JP2013054703A - Work support system - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a work support system, when a worker shifts to a next work earlier than a predetermined time in performing each predetermined work, eliminating a time loss of the worker in a support operation and actuating supply means concurrently with the movement of the worker, thus improving working efficiency.SOLUTION: The work support system includes: prediction means 10 for predicting behavior of a worker using Markov model generated from moving data of the worker; and supply means 20 for supplying the worker with components, tools, and other necessary objects on the basis of the behavior of the worker that has been predicted by the prediction means 10.

Description

  The present invention relates to a work support system that predicts a current work situation for a worker who performs a series of work, and supplies necessary items such as tools and parts to a worker's hand based on the predicted result. .

  Factory automation with industrial robots aimed at improving production efficiency is progressing. However, there are many tasks in the factory that are difficult for robots to perform, such as assembling parts that are difficult to handle by robots, tasks that require skillful adjustment, and tasks that require situational judgment. However, there are still many processes that require human intervention and are not automated.

  An example of this is an automobile assembly process. In such a process, the worker must select an assembly part such as a bolt and take out a tool for assembling the part, in addition to the part assembly work on the vehicle body. Since wagon carts that store and store parts and tools are located at the edge of the work space, workers must move between the work site and the wagon cart to supply parts and change tools. . In this way, the worker must perform work associated with the work in addition to the original assembly work, and the worker is forced to spend a lot of time and effort on these work.

  Therefore, the present inventors have researched and developed a work support system that delivers necessary tools and parts to a worker's hand using a robot arm according to the work (for example, Patent Document 1 and Non-Patent Document 1). Hereinafter, the assembly process of an automobile will be described as an example.

  The worker needs to perform a lot of work, and each work is summarized for each work content in the work process chart. The work process table also defines a work place designated by a position with respect to the vehicle body in each work, parts and tools necessary for the work, and the like. Therefore, the worker finishes the work contents for one vehicle body within a predetermined time, and shifts to the assembly work to the next vehicle body.

  Although workers work faithfully in the work schedule, there are actually differences among workers. Therefore, in order to realize cooperative work between the worker and the robot, it is necessary for the robot to deliver the component and tool to the hand by adaptively changing the supply time and supply location of the component and tool according to the worker.

  Therefore, the present inventors have considered that the robot supports according to the worker based on the statistical analysis result of the worker's movement as follows. In this system, the vehicle body coordinate system is discretized with a fixed-width mesh, and the ratio of the workers measured in each cell is expressed using three types of statistical models.

  The first statistical model uses a probability (referred to as “the presence rate of the worker with respect to the position”) that expresses in which place (cell) the worker is working in each work. Thus, a place (cell) that is highly likely to perform each work is shown. Thereby, the supply position of components or tools is determined.

  The second statistical model obtains the probability of performing each work at a certain time (referred to as “work execution rate with respect to time”) in order to determine the supply time.

  The third statistical model uses the probability (referred to as “work execution rate for position”) that indicates which work is likely to be performed from the position of the worker. The work content is estimated from the position, and the work progress is estimated.

  By using this third statistical model, when the worker moves to the next work place earlier than the next supply time, it recognizes that the worker has moved from the worker's position to the next work place. The supply time is corrected, and parts and tools are supplied to the worker in accordance with the work progress of the worker earlier than the supply time before the correction.

  In this system, the actual time is corrected for the supply time in accordance with the work progress. As a result, when the work progress of the worker is proceeding at a faster pace than usual, the parts and tools can be supplied to the worker ahead of the original supply time.

WO2010 / 134603A1

Yasushi Tanaka, Jun Kinukawa, Yuta Kawai, Yusuke Hagiwara, Kazuhiro Ogura, “Human Cooperation / Coexistence Work Support Partner Robot at Production Site -PaDY-,-6th Report: Proposal of Worker Position Estimation Method Using Particle Filters- ”Proceedings of the 11th Society of Instrument and Control Engineers System Integration Division, 1E1-2, Miyagi, December 2010

  In the system disclosed in Patent Document 1, it is possible to estimate the work progress of the worker using three statistical models and to perform work support in accordance with the worker. However, the correction of the supply time is realized by estimating the work progress from the place where the worker exists using the “work execution rate with respect to the position”. Therefore, in this method, it cannot be estimated that the worker has moved to the next work until the moment when the worker finishes moving to the place where the next work is performed. In other words, since the robot arm starts to operate immediately after the movement of the worker is completed, the worker must wait for the time for the robot arm to move to the supply location with respect to the time at which the worker originally wants to receive the parts and tools. Don't be.

  Therefore, in the present invention, when the worker moves to the next work earlier than a predetermined time when performing each predetermined work, the time loss of the worker in the support operation is eliminated, and the movement of the worker is adjusted. An object of the present invention is to provide a work support system that operates a robot arm or other supply means to improve work efficiency.

  Based on the history of research and development of work support systems, the present inventors examined adopting a Markov model having a state transition probability as a technique for predicting the behavior of a worker who performs repetitive work. The worker repeats the same work every time as in the assembly process, but works with ambiguity with respect to the work position, work time, and the like. This is because the operator's state (for example, position) is considered to change to another state with ambiguity as the operation proceeds because a plurality of operations are performed according to the procedure.

  In order to achieve the above object, the work support system of the present invention uses a Markov model generated from worker movement data to predict the worker's behavior, and the worker's behavior predicted by the prediction unit. And a supply means for supplying parts, tools, and other necessary items to the worker.

  As one specific configuration, the prediction means calculates an expected value of a period (hereinafter referred to as “state duration length”) that remains an arbitrary state variable from the Markov model, and The movement trajectory of the worker is predicted by determining the movement destination when the state duration time elapses.

  Preferably, the prediction means extracts a state variable having a low state transition probability from the Markov model, and obtains an expected value of the worker's state for each work from the extracted state variable distribution; From the expected value of the worker's state for each operation obtained by the first calculation processing unit, the second calculation processing unit that estimates the current work performed by the worker and the Markov model By calculating the expected value for a certain period (hereinafter referred to as “state duration length”) and determining the destination when the state duration length has elapsed, the predicted movement trajectory of the worker is determined. Based on the third calculation processing unit to be calculated, the current operation estimated by the second calculation processing unit and the predicted movement trajectory calculated by the third calculation processing unit, the arrival time and the next operation to move to the next work location Fourth calculation to find the predicted trajectory to a place It includes a processing section, a.

  Further, the supply trajectory calculated by the fourth calculation processing unit and for calculating the supply timing and the supply trajectory of the supply means based on the arrival time to move to the next work place and the predicted movement trajectory to the next work place A calculation means is provided.

  As another specific configuration, the predicting means extracts a state variable having a low state transition probability from the Markov model, and obtains an expected value of the worker's state for each work from the extracted state variable distribution. And a supply position determining means for determining a supply position from an expected value obtained by the first calculation processing unit and an expected value of a position where an operator for each operation exists.

  According to the present invention, since the predicting means predicts the worker's behavior using the Markov model generated from the movement data of the worker, the supply means is based on the predicted operator's behavior, such as parts, tools, and the like. Supply the necessary items. Therefore, even when the worker performs each predetermined task and moves to the next task earlier than a predetermined time, the time to move to the next task can be predicted before moving to the next task. Therefore, it is possible to supply necessary articles as the worker moves without causing the worker to lose time.

  Further, according to a specific configuration, a Markov model is generated from the movement data of the worker, and the movement trajectory is predicted every time the worker enters a new state (for example, a cell). By predicting the worker's destination and travel time at an early stage, predicting the start time and location of the next work and generating the supply operation trajectory, the parts and tools can be moved without delay from the worker's work progress. Supply. Thus, even when the worker finishes work earlier than the supply time determined based on the worker model created by the method disclosed in Patent Document 1, before moving to the next work place When the movement to the next work place is completed, the time and place where the next work is started are predicted, and the supply means such as the robot arm is driven based on this. Therefore, parts and tools can be supplied without delay from the progress of the work of the worker, and work efficiency is improved. Also, by predicting the movement trajectory of the worker, it is possible to generate a robot arm trajectory with reduced risk of collision with the worker.

It is a block block diagram of the work assistance system which concerns on embodiment of this invention. It is a figure for demonstrating discretization of a work space. It is a figure which shows typically the update of the Markov model in a Markov model production | generation means. It is explanatory drawing regarding a worker's movement track | orbit prediction.

Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings.
The work support system 1 according to the embodiment of the present invention is based on a prediction unit 10 that predicts a worker's behavior using a Markov model generated from the movement data of the worker, and a worker's behavior predicted by the prediction unit 10. Supply means 20 for supplying parts, tools and other necessary articles to the operator.

  In this work support system 1, first, the predicting means 10 models a worker's movement during a series of work using a Markov model, and based on the model, the worker position and the worker's prediction for each work. One or more of predicting the trajectory, the start time of the next operation, etc., and specifying the current position based on the model are performed. In this way, based on one or a plurality of results predicted and estimated by the prediction unit 10, the supply unit 20 is controlled to supply tools, parts, and other necessary articles to the operator.

  The work support system 1 further includes a measuring unit 21 that measures the position of the worker, and a worker position estimating unit that estimates the position of the worker using a particle filter or the like based on a result input from the measuring unit 21. 22, a value input from the prediction means 10, a supply position determination means 23 for determining a supply position based on an expected value of the presence rate of the worker for the work, and determination of supply timing by the supply means 20 Supply trajectory calculation means 24 for calculating the supply trajectory and outputting the supply trajectory to the supply means 20.

  Assuming that the worker's workspace is divided into a plurality of cells as state variables, the prediction means 10 generates a Markov model from the worker's movement data and updates it as necessary. The Markov model is used to predict the arrival time, movement trajectory, and the like for the worker to move to the next work location.

  Here, the Markov model is a probabilistic model described by a Markov chain, and the Markov chain is one of the probabilistic methods for discretely representing states and determining the future state only by the current state. is there.

An example of how a Markov model is generated will be described by taking as an example a case where the worker's workspace is divided into a plurality of cells as state variables.
FIG. 2 is a diagram for explaining the discretization of the work space. As shown in FIG. 2, state variables are defined by discretizing the work space. For example, if the work space is a 4 m × 8 m space, it is divided into 32 × 64 cells, each cell is given a cell number, and this is defined as a state variable. Then, it is described in a state transition probability matrix whether the worker exists in a certain cell at a certain time and whether the worker stays in the current cell or moves to another adjacent cell. Creating or updating this state transition probability matrix is called Markov model generation / update. The movement trajectory of the worker is predicted based on this state transition probability matrix.

  The operator's position is measured by the measuring means 21 using a Laser Range Finder (LRF) set at a height above the operator's knee. Then, the worker position estimating means 22 applies the clustering method and the particle filter to the measured point group to estimate the worker position. As this estimation method, those disclosed in Patent Document 1, those disclosed in Non-Patent Document 1, and other equivalents can be used.

  In the estimation method disclosed in Patent Document 1, the position of the worker is measured using a sensor in which one LRF is attached to the waist height and one to the foot height. In this method, the clustering method is first applied to the LRF measurement points of the lumbar region and the foot region, and a cluster of the lumbar region and the foot region is generated. The worker's representative point is obtained from the average value of the points in the cluster, and the worker's position is specified from the representative point of the worker.

  The estimation method disclosed in Non-Patent Document 1 is an improvement of the method of Patent Document 1 described above. First, an operator is distinguished from an obstacle and an operator using a background difference method from data points obtained by LRF. And the number of clusters generated by applying the clustering method to the extraction result is defined as the number of workers. Next, each worker is tracked by a particle filter to estimate the positions of the partner worker and other workers. Here, the particle filter is a method for estimating the current state by using a large number of particles and using prediction from a past state and current observation information.

  Specifically, the prediction unit 10 includes a Markov model generation unit 15, a first calculation processing unit 11 that calculates an expected value of the worker location for the work using the Markov model generated by the Markov model generation unit 15, and A second calculation processing unit 12 for estimating the current work of the worker, a third calculation processing unit 13 for calculating the predicted movement trajectory of the worker using the Markov model generated by the Markov model generation means 15, A fourth calculation processing unit 14 that calculates a predicted arrival time and a predicted movement trajectory to the work place.

  The first calculation processing unit 11 extracts a state variable having a low state transition probability from the Markov model generated by the Markov model generation means 15, and from the distribution of the extracted state variables, the expected value of the worker's state for each operation Ask for. The case where the state variable is a cell will be specifically described as an example. The first calculation processing unit 11 extracts a cell having a high probability of staying in the same cell in the Markov model, and calculates the distribution of the extracted cell as many as the number of operations. Approximation is performed with a Gaussian distribution, and the coordinates of this center are set as the expected value of the position where the worker exists for each work. Although the cell distribution is approximated by a Gaussian distribution, the cell distribution may be obtained by an approximate expression other than the Gaussian distribution.

  The second calculation processing unit 12 includes a Gaussian distribution and other approximate distributions regarding the worker position for each work obtained by the first calculation processing unit 11 and the current position of the worker input from the worker position estimation means 22 ( Xw, Yw) are compared to estimate the work that the worker is currently performing.

  The third calculation processing unit 13 calculates an expected value of a period (referred to as “state duration length”) that remains an arbitrary state variable from the Markov model generated by the Markov model generation unit 15; and The movement trajectory of the worker is predicted by determining the movement destination when the state duration time elapses. The case where the state variable is a cell will be specifically described as an example. In the Markov model generated by the Markov model generation means 15, the third calculation processing unit 13 is a period during which an operator stays in each cell (“state duration time length”). ) And the destination cell when the state duration time has elapsed is determined to be the cell with the highest transition probability at that moment. By repeating this, the predicted movement trajectory of the worker is calculated.

  Based on the current work estimated by the second calculation processing unit 12 and the predicted movement trajectory calculated by the third calculation processing unit 13, the fourth calculation processing unit 14 determines the arrival time to move to the next work place and the next Find the predicted trajectory to the work place.

  The supply position determination means 23 determines the supply position (Xn, Yn) based on the value calculated by the first calculation processing unit 11 and based on the expected value of the presence rate of the worker for the work.

  The supply trajectory calculation means 24 determines the supply timing by the supply means 20 and calculates the supply trajectory.

  The configuration of the work support system 1 according to the embodiment of the present invention is as described above. With this configuration, the arrival time for the worker to move to the next work location using the Markov model generated from the worker's movement data. And whether the movement trajectory is predicted will be described.

  Now, it is assumed that the work space is divided into a plurality of cells, cell numbers are assigned, and the cell numbers are defined as state variables. Further, as shown in FIG. 2, when the work space of the vehicle body coordinate system Σc composed of coordinates Xc and coordinates Yc is divided into cells and the operator repeats a series of operations for each vehicle, the measuring means 21 Thus, for example, since the position of the worker is measured by LRF, the worker's movement trajectory data for one vehicle can be obtained by the worker position estimating means 22. The Markov model generation means 15 receives the movement trajectory data of the worker for one vehicle. The Markov model generation means 15 receives the input and is described by the state transition probability matrix, thereby generating and updating the Markov model.

  In order to predict the movement route of the worker, the Markov model generation means 15 generates a Markov model from the behavior data of the worker. Now, regarding the movement of the worker, as shown in FIG. 2, the work space is divided into a plurality of cells, and a cell number is assigned to each cell to define a state variable. An operator's work space, for example, a work space in the vehicle body coordinate system is discretized in a mesh shape, and a state variable is defined in each cell. For each cell, the number of times the worker is present is counted every discrete time interval Δt (seconds), and a Markov model is statistically generated.

In the present embodiment, the state transition probability a _ij is unique to each cell and is a value that does not change with time. In other words, when an operator enters a cell, the time from the start of the operation of this process does not matter at all, it stays in that cell for the time of the expected value of a certain state duration, and the cell with the highest state transition probability To model the movement of the worker.

ここで、ｉは時間ｔ−Δｔのときの状態を示し、ｊは時間ｔのときの状態を示す。Ｎ_ijは作業者が存在したセルがｉからｊへ遷移した回数を示し、Ｎ_iは作業者がセルｉ に存在した回数を離散時間間隔ごとにカウントしたものである。この状態遷移確率ａ_ijを全iとjの組み合わせについてすべて求め、行列[ａ_ij]としてあらわしたものが本実施形態におけるマルコフモデルである。なお、ｉ，ｊは状態変数の数がｍ個であれば、ｉ，ｊは１≦ｉ，ｊ≦ｍの関係式を満たす自然数である。 More specifically, the state transition probability a _ij is calculated using Equation (1).

Here, i indicates a state at time t-Δt, and j indicates a state at time t. N _ij indicates the number of times that the cell in which the worker was present has transitioned from i to j, and N _i is the number of times that the worker has been present in cell i 2 at each discrete time interval. The Markov model in this embodiment is obtained by _obtaining all the state transition probabilities a _ij for all combinations of i and j and representing them as a matrix [a _ij ]. Note that i and j are natural numbers satisfying the relational expression of 1 ≦ i and j ≦ m if the number of state variables is m.

  Here, the discretization of the state space needs to be set in detail in order to fully model the characteristics of the worker's movement. It may be determined in consideration of. For example, if the state space of 4 m × 8 m is divided into 32 × 64 cells, the work space can be sufficiently covered.

  In this way, a Markov model is generated from the movement data of the worker. As shown in FIG. 3, the Markov model generation means 15 may update the Markov model each time the movement trajectory data of the worker for one vehicle is input, but it does not necessarily have to be updated. . The Markov model is updated in order to cope with changes in the Markov model because the behavior and movement trajectory of the worker can change due to the operator's familiarity with work and fatigue. Of course, a Markov model may be generated for each worker and used as a worker-specific one, or the model does not need to be changed in the case of work with some reproducibility or sensitive to the movement of the worker. In some cases, it is not necessary to update. In order to reflect the latest worker's habit in the Markov model, for example, a forgetting Markov model learning algorithm that forgets several percent of the past Markov model at every update timing is used. As this algorithm, for example, a well-known SDLE (Sequentially Discounting Laplace Estimation) algorithm can be used.

  Next, the work behavior is modeled from the generated Markov model and the work behavior is estimated. The first calculation processing unit 11 calculates an expected value for the position of the worker with respect to the work. For this expected value, first, only the number of operations is known, cells having a high probability of staying in the same cell in the Markov model are extracted, and the distribution of the extracted cells is approximated by a Gaussian distribution or other distributions corresponding to the number of operations. Specifically, the parameters of the Gaussian distribution in each work are calculated using the EM algorithm, and the coordinates of the center of the Gaussian distribution are set as the expected value of the position where the worker exists for each work. Here, the EM algorithm is a technique for estimating a parameter of a probability model based on a maximum likelihood method in statistics, and an expectation (expectation, E) step and a maximization (maximization, M) step are alternately performed. The calculation is advanced by repeating.

  As for the current position of the worker, the second calculation processing unit 12 uses the position of the worker obtained from the measuring means 21 and the worker position estimating means 22 as described above, and the distribution already obtained in each work. By comparing the Gaussian distributions of the worker positions, it is possible to calculate which work is likely to be performed if the worker is at that position. Therefore, it can be estimated that the work is being performed.

  The third calculation processing unit 13 predicts the movement trajectory of the worker. The movement trajectory of the worker is performed using a Markov model generated from the movement data of the worker. This prediction consists of the following two steps. One is a step for calculating an expected value of the state duration time, and one is a step for determining a transition destination according to the state transition probability.

The step of calculating the expected value of the state duration time will be described.
The state duration length means a period during which the state i remains in the Markov process. When the state duration is designated, the probability p _i (d) that the state i continues for the designated time d (seconds) can be obtained from the equation (2).

Here, π _i represents an initial probability. The expected value d _exp of the state duration time length represents the average of the period during which the state i continues, and can be obtained from Equation (3).

Since the expected value of the state duration time is an expected value of the length of the period during which the worker stays in this cell, it is considered that the cell moves to another cell after this d _exp . Therefore, in the moving trajectory prediction method, this expected value is used to predict the moving time.

The step of determining the transition destination according to the state transition probability will be described.
From the viewpoint of the state duration length, it is considered that the state transitions to a state j other than the state i immediately after the state i continues by the expected value d _exp of the state duration time length. That is, this state j is any state that satisfies j ≠ i. The state transition destination is determined according to the state transition probability to a state having the maximum state transition probability.

  Therefore, the movement trajectory of the worker can be obtained by sequentially repeating the above step (Step 1) for calculating the expected value of the state duration time and the step (Step 2) for determining the transition destination according to the state transition probability. .

  FIG. 4 is a diagram illustrating an example for explaining the movement trajectory prediction of the worker. (A) shows a state in which Step 1 and Step 2 are repeated to obtain a predicted observation sequence, and (b) shows a prediction on a discretized state space based on the predicted observation sequence. The movement location and movement route of the worker to be performed are shown.

  As a prediction procedure, first, assuming that the current worker exists in the state 3 on the discretized state space, the time during which the worker remains in the state 3 is calculated by Step 1. Next, the transition destination according to the state transition probability is determined by Step2. Assume that the transition destination is now in state 9. Then, Step 1 is performed again, the time during which the worker continues to exist in the state 9 is obtained, and Step 2 is applied again to determine the next transition destination.

  In this way, when the two steps are repeated and the total expected value of the state duration time reaches the target prediction time Tp, the final transition state is the movement of the worker after the predicted Tp time. Is a place. In the figure, the predicted movement location of the worker is the state 20, the predicted movement paths are 3, 9, 14, and 20, and the movement trajectory including the time spent in each cell is 3, 3 , 3, 9, 9, 14, 14, 14, 14, 20, 20, 20, 20.

  Next, the fourth calculation processing unit 14 predicts arrival at the next work location from the worker's current position obtained by the second calculation processing unit 12 and the predicted movement trajectory obtained by the third calculation processing unit 13. Calculate time and predicted trajectory. Specifically, the next work is identified from the current work, the time at which the movement to the next work location ends is obtained, and the arrival time is obtained from the difference between this and the current time. The predicted movement trajectory to the next work location uses the trajectory from the current location to the next work location out of all the predicted trajectories. Note that the predicted movement trajectory obtained by the third calculation processing unit 13 is the operator's predicted movement trajectory through a series of operations, and the predicted movement trajectory obtained by the fourth calculation processing unit 14 is the work to the next work place. This is the predicted movement trajectory of the person. That is, the fourth calculation processing unit 14 extracts a predicted trajectory toward the next work from a series of predicted trajectories of the workers.

  As described above, the prediction means 10 obtains the expected value of the worker presence rate with respect to the worker, the predicted arrival time to the next work place, and the predicted movement trajectory.

Therefore, the supply position determination unit 23 determines the position having the highest “expected value of the presence rate of the worker for the work” input from the first calculation processing unit 11 as the supply position (Xn, Yn). Thereafter, an offset (Δx _n , Δy _n ) such as a position where the operator can easily reach his / her hand is added to obtain the final supply position. This offset value is also used for fine adjustment in consideration of ease of work of the operator by teaching or the like.

  The supply trajectory calculation means 24 determines the timing of supplying parts, tools and other necessary articles to the worker by the supply means 20 and calculates the supply trajectories x (t) and y (t) of the supply means 20. To do. The timing is determined from the predicted arrival time at the next work place input from the predicting means 10 and the timing for supplying parts, tools and other articles by the supplying means 20. In the simplest case, the estimated arrival time of the operator can be used as it is for the supply timing, and the supply can be performed early by intentionally providing a time for advance. The trajectory calculation of the supply means 20 includes the predicted movement trajectory of the worker to the next work place input from the prediction means 10 and the supply position offset information (ΔXn, ΔYn to the supply position input from the supply position determination means 23. The supply trajectory of the supply means 11 is calculated from the final supply position to which () is added and the determined supply timing. Here, the supply trajectory may simply be a trajectory obtained by interpolating a straight line from the replenishment position to the final supply position of tools, parts, and other necessary items, with a time axis, for example, using a polynomial, or an operator's prediction. It may be a trajectory created in consideration of the moving trajectory so that collision does not easily occur.

  The supply means 20 includes a robot arm, a mobile robot and other robots, a belt conveyor and other transport mechanisms, and according to the supply trajectories x (t) and y (t) input from the supply trajectory calculation means 24, the tips of these mechanisms. For example, the hand of the robot arm is moved. In addition to the transport mechanism, the supply means 20 stores a list of necessary articles such as parts and tools according to the work, and the list from the list by the input from the supply trajectory calculation means 24 is stored. The necessary articles are selected and supplied to the operator's hand by the transport mechanism.

  According to the embodiment of the present invention, when the work progress is estimated and the work is progressing quickly by the worker, the parts, tools, etc. are given to the worker without delay in accordance with the work progress at that time. You can supply what you need.

  In the above-described embodiment of the present invention, the case where the worker's work space is divided into a plurality of cells as state variables has been described as an example. It is also possible to acquire the work situation from the acquired motion pattern and use the estimated value as a state variable. Even if the state variables are set in this way, it is possible to predict the arrival time and movement trajectory of the worker moving to the next work location using the Markov model generated from the movement pattern of the worker, and the prediction result Accordingly, the movement path of the worker may be obtained according to the control, and the supply means may be controlled.

1: work support system 10: prediction means 11: first calculation processing section 12: second calculation processing section 13: third calculation processing section 14: fourth calculation processing section 15: Markov model generation means 20: supply means 21: measurement Means 22: Worker position estimation means 23: Supply position determination means 24: Supply trajectory calculation means

Claims (5)

A prediction means for predicting the behavior of the worker using a Markov model generated from the movement data of the worker;
Based on the worker's behavior predicted by the prediction means, supply means for supplying parts, tools and other necessary items to the worker;
A work support system comprising:   The prediction means calculates an expected value of a period (hereinafter referred to as “state duration length”) that remains an arbitrary state variable from the Markov model, and the state duration length has elapsed. The work support system according to claim 1, wherein the movement trajectory of the worker is predicted by determining a movement destination at the time. The prediction means is
A first calculation processing unit that extracts a state variable having a low state transition probability from the Markov model, and obtains an expected value of the worker's state for each operation from the extracted state variable distribution;
A second calculation processing unit that estimates an operation currently performed by the worker from an expected value of the worker's state for each operation obtained by the first calculation processing unit;
From the Markov model, an expected value of a period that remains an arbitrary state variable (hereinafter referred to as “state duration length”) is calculated, and a destination is determined when the state duration length has elapsed. A third calculation processing unit for calculating the predicted movement trajectory of the worker,
Based on the current work estimated by the second calculation processor and the predicted movement trajectory calculated by the third calculation processor, the arrival time to move to the next work place and the predicted movement trajectory to the next work place are calculated. A fourth calculation processing unit to be obtained;
The work support system according to claim 1, comprising:   A supply trajectory calculated by the fourth calculation processing unit for calculating the supply timing and the supply trajectory of the supply means based on the arrival time to move to the next work place and the predicted movement trajectory to the next work place. The work support system according to claim 3, further comprising a calculation unit. The prediction means includes a first calculation processing unit that extracts a state variable having a low state transition probability from the Markov model, and obtains an expected value of the worker's state for each operation from the extracted distribution of the state variable,
The work according to claim 1, further comprising a supply position determination unit that determines a supply position from an expected value obtained by the first calculation processing unit and an expected value of a position where an operator for each work exists. Support system.

Images