of the evaluated output parameters and updates the output distribution data (14) until the output distribution data (14) converges; and a result output unit (16) that outputs the converged output distribution data (14).

Description

ROBOT SIMULATION DEVICE, ROBOT SIMULATION SYSTEM, AND ROBOT SIMULATION METHOD

The present invention relates to a robot simulation device, a robot simulation system, and a robot simulation method.

By simulating the behavior of the working robot on a computer before operating the working robot, problems with behavior in the real world can be discovered in advance.

Claim 1 of Patent Document 1 describes the following information processing apparatus.
Based on the flexible object's weight, length, stiffness, their tolerances, and the direction of gravity, the existence range of the flexible object is calculated, and a probability distribution of at least one of the weight, the length, and the stiffness is calculated. a calculation unit that calculates the existence probability of the flexible object in the existence range;
and a display device that displays the existence range and displays the location with the low existence probability in a lighter color than the location with the high existence probability.

Claim 1 of Patent Document 2 describes the following work motion optimization device for a robot.
a robot unit comprising a robot and a working environment surrounding the robot;
a robot controller unit that operates the robot unit according to an input work command;
a work 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 generation unit that generates work command candidates that are estimated to result in the most favorable behavior 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 generated by generating the work command candidate, and repeats this to generate a new work command, A work motion optimization device for a robot, characterized by obtaining an optimized work command.

According to Patent Literature 1, the tolerance of flexible objects is reflected in the simulation.
According to Japanese Patent Application Laid-Open No. 2002-200021, it is possible to optimize the work command to the robot based on the simulation result of the operation of the robot unit.

JP 2019-70963 A JP 2009-125920 A

　In robot work in an unknown environment such as a disaster site, it is not possible to accurately determine the physical property values and boundary conditions that are the inputs to the simulation. For example, rubble is scattered at the site of a disaster, and the number and size of the rubble can only be understood by actually visiting the site. Therefore, the simulation technology as in Patent Documents 1 and 2, which assumes a known environment although some tolerance occurs, cannot be applied to an unknown environment as it is.

In other words, in a simulation that assumes an unknown environment, the accuracy of the simulation is less important than the results that are robust under that environment (which is reasonable to some extent in any environment) no matter what kind of environment occurs as an unknown environment. is required. A simulation system that supports the flexibility of such an environment is required.

Therefore, the main object of the present invention is to provide a robot simulation corresponding to an unknown environment.

In order to solve the above problems, the robot simulation device of the present invention has the following features.
The present invention includes a distribution input unit that receives, as input distribution data, simulation input parameters that define the motion of a robot;
an input generator that generates a set of input parameters according to the input distribution data;
a simulation execution unit that, for each generated input parameter, evaluates a corresponding output parameter;
an output distribution estimation unit that estimates output distribution data obtained by stacking sets of evaluated output parameters, and updates the output distribution data until the output distribution data converges;
and a result output unit for outputting the converged output distribution data.
Other means will be described later.

According to the present invention, it is possible to provide a robot simulation corresponding to an unknown environment.

1 is a configuration diagram of a robot simulation device according to this embodiment; FIG. FIG. 4 is an explanatory diagram of a simulation execution unit according to the embodiment; 1 is a configuration diagram of a robot simulation system according to this embodiment; FIG. It is a flow chart which shows the operation of the robot simulation device concerning this embodiment. It is an input screen figure of a distribution input part and an actual data input part regarding this embodiment. It is a graph which shows an example of input distribution regarding this embodiment. It is a graph which shows an example of the input parameter regarding this embodiment. It is a table which shows an example of simulation DB regarding this embodiment. It is a graph which shows the process of the simulation result reading part regarding this embodiment. 4 is a histogram generated by an output distribution estimator according to the embodiment; It is output distribution data estimated by an output distribution estimating unit according to the present embodiment. 5 is a graph showing convergence evaluation of output distribution data according to the present embodiment; It is a graph which shows the point cloud data of the experimental result stored in experiment DB regarding this embodiment. 14 is a graph showing a result of Bayesian estimation performed by a Bayesian estimation unit according to the present embodiment from the experiment DB of FIG. 13; FIG. 4 is an explanatory diagram of output distribution data estimated by an output distribution estimating unit according to the embodiment; It is a Gantt chart which shows the processing example of the work process optimization part regarding this embodiment.

An embodiment of the present invention will be described below.

FIG. 1 is a configuration diagram of a robot simulation device 100. As shown in FIG.
The robot simulation apparatus 100 includes, as processing units, a distribution input unit 11A, a real data input unit 11B, a distribution estimation unit 11C, an input generation unit 12A, an output distribution estimation unit 13, a simulation execution unit 14A, and a simulation result It has a reading unit 14B, an experiment result reading unit 14D, a Bayesian estimation unit 14E, a work process optimization unit 15A, and a result output unit 16. FIG.
The storage unit of the robot simulation device 100 stores an input distribution (input distribution data) 11D, input parameters 12B, a simulation DB 14C, an experiment DB 14F, output distribution data 14, and optimum process data 15B.

The distribution input unit 11A accepts input of the input distribution 11D. The input distribution 11D is input data that is varied according to unknown environmental conditions. It should be noted that although the range of possible values of the input distribution 11D can be predicted to some extent, the exact values corresponding to the unknown environment for this simulation cannot be specified.

Instead of the input distribution 11D, the actual data input unit 11B receives input of actual data (measured data) obtained in advance under an assumed environment different from the unknown environment in which the current simulation is performed. For example, in the case of a simulation of debris removal, a model of the debris is created under conditions similar to those of the actual disaster site (the original materials of the debris, how it breaks down, etc.), and the weight of the model is measured as actual data. You may input from the data input part 11B. In general, preparation of actual data is costly, so the number of actual data (100, etc.) is smaller than the number of input parameters 12B (10,000, etc.) to the simulation process described later.
The distribution estimation unit 11C estimates an input distribution 11D from the actual data received by the actual data input unit 11B, using a kernel density estimation method (KDE: Kernel Density Estimation) or the like.

The input generator 12A generates input parameters 12B for the current simulation according to the probability density of the specified input distribution 11D. This generating process is, for example, a process of generating a random number a specified number of times (for example, 10,000 times) by the Monte Carlo method, and the generated random number has a value different from other random numbers.
As the number of generated random numbers increases, the input distribution 11D approaches and accuracy improves, but on the other hand, the amount of simulation calculation also increases. Therefore, it is desirable that the user considers the performance of the computer in advance, and sets so as to generate an appropriate number of random numbers.

The input parameter 12B is a parameter used in this simulation. For example, when the robot grips debris scattered at a disaster site, the weight of the gripped object, the shape of the gripped object, and the gripping task Examples include the coefficient of friction of the ground and the indoor temperature in which the gripping work is performed. These input parameters 12B are parameters that affect the output distribution data 14 (work time, possibility of work such as overturning due to the weight of an object grasped by the work robot, etc.), which is the result of this simulation.

The output distribution estimator 13 performs robust evaluation of the output distribution data 14 by probabilistic simulation for unknown environmental conditions indicated by the input parameters 12B. The output distribution estimator 13 may use any one of the following (Method 1) to (Method 3) as probabilistic simulation means.
(Method 1) The simulation executing section 14A receives the input parameter 12B, executes a simulation on a computer, and returns the result to the output distribution estimating section 13. FIG.

(Method 2) The simulation result reading unit 14B receives the input parameter 12B, refers to the simulation DB 14C, reads the simulation result registered in advance by machine learning or the like, and sends it back to the output distribution estimating unit 13. Therefore, the simulation result reading unit 14B registers the input parameters 12B of the simulation executed in advance by the simulation executing unit 14A and the result thereof in the simulation DB 14C.
(Method 3) The experiment result reader 14D receives the input parameter 12B, acquires distribution data of Bayesian estimation estimated by the Bayesian estimator 14E, and returns the distribution data to the output distribution estimator 13. FIG. Therefore, the Bayesian estimation unit 14E registers, in the experiment DB 14F, actual data (actually measured data) under the experimental environment input in advance by the actual data input unit 11B and experimental results of the actual data.

The output distribution estimation unit 13 uses the kernel density estimation method ( KDE: Kernel Density Estimation) or the like estimates output distribution data 14 by stacking a set of evaluated output parameters.
By outputting the output distribution data 14 on the screen, the result output unit 16 allows the user to comprehend robust simulation results corresponding to an unknown environment, so that the user can utilize them for examining the process of robot work.

Furthermore, the output distribution data 14 may not only be output on the screen as it is, but may also be used as a material for optimizing the process plan of the robot.
The work process optimization unit 15A receives the output distribution data 14 as input, performs process optimization (eg, mixed integer programming, A* algorithm, etc.) from the initial plan of the robot process plan, and saves the result as optimum process data 15B. and
This optimization process is, for example, starting from an initial plan for executing the same work, adjusting individual work times, or changing the order of work without changing the content of the work, thereby optimizing the This is a process that shortens the time.
The result output unit 16 outputs the optimum process data 15B optimized based on the output distribution data 14 in the form of a Gantt chart or the like. As a result, even if there are a plurality of robust tasks corresponding to an unknown environment, it is possible for the user to comprehend a reasonable work process that links them.

FIG. 2 is an explanatory diagram of the simulation execution unit 14A.
The simulation execution unit 14A simulates the task of the work robot 111B clearing up the rubble 111A scattered at the work site. The output parameters (work time and work availability) as the simulation result change depending on the number and size of the rubble 111A specified as the input parameter 12B.
That is, the simulation execution unit 14A evaluates the corresponding output parameter for each generated input parameter 12B.

FIG. 3 is a configuration diagram of the robot simulation system.
The robot simulation system is configured by connecting the computer 900 of the robot simulation device 100 and the input/output device 916 .
Robot simulation apparatus 100 is configured as computer 900 having CPU 901 , RAM 902 , ROM 903 , HDD 904 , communication I/F 905 , input/output I/F 906 and media I/F 907 .
Communication I/F 905 is connected to an external communication device 915 . Input/output I/F 906 is connected to input/output device 916 . A media I/F 907 reads and writes data from a recording medium 917 . Furthermore, the CPU 901 controls each processing unit by executing a program (also called an application or an app for short) read into the RAM 902 . This program can be distributed via a communication line or recorded on a recording medium 917 such as a CD-ROM for distribution.

FIG. 4 is a flow chart showing the operation of the robot simulation device 100. As shown in FIG.
11 A of distribution input parts receive the input of the input distribution 11D (S101).
The input generator 12A generates input parameters 12B (S102).

The output distribution estimation unit 13 branches depending on the type of simulation output acquisition (S111).
(Method 1) If it is "execution of simulation" in S111, the output distribution estimation unit 13 acquires the output of the simulation executed by the simulation execution unit 14A (S112; execution of simulation).
(Method 2) If "obtain simulation" in S111, the output distribution estimation unit 13 obtains the simulation output read from the simulation DB 14C by the simulation result reading unit 14B (S113; simulation acquisition).
(Method 3) If "acquire experiment" in S111, the output distribution estimating unit 13 acquires the experimental result read by the experimental result reading unit 14D (estimation result of the Bayesian estimating unit 14E) (S114: Acquiring experimental result).

The output distribution estimation unit 13 estimates the output distribution data 14 from the results obtained in S112 to S114 (S115; estimation of output distribution).
The output distribution estimator 13 determines whether or not the output distribution data 14 estimated this time has converged (S116, see FIG. 12 for details). If Yes in S116, the process proceeds to S121, and if No, the process returns to S111 to estimate the output distribution data 14 again. That is, the output distribution estimator 13 updates the output distribution data 14 until the output distribution data 14 converge.

The result output unit 16 branches according to the type of output result specified by the user (S121). When the output distribution data 14 is specified (distribution in S121), the result output unit 16 outputs a graph of the output distribution data 14 (S122).
When process optimization is designated (step in S121), the work process optimization unit 15A calculates the optimum process for robot work using the output distribution data 14, and outputs the result to the result output unit 16. (S123).

FIG. 5 is an input screen diagram of the distribution input section 11A and the actual data input section 11B used for the input processing of S101.
The input screen 121 includes input fields for one or more input parameters (two in FIG. 5), an "execute" button for confirming the input after completing the input in each input field, and an "execute" button for discarding the input. a "Finish" button.
Each input field (input parameters 1 and 2) includes the following input fields.
Input column for parameter name: Here, input parameter 1 "grasp object weight" and input parameter 2 "grasp object size" are entered.
・Radio buttons to specify the input data format. Here, the data format of the probability distribution name and its parameter group (mean, standard deviation), the data format that directly specifies the probability distribution by specifying a CSV (Comma-Separated values) file, or the data format that specifies the CSV file and executes One can be selected from three types: a data format in which data is specified directly;
Also, one can be selected from the pull-down menu for inputting the probability distribution name in the input screen 121, and various distributions such as uniform distribution can be selected in addition to the normal distribution of the mountain curve.

The distribution input unit 11A receives an input of the input parameter 1 "grasped object weight" in the data format of the probability distribution name "normal distribution".
The actual data input unit 11B accepts an input in a data format in which a CSV file is designated and actual data is directly designated for the input parameter 2 “grasped object size”. In this case, the distribution estimator 11C estimates the input distribution 11D from the actual data received by the actual data input unit 11B, using a kernel density estimation method or the like.

FIG. 6 is a graph showing an example of the input distribution 11D received by the distribution input unit 11A from the input screen of FIG.
The horizontal axis of the graph 131 is the input parameter 1 “grasped object weight”, and the vertical axis is the probability density of rubble generation at each weight. The distribution input unit 11A may accept a data format that directly specifies the probability distribution of the graph 131 by specifying the correspondence data 132 between the horizontal axis and the vertical axis, such as a CSV file.
Since this simulation assumes an unknown environment, the graph 131 is determined based on past experience, literature, opinions of experts, etc., without assuming a specific environment. For this reason, the horizontal axis of the graph 131 does not have a fixed value, but is distribution data including variations.

FIG. 7 is a graph showing an example of the input parameter 12B.
The horizontal axis of the graph 141 is the data number (No.) of the input parameter, and the vertical axis is the numerical value of the input parameter 1 "grasped object weight" corresponding to the data number. The input generation unit 12A generates a numerical value set (random number set) 142 of gripped object weights that conforms to the input distribution 11D. A graph 141 is a graphical representation of the numerical value set 142 .

FIG. 8 is a table showing an example of the simulation DB 14C.
The table 151 of the simulation DB 14C contains, for each data number (No.) of the input parameter, the input parameter (here, the weight of the gripped object) and the execution result of the simulation for that input parameter (here, work time and work availability). and As a general tendency of simulation, the heavier the gripped object is, the longer the work time is, or the work is rejected.
The simulation executing unit 14A prepares in advance data in the simulation DB 14C to be acquired by the simulation result reading unit 14B in S113.

FIG. 9 is a graph showing processing of the simulation result reading unit 14B.
The horizontal axis of the graph 161 is the input parameter 1 “grasped object weight”, and the vertical axis is the simulation execution result (here, work time) corresponding to the weight. Graph 161 is a curve graph. Each point in the line graph is the coordinates of the point indicated by the combination of "weight" and "working time" in FIG.

Since the curve interpolating between the points of the graph 161 has a continuous distribution with only one input parameter here, the simulation execution unit 14A can obtain it by a simple interpolation method (eg, spline interpolation).
On the other hand, when dealing with multidimensional input parameters or complicated (non-continuous) distribution input parameters, the simulation execution unit 14A is caused to execute an interpolation method in advanced machine learning such as random forest. This is because if the distribution is discontinuous or not smooth, a slight difference in the input parameters may result in completely different values in the output parameters.

The output distribution estimating unit 13 inputs a numerical value ("120 kg" in FIG. 9) for each data number (No.) to the graph 161, obtains an intersection with the graph 161, and calculates an output parameter numerical value (Fig. 9, "350 minutes") (see arrow in Figure 9). That is, if the number of input parameters 12B is 1000, the number of output parameters obtained by the output distribution estimator 13 is also 1000. In addition, in FIG. 9, in order to make the explanation easier to understand, one type of output parameter is obtained from one type of input parameter. On the other hand, in practice, there are many types of input parameters and multiple types of output parameters.

FIG. 10 is a histogram generated by the output distribution estimation unit 13. FIG.
The horizontal axis of the histogram 162 is the output parameter (work time), and the vertical axis is the frequency with which the work time is output. The output distribution estimating unit 13 generates a histogram 162 by accumulating the frequency for each numerical value of the output parameter obtained in FIG. 9 (for example, “350 minutes”).
The output distribution estimation unit 13 generates a histogram 162 as shown in FIG. 10 for the output of the simulation execution unit 14A of (method 1) and the output of the simulation result reading unit 14B of (method 2). On the other hand, for the output of the experiment result reader 14D of (Method 3), instead of the histogram 162, a graph in which the distribution is accumulated is generated (see FIG. 15 for details).

FIG. 11 shows the output distribution data 14 estimated by the output distribution estimator 13 .
The output distribution estimator 13 uses a histogram 162 of the output distribution data 14 as shown in FIG. 10 (a histogram obtained by stacking a set of evaluated output parameters), and uses a kernel density estimation method or the like to gradually smooth out the vicinity of each peak of the histogram 162. Generate a graph 163 of connecting curves. This graph 163 is the estimation result of the output distribution data 14 .
With this output distribution data 14, for example, a robust evaluation of work time can be made such as "90% of the work time is within 350 minutes in any unknown environment".
Although the graph 163 has one peak in the example of FIG. 11, there may be multiple peaks in another example. In that case, it is possible to make an evaluation such as "There is a 20% chance that the work time will be about 350 minutes, and a 70% chance that it will be about 550 minutes."

FIG. 12 is a graph showing convergence evaluation of the output distribution data 14 as an example of S116 in FIG.
The horizontal axis of the graph 164 indicates the number of simulations (the number of times any one of S112, S113, and S114 is executed, also called the number of Monte Carlo methods), and the vertical axis indicates the convergence evaluation value of S116.
The shape of the output distribution data 14 is updated each time the number of times of simulation is increased. Therefore, the output distribution estimation unit 13 compares the difference in graph shape (convergence evaluation value such as KL divergence) between the output distribution data 14 of the simulation results up to the previous time and the output distribution data 14 of the simulation results up to this time. . Then, when the difference becomes sufficiently small (below the threshold in the graph 164), the output distribution estimator 13 determines that the output distribution data 14 has converged (Yes in S116).

The details of how the output distribution estimator 13 creates the output distribution data 14 for (Method 1) and (Method 2) have been described above with reference to FIGS. Next, with reference to FIGS. 13 to 15, the details of how the output distribution estimator 13 creates the output distribution data 14 for (Method 3) will be described.
FIG. 13 is a graph showing point cloud data of experimental results stored in the experiment DB 14F. The horizontal axis of the graph 171 is the input parameter 1 “grasped object weight”, and the vertical axis is the experimental result (here, work time) corresponding to the weight.

FIG. 14 is a graph showing the results of Bayesian estimation performed by the Bayesian estimation unit 14E from the experiment DB 14F of FIG.
The Bayesian estimation unit 14E reads the point cloud data of the graph 171 of FIG. Generate a graph 172 that estimates the variability.

In the example of FIG. 14, the Bayesian estimation unit 14E detects a section 172W sandwiched between two upper and lower wavy lines (a wavy line passing through point P1 and a wavy line passing through point P3) that includes 95% of the point cloud data. is the estimation result of the variation. A solid line passing through the point P2 in this section 172W indicates a point where the appearance frequency of the point cloud data is estimated to peak. By using Bayesian estimation in this way, the probability distribution can be evaluated even with a small amount of measured data.
Then, the output distribution estimating unit 13 obtains the output parameter of the working time corresponding to one input parameter of weight=350 kg, for example, as a range of points "points P1 to P2 to P3" which are intersection points in the section 172W.

FIG. 15 is an explanatory diagram of the output distribution data 14 estimated by the output distribution estimating section 13 based on the distribution read from the graph of FIG. 14 by the experiment result reading section 14D.
The horizontal axis of the graph 173 is the output parameter (work time), and the vertical axis is the frequency (probability) of occurrence of that output parameter.
The output distribution estimating unit 13 generates a distribution curve 173A showing the variation (appearance frequency of point cloud data) of the work time (the peak at 120 minutes) within the section 172W in the range of "points P1 to P2 to P3" obtained in FIG. is added to the output distribution data 14 as a simulation result for one time. This distribution curve 173A is obtained by projecting the output parameter of working time corresponding to one input parameter of weight=350 kg in FIG. 14 onto the vertical axis of FIG.
Similarly, the output distribution estimator 13 creates a graph 173B of the output distribution data 14 by successively superimposing (adding) each simulation result on the output distribution data 14 .

FIG. 16 is a Gantt chart showing a processing example of the work process optimization unit 15A as an example of S123 in FIG.
The Gantt chart 201 of the initial plan, the Gantt chart 202 without optimization, and the Gantt chart 203 with optimization will be described below in this order. Each of the Gantt charts 201 to 203 shows the work processes of the robot R1 in charge of dismantling and grasping the rubble and the robot R2 in charge of transporting the rubble.

The output distribution data 14 as a simulation result is not reflected in the Gantt chart 201 of the initial plan, and a process is registered in which variation in working time, which is a factor of the risk of delaying the work process, is not considered.
Specifically, for the robot R1, a process of dismantling rubble at point A, a process of loading the rubble onto robot R2, and a process of moving to point B are registered. For the robot R2, a process of waiting for the robot R1 to dismantle the rubble, a process of waiting to load the rubble from the robot R1, a process of transporting the rubble, and a process of moving to the point C are registered.
The work time for each process such as the rubble dismantling process at point A is based on the pre-registered average work time value Ta.

In the non-optimized Gantt chart 202, the output distribution data 14 as a simulation result is reflected in the work time of each process. For example, the work time for the rubble dismantling process at point A is obtained by adding the standard deviation (variation) time Tb indicated by the output distribution data 14 to the pre-registered average value Ta of the work time.
Similarly, it is necessary to consider the variation in the output distribution data 14 and change the work other than the rubble dismantling process at the point A. However, in FIG. 16, in order to simplify the explanation, it is shown that the working times of other processes are not changed.

In the non-optimized Gantt chart 202, compared to the Gantt chart 201 of the initial plan, the work time of each process is slightly longer or shorter, but the order of each process does not change, and each process is simply executed. It will be postponed. For example, when considering the variation in work time that occurs in the "demolition of rubble at point A" process of robot R1, the waiting time for "waiting for loading" of robot R2, which is the subsequent process, becomes longer. The work (reference numeral 202C) at time C of robot R2 is also delayed.

The Gantt chart 203 with optimization is optimized to shorten the total work time by partially replacing the processes of the robot R2 from the Gantt chart 202 without optimization.
Specifically, the work process optimization unit 15A determines that the waiting time of the robot R2 will be long because there is a large variation in the "demolition of rubble at point A" by the robot R1 (the risk of delay is high). Recognize from the Gantt chart 202 .

Then, in order to make effective use of the waiting time of the robot R2, the work process optimization unit 15A, before performing the process of waiting for the loading of rubble from the robot R1 at the point A, says, "Move to the point C, where the rubble has already been dismantled." Execute the process of transporting the debris and returning to point A.
As a result, the work at time point C (reference numeral 202C) in the Gantt chart 202 without optimization is moved forward to the work at time point C (reference numeral 203C) before the work at point A in the Gantt chart 203 with optimization. By doing so, the total work time is shortened.
In this way, the work process optimization unit 15A rearranges the order of the processes executed by the robots and rearranges the allocation of processes (work) to the robots, taking into consideration variations (delay risk) in the output distribution data 14. The robot's work plan can be improved by performing "process optimization" that

The robot simulation apparatus 100 according to the present embodiment as described above, based on the input distribution 11D in which the probabilistic simulation input conditions (physical property values, boundary conditions, etc.) are varied for unknown environmental conditions, Output parameters such as work time and work availability are evaluated as output distribution data 14 . As a result, robust evaluation is possible even in an unknown environment in which physical property values and boundary conditions, which are inputs to the simulation, cannot be determined accurately.

In addition, the input generation unit 12A generates an appropriate number of random numbers according to the input distribution 11D by the Monte Carlo method. As a result, the number of simulations within a practical calculation time can be performed.
Furthermore, in addition to the simulation processing performed each time by the simulation execution unit 14A (Method 1), the simulation execution unit 14A can also read the results of simulations that have already been executed (Method 2), or read the results of experiments that have already been input (Method 3). By using it, the number of times of simulation can be suppressed and evaluation can be performed in a practical time.

It should be noted that the present invention is not limited to the above-described embodiments, and includes various modifications. For example, the above-described embodiments have been described in detail in order to explain the present invention in an easy-to-understand manner, and are not necessarily limited to those having all the described configurations.
In addition, it is possible to replace part of the configuration of one embodiment with the configuration of another embodiment, and it is also possible to add the configuration of another embodiment to the configuration of one embodiment.
Moreover, it is possible to add, delete, or replace a part of the configuration of each embodiment with another configuration. Further, each of the above configurations, functions, processing units, processing means, and the like may be realized by hardware, for example, by designing a part or all of them using an integrated circuit.
Further, each configuration, function, and the like described above may be realized by software by a processor interpreting and executing a program for realizing each function.

Information such as programs, tables, and files that realize each function can be stored in recording devices such as memory, hard disks, SSDs (Solid State Drives), IC (Integrated Circuit) cards, SD cards, DVDs (Digital Versatile Discs), etc. can be placed on a recording medium of You can also use the cloud.
Further, the control lines and information lines indicate those considered necessary for explanation, and not all control lines and information lines are necessarily indicated on the product. In fact, it may be considered that almost all configurations are interconnected.
Furthermore, the communication means for connecting each device is not limited to a wireless LAN, and may be changed to a wired LAN or other communication means.

11A distribution input unit 11B actual data input unit 11C distribution estimation unit 11D input distribution (input distribution data)
12A Input generation unit 12B Input parameter 13 Output distribution estimation unit 14 Output distribution data 14A Simulation execution unit 14B Simulation result reading unit 14C Simulation DB
14D Experiment result reader 14E Bayesian estimation unit 14F Experiment DB
15A work process optimization unit 15B optimum process data 16 result output unit 100 robot simulation device

Claims (8)

1. a distribution input unit that receives, as input distribution data, simulation input parameters that define the motion of the robot;
   an input generator that generates a set of input parameters according to the input distribution data;
   a simulation execution unit that, for each generated input parameter, evaluates a corresponding output parameter;
   an output distribution estimation unit that estimates output distribution data obtained by stacking sets of evaluated output parameters, and updates the output distribution data until the output distribution data converges;
   and a result output unit that outputs the converged output distribution data.
2. The input generation unit generates random numbers according to the input distribution data by a Monte Carlo method, and uses the set of random numbers as a set of input parameters,
   2. The robot simulation apparatus according to claim 1, wherein the output distribution estimator estimates the output distribution data by a kernel density estimation method from a histogram obtained by stacking a set of evaluated output parameters.
3. The robot simulation device further has a simulation DB and a simulation result reader,
   The simulation DB stores correspondence data between past input parameters and past output parameters as a result of pre-simulation by the simulation execution unit,
   Instead of the process of evaluating the output parameters by the simulation execution unit, the simulation result reading unit executes the process of reading the corresponding output parameters from the simulation DB for each input parameter generated this time. The robot simulation device according to claim 1.
4. The robot simulation device further has an experiment result reading unit, a Bayesian estimation unit, and an experiment DB,
   The experiment DB stores a set of correspondence data between past input parameters and past output parameters,
   The Bayesian estimation unit creates distribution data of output parameters corresponding to input parameters from a set of corresponding data read from the experiment DB by Bayesian estimation,
   Instead of the process of evaluating the output parameters by the simulation execution unit, the experiment result reading unit reads the distribution data of the corresponding output parameters from the Bayesian estimation unit for each input parameter generated this time,
   2. The robot simulation apparatus according to claim 1, wherein the output distribution estimating unit creates the output distribution data by accumulating distribution data of read output parameters.
5. The robot simulation device further has a work process optimization unit,
   The work process optimization unit refers to the output distribution data created by the output distribution estimation unit, changes the work time of each process executed by the robot, and adjusts the work time after the change to each process. and at least one of a process of rearranging the order of processes and a process of rearranging the assignment of processes to each robot, thereby shortening the total work time of each process executed by the robot. robot simulation equipment.
6. 2. The robot simulation apparatus according to claim 1, wherein the simulation execution unit evaluates a working time of the robot or whether the robot can work or not as an output parameter.
7. 7. An input/output device comprising a robot simulation apparatus according to any one of claims 1 to 6, means for transmitting input data to said distribution input section, and means for receiving output data from said result output section. A robot simulation system comprising: a device;
8. The robot simulation device has a distribution input unit, an input generation unit, a simulation execution unit, an output distribution estimation unit, and a result output unit,
   The distribution input unit receives, as input distribution data, simulation input parameters that define the motion of the robot,
   The input generator generates a set of input parameters according to the input distribution data,
   The simulation execution unit evaluates corresponding output parameters for each generated input parameter,
   The output distribution estimating unit estimates output distribution data by stacking sets of evaluated output parameters, and updates the output distribution data until the output distribution data converges;
   The robot simulation method, wherein the result output unit outputs the converged output distribution data.

Images