JP3872387B2 - Control device and control method of robot coexisting with human - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To realize a control technique without unnecessary restriction of movement of a robot, while preventing incidents such that a human error occurs and a human enters a non-entry region during a cooperation work of human beings and a cooperating robot. <P>SOLUTION: This device comprises: a means 22 for storing "a space requiring control by priority of safety" and "a space allowing control by priority of productivity" in a space having dimensions such as a distance to a non-entry region and speed; a storage means 17 of an avoiding trajectory group; a storage means 18 of an entry trajectory group; and a means 12 for judging whether an actual trajectory of the robot belongs to the avoiding trajectory group or the entry trajectory group. The device also has a means 26 for allowing a robot operation to continue unless it is judged to belong to the entry trajectory group when a movement condition of the robot is in "a soft safety control allowing space", and for restraining continuous robot operation unless it is judged to belong to the avoiding trajectory group in "a careful safety control requiring space". <P>COPYRIGHT: (C)2004,JPO

Description

[0001]
BACKGROUND OF THE INVENTION
The present invention relates to a control technology for a robot coexisting with a human.
[0002]
[Prior art]
For example, in order to efficiently perform the work of transporting and positioning a heavy object, a robot that detects a human operation and exerts a force to reduce the force required for the operation has been developed. This type of robot is a robot that changes its trajectory by human operation, and can be said to be a robot that works together with a human while coexisting with the human. Robots that change their trajectory by human operations can be used in combination with the advanced judgment and control power of humans and the power that robots do not know about fatigue, and are beginning to be put into practical use for various tasks. . The inventors of the present invention have developed a robot that works in cooperation with humans to carry the vehicle instrument panel and position it on the vehicle body, and has achieved great results. This robot is called a skill assist robot because it assists heavy load work while utilizing skills derived from human advanced judgment and control power. Robots that collaborate with humans of this kind are expected to be used in many fields.
[0003]
[Problems to be solved by the invention]
As the range of use of skill-assisted robots that collaborate with humans expands, it becomes necessary to regulate the movement of the robots so that they do not fall into situations that could endanger humans. For example, when a heavy object is transported through the opening, a human hand that moves together with the robot must not be sandwiched between the robot and the opening.
The trajectory of a skill assist robot that detects a human operation and reduces the force required for the operation is basically determined by the human operation. This is quite different from the current industrial robot that moves along the previously taught trajectory. Even when a heavy object is transported through the opening illustrated above, the trajectory intended by a person differs from person to person (thus, a person can have skills), and even the same person repeats work. Each time you do it, the trajectory will be different. No suitable technique has been developed for determining whether the trajectory is a trajectory for which danger is a concern or a trajectory that can be safely moved.
[0004]
In the present invention, a future working environment (such as a frame in which a robot or a part of a human body forms an opening of a vehicle, etc.) with respect to a trajectory that can be changed in various ways depending on a human being operated or sometimes changes. We have created a robot control device and control method that can determine whether the trajectory is a concern or a safe trajectory, and can regulate the robot trajectory based on the determination result. . In particular, instead of unnecessarily regulating with emphasis on safety unnecessarily, in a range where danger is not a concern, humans can freely control the trajectory and work comfortably and efficiently while utilizing human skills. To be able to. At the same time, if the trajectory is truly concerned about danger, the subsequent movement is restricted to prevent humans from falling into a dangerous situation due to human error.
[0005]
Means for Solving the Problems and the Invention The invention of claim 1 relates to a robot control device that changes a trajectory by a human operation. This controller isWhen a part of the robot is gripped by a human and the robot is moved, the point of interest set on the robot, the person gripping the robot, or the workpiece gripped by the robot does not enter the intrusion prohibited area. To control the robot.
  This control device must be within the movable range of the point of interest and prohibit the point of interest from entering.In a space with dimensions of distance, speed and direction to the no entry area, Because the combination of “distance, speed, and direction” is a combination that is highly likely to enter the entry-prohibited area, because “space requiring safety control” is not found, “the avoidance trajectory to be described later has not been discovered.Requires control that prioritizes safetyFirstSpace ","Because the avoidance trajectory has been discoveredAllow control to prioritize productivitySecondspace"Divided intoMemorizing means,The point of interest rushesProhibited areaTo rush intoAvoidAvoiding points of interestOrbital groupAn avoidance trajectory group model describing a learned avoidance trajectory groupMeans for storingThe point of interest rushesEnter the prohibited areaEntering the point of interestOrbital groupAn intrusion trajectory group model describing a learned ingress trajectory groupMeans for storingOf the point of interest measured while a human is manipulating and adjusting the trajectory of the robot.Position, speed and directionDescribed inIn factofOrbitCalculate the probability belonging to each of the avoidance trajectory group model and the inrush trajectory group model, and the actual trajectory measured isBelongs to the avoidance trajectory groupofOr,Belonging to the rush track groupofOrOr is it impossible to judgeMeans for determining. further,The position, speed, and direction of the point of interest measured while a human is operating to adjust the robot trajectory are the second space.While in, The actual trajectory measuredWhile it is allowed to continue the robot operation by humans unless it is determined to belong to the rush trajectory group,The position, velocity, and direction of the measured point of interest are the first space.While in, The actual trajectory measuredAnd means for restricting the continuation of the robot operation by a human unless it is determined to belong to the avoidance trajectory group.
[0006]
  Through the inventor's research,The point of interest isAssume a space with dimensions of distance, speed and direction to the area where entry is prohibitedThenIn that spaceIt is possible to set “space that requires safety control”.AndCreate a space that requires safety control"Space that allows control giving priority to productivity" and "Space that requires control giving priority to safety"Divided intoIn other words, the trajectory changes depending on the human being operated or depending on the time.Turn intoHowever, it was confirmed that necessary safety control could be taken uniformly.
  By learning in advance,The point of interest isSuddenProhibited areaCan avoid rushing into the areaAvoiding points of interestTracks can be stored. By learning in advance,The point of interest isSuddenProhibited areaIt is possible to store a rush trajectory group of a point of interest that rushes into the area. However, it is impossible to learn about all of the trajectory groups that can be changed variously by the man who operates or sometimes, and not to learn whether it is an avoidance trajectory or a rush trajectory.
  Various mathematical methods have been developed to describe the trajectory, and even when trajectories that do not exactly match the learning trajectory avoidance or intrusion trajectories were observed, they were observed using mathematical modeling techniques.The actualWhether the trajectory is an avoidance trajectory groupOrIt can be determined whether it is a rush track group. However, the reliability of the determination is high only for the trajectory similar to the learning trajectory avoidance trajectory group or the rush trajectory group, and a reliable determination cannot be performed for a trajectory that is not similar. It is an avoidance trajectory for all possible trajectories when the trajectory changes variously depending on the man who operates or sometimesIsOr rush orbitIsYou cannot learn as much as you can.
  In the present invention, the restriction that it cannot be fully learned is that “safe control is required.Set the space to "Complementing by distinguishing between “a space that allows control that prioritizes productivity” and “a space that requires control that prioritizes safety”, humans can freely control the trajectory and work comfortably and efficiently When it was possible to do so, it allowed free operation, and when there was a real danger, it succeeded in regulating the trajectory so that it did not fall into a dangerous situation.
[0007]
It is preferable that the learned rush track group and avoidance track group are described and stored in a hidden Markov model.
In this case, it is easy to determine whether the trajectory similar to the avoidance trajectory group or the rush trajectory group in the learning corner is the avoidance trajectory or the rush trajectory.
The hidden Markov model is a method for describing the trajectory group. For example, a pattern identifying method using a DP matching method or the like may be employed.
[0008]
  In the present invention, a robot control method that changes the trajectory by human operation has also been created. in this way,When a part of the robot is gripped by a human and the robot is moved, the point of interest set on the robot, the person gripping the robot, or the workpiece gripped by the robot does not enter the intrusion prohibited area. To control the robot.
This control method needs to prohibit the point of interest from entering the point of interest within the movable range.In a space with dimensions of distance, speed and direction to the no entry area, Because the combination of “distance, speed, and direction” is a combination that is highly likely to enter the entry-prohibited area, because “space requiring safety control” is not found, “the avoidance trajectory to be described later has not been discovered.Requires control that prioritizes safetyFirstSpace ", "Because an avoidance trajectory has been discoveredAllow control to prioritize productivitySecondspace"Divided intoA step of measuring in advance;The point of interest rushesProhibited areaTo rush intoAvoidAvoiding points of interestLearn the orbital group in advanceModelingThe placing step,The point of interest rushesEnter the prohibited areaEntering the point of interestLearn the orbital group in advanceModelingAnd the step of placing. And while actually manipulating and adjusting the robot trajectoryOf the measured point of interestPosition, speed and directionDescribed inIn factofOrbitCalculate the probability belonging to each of the avoidance trajectory group model and the inrush trajectory group model, and the actual trajectory measured isBelongs to the avoidance trajectory groupofOr,Belonging to the rush track groupofOrOr is it impossible to judgeA step of determiningThe position, speed, and direction of the point of interest measured while a human is operating to adjust the robot trajectory are the second space.While in, The actual trajectory measuredWhile it is allowed to continue the robot operation by humans unless it is determined to belong to the rush trajectory group,The position, velocity, and direction of the measured point of interest are the first space.While in, The actual trajectory measuredUnless it is determined to belong to the avoidance trajectory group, a process of restricting the continuation of the robot operation by a human is performed.
[0009]
According to this method, in the case of “requiring careful safety control”, unless it is determined to belong to the avoidance trajectory group, that is, unless it is confirmed to be safe, the continuation of the robot operation is restricted, which is truly dangerous. If you are concerned, you can regulate the trajectory so that you do not fall into a dangerous situation. On the other hand, in the case of “flexible safety control is allowed”, the robot operation is allowed to continue unless it is determined to belong to the rush track group, that is, unless it is confirmed that there is a risk of danger. Even if robot operation is allowed, if there is a real danger, it is necessary to move to a space that requires careful safety control before falling into a dangerous situation of concern. The necessary safety can be ensured.
This method allows humans to freely control the trajectory in a wide range (under a variety of conditions) and to proceed comfortably and efficiently, and when danger is truly a concern. The orbit can be regulated so that it does not fall into a dangerous situation.
[0010]
  The position, speed, and direction of the measured point of interest“We need control that prioritizes safetyFirstWhen an avoidance trajectory is found when in space,Avoidance trajectoryIt's been foundSometimes the position, speed and direction of the point of interest measuredState space, “control that gives priority to safetyFirstAllow control that gives priority to productivity from spaceSecondIt is preferable to further include a step of changing to “space”.
  When this process is added, the boundary between "space requiring control that prioritizes safety" and "space allowing control prioritizing productivity" is learned and corrected to a boundary suitable for the actual situation. Can do.
[0011]
Preferably, the method further includes a step of adding a newly discovered avoidance trajectory to a previously learned avoidance trajectory group.
If this step is added, the chance of erroneously determining that it is a rush track group is reduced although it is actually avoidable and not a rush track group. As a result, the chance of being erroneously determined to be a rush track group when being in “a space that allows control giving priority to productivity” and the operation being restricted is reduced.
[0012]
DETAILED DESCRIPTION OF THE INVENTION
First, the main forms of the technology shown in the examples are listed.
(Mode 1) A handle that is grasped and operated by a human being is prepared on the arm of the robot, and the motion state of the robot or a part of the human body is specified by the position, moving speed, and moving direction of the handle.
(Mode 2) When safety control is required, the movement of the robot is forcibly stopped.
(Mode 3) When safety control is required, the robot trajectory is forcibly switched to an avoidance trajectory.
[0013]
【Example】
FIG. 1 shows a skill assist robot to which the control device and the control method of this embodiment are applied. The arm 4 of the skill assist robot 2 is provided with a handle 6 that is grasped and operated by a human. The robot 2 can grip and carry a heavy object such as a vehicle seat. The person 8 grasps the handle 6 and moves it at a speed at which he / she wants to move it. The robot 2 detects a human operation and exerts a power that reduces the force required for the operation. The human 8 can collaborate with the robot 2 to move a heavy object at a desired position by moving it in a desired direction at a desired speed with a light operating force.
[0014]
When the human 8 collaborates with the robot 2, the superior judgment power and control power of the human 8 can be used in combination with the power that the robot 2 exerts without breaks. Since the work of the human 8 can be utilized while the power of the robot 2 is utilized, the work can be advanced. Unlike a normal industrial robot, the skill assist robot 2 can be said to be a robot whose trajectory is changed by a human operation.
[0015]
In this embodiment, the person 8 and the skill assist robot 2 cooperate to convey a heavy object such as a vehicle seat from the outside of the vehicle body into the vehicle body and position it at a predetermined position in the vehicle body. In this work, the handle 6 gripped by the human 8 needs to pass through the opening 12 of the vehicle body 10, and if the human 8 inadvertently makes a human error, the hand of the human 8 gripping the handle 6 There is a possibility of being pinched between 12 edges. Safety control is required to prevent a dangerous situation even if a human error occurs.
[0016]
In this embodiment, various learning processes are performed in advance to contribute to safety control.
(Avoidance trajectory group learning process)
The human hand holding the handle 6 by actually operating the skill assist robot 2 is moved along a trajectory passing through the edge of the vehicle body opening 12. At this time, k1 avoidance trajectory groups are taught.
FIG. 2 illustrates a trajectory in a moving surface of a human hand holding the handle 6 (generally not a plane, but a planar motion example will be described here). The trajectory is the distance r from the reference point position 16 of the entry prohibition area 14 (in this case, the vehicle body 10) to the human hand holding the handle 6, the moving speed v of the human hand holding the handle 6, and the handle 6. It can be identified by observing the moving direction φ of the human hand along a time series. FIG. 2 illustrates that the human hand holding the handle 6 is at a point Pt on the orbit at time t, the distance at that time is rt, the speed is vt, and the direction is φt. ing. The moving direction of the human hand holding the handle 6 is zero in the direction in which the vehicle body extends, and the angle formed by the vehicle body is adopted.
The observed elements are called observation symbols. The observation symbol is (rt, vt, φt). Note that the position of the human hand holding the handle 6 at the start of work is indicated by reference numeral 18. In FIG. 2, one avoidance track S and one entry track D are illustrated.
[0017]
(Modeling of avoidance trajectory group)
The hidden avoidance trajectory group can be modeled by the hidden Markov model. The hidden Markov model is modeled by equation (1).
λ1 = {A1, B1, π1} (1)
π1 is a probability distribution initially in the state s1, A1 is a transition probability between the two states s1 and s2 under the model λ1, B1 is an observation symbol Ot (t = 1 to t1: t1 is one time in the state s2 This is the probability of outputting the last movement time in the work. The state s1 is a state “safety control giving priority to safety” described later, and the state s2 is a state “safety control giving priority to productivity is allowed”.
In Equation (1), a suitable initial value is selected, the observed symbol Ot of the observed avoidance trajectory group is substituted, and the learned avoidance trajectory group S is modeled by using the Baum-Welch re-estimation method. The following formula (1) can be obtained.
Details of this modeling are described in the following two papers published by the present inventors. “Proposal of accident prevention method based on evidence theory based on Hidden Markov Model”, Y. Yamada, Yuki Sonohara, Tetsuya Morizono, Yoji Umeya, 2nd System Integration Division Academic Lecture (SI2001), pp517-518, (2001.12 )
“Accident prevention method due to behavioral slip based on evidence theory based on hidden Markov model”, Y. Yamada, Yuki Sonohara, Tetsuya Morizono, Yoji Umeya, Proceedings of the 7th Robotics Symposia Lecture, pp167-172, (2002.3)
[0018]
(Learning process of rush trajectory group)
The human hand holding the handle 6 by actually operating the skill assist robot 2 is moved along a trajectory that collides with the edge of the vehicle body opening 12. This track is referred to as a rush track D that enters the rush prohibition region 14. At this time, a plurality (k1) of rush trajectory groups are taught.
FIG. 2 illustrates one entry track D. The rush trajectory D also includes the distance r from the reference point 16 of the rush prohibition area 14 to the human hand gripping the handle 6, the moving speed v of the human hand gripping the handle 6, and the movement of the human hand gripping the handle 6. It can be specified by observing the direction φ along the time series.
[0019]
(Modeling of rush track)
A rushing trajectory group is modeled by Equation (2) using a hidden Markov model.
λ2 = {A2, B2, π2} (2)
π2 is a probability distribution initially in the state s1, A2 is a transition probability between the two states s1 and s2 under the model λ2, and B2 is the observed symbol Ot (t = 1 to t1) in the two states s1 and s2. The probability of output.
In Equation (2), select an appropriate initial value, substitute the observed symbol Ot of the observed rush trajectory group, and model the learned rush trajectory group D by using the Baum-Welch re-estimation method. Equation (2) can be obtained.
[0020]
(Selection of space that requires control that prioritizes safety and space that allows control that prioritizes productivity)
Assume a space having a dimension of a distance r, a velocity v, and a direction φ to the reference point 16 of the entry prohibition region 14. The space is illustrated in FIG. 3, the distance r from the reference point 16 of the entry prohibition region 14 to the human hand holding the handle 6 on the x axis, and the moving speed of the human hand holding the handle 6 on the y axis. The moving direction φ of a human hand holding the handle 6 is taken on the z axis.
In the present embodiment, safety control is not required if the distance r from the reference point position 16 of the entry prohibition area 14 to the human hand holding the handle 6 is 50 cm or more, and safety control is required if it is less than 50 cm. It is going to be. Further, since the conveyance speed during actual work does not exceed 1 m / sec, a space of 1 m / sec or more is not taken into consideration. Since the safety control is not required when the traveling direction is away from the vehicle body, the safety control is required only when φ is in the range of 0 to 180 °.
In this embodiment, the space where safety control is required is divided into two types. One is a “space SP that requires control giving priority to safety”. This space SP is an area where the margin is low even in a space where safety control is required, and it is assumed that it is necessary to perform safety control in a careful manner with an emphasis on safety. In this embodiment, if the moving speed is 0.6 m / second or more, it is assumed that the space SP is uniformly “a space SP that requires control giving priority to safety”. In addition, when the distance to the handle 6 is less than 25 cm, the space SP is always “a space SP that requires control giving priority to safety” regardless of the speed. In the present embodiment, there is no boundary that may or may not be a “space SP that requires control giving priority to safety” depending on the moving direction of the human hand holding the handle 6. If necessary, a boundary of “a space SP that requires control giving priority to safety” can be provided depending on the moving direction φ of the human hand holding the handle 6.
In the present embodiment, among the spaces where safety control is required, spaces other than “space SP that requires control giving priority to safety” are defined as “space FW that allows control giving priority to productivity”. “Space FW that allows control to give priority to productivity” requires control for safety, but has a relatively high margin and activates control for safety unless there is a concern about imminent danger. This space is assumed to be unnecessary.
By the above, the preparatory process required for safety control is completed. The following steps are performed in real time during actual work.
[0021]
(Orbital observation)
A human error occurs during actual work, and the robot 2 is moved along a trajectory in which a human hand holding the handle 6 is sandwiched between the robot 2 and the edge of the opening 12 due to an erroneous operation by the human. It can be moved. Recognizing that such a trajectory is present, the robot 2 is controlled so as not to enter the entry prohibition area 14. In order to make it possible, during actual work, the distance r from the reference point 16 of the entry prohibition area 14 to the human hand holding the handle 6, the moving speed v of the human hand holding the handle 6, the handle 6 Continue to observe the moving direction φ of the human hand holding the That is, the observation symbols (r, v, φ) are continuously observed along the time series.
[0022]
(Judgment of observed orbit)
Using the equation (1) modeling the avoidance trajectory group S modeled in advance and the time series of the observed observation symbols, the probability that the observed trajectory is the avoidance trajectory can be determined.
The probability p (Ot / λ1) is expressed by the following equation (3)
p (Ot / λ1) = p (Ot, s1 / λ1) + p (Ot, s2 / λ1) (3)
Can be calculated with Here, the value on the right side can be calculated from the forward variable. Similarly, using the equation (2) that models the intrusion trajectory group D modeled in advance and the time series of the observed observation symbols, the probability that the observed trajectory is the intrusion trajectory can be determined. it can.
The probability p (Ot / λ2) is expressed by the following equation (4)
p (Ot / λ2) = p (Ot, s1 / λ2) + p (Ot, s2 / λ2) (4)
Can be calculated with Again, the value on the right side can be calculated from the forward variable. The details of the above probability calculation are described in detail in the above-mentioned paper published by the present inventor.
If the observed trajectory corresponds to the modeled avoidance trajectory group S or the modeled intrusion trajectory group D, the probability is calculated by the above formulas (3) and (4). It can be determined whether the observed trajectory is the avoidance trajectory S or the rush trajectory D.
[0023]
(Impossible to judge)
However, when the observed trajectory is neither the modeled avoidance trajectory group S or the modeled intrusion trajectory group D, that is, if it is significantly different from the taught trajectory groups S and D, Both the probability calculated by the equation (3) and the probability calculated by the equation (4) become zero, and it is determined whether the observed trajectory is the avoidance trajectory S or the rush trajectory S. I can't.
[0024]
In the case of conventional techniques, if the determination is impossible, the safety control must be activated, and the safety control is frequently activated. Work is frequently stopped, work efficiency is reduced, and work comfort is impaired. However, if safety control is not activated when determination is impossible, safety is not ensured when a human error that enters the entry prohibition area occurs.
[0025]
(Utilization of evidence theory)
In this embodiment, when it is impossible to determine whether the observed trajectory is an avoidance trajectory or a rush trajectory, it is possible to prevent safety control from being activated unnecessarily, and at the same time, safety control is actually necessary. Evidence theory is applied to ensure that safety control is triggered without delay.
Here, different evidence standards are adopted for “a space requiring control giving priority to safety” and “a space allowing control giving priority to productivity”. That is, as shown in FIG. 4, when it is impossible to make a determination while in a “space that requires control giving priority to safety”, it is said that “it is necessary to activate safety control in a rush track” while “production” When it is impossible to make a decision while in the space that allows control that prioritizes sex, it is not necessary to activate safety control because it is not proved that the vehicle is in a rushing trajectory and that there is no concern about danger. "
[0026]
(Experimental result)
The experimental results are shown in the lower column of FIG. In the experiment, five types of avoidance track groups and five types of rush track groups were learned in advance. “Space SP that requires control giving priority to safety” and “Space FW that allows control giving priority to productivity” were set as described above. For safety control, forced stop control was adopted.
The actual operation was performed many times with the above preparation. It was assumed that the rush trajectory was selected due to human error in 20 of them.
When logic that does not activate evidence control and does not trigger safety control when it cannot be determined (shown in the column of no evidence theory in FIG. 4), 4 out of 20 human errors have occurred. I couldn't prevent the accident.
Evidence theory is adopted, safety control is not activated when it cannot be determined while it is in “space FW that allows control giving priority to productivity”, and it is in “space SP that requires control giving priority to safety”. When the logic that activates the safety control is adopted when it cannot be judged in the meantime, it was possible to prevent an accident by forcibly stopping all 20 human error occurrences. The number of accidents was reduced from 4 to 0.
[0027]
(Learning the boundary between the space SP that requires control giving priority to safety and the space FW that allows control giving priority to productivity)
As a result of repeated experiments, it is understood that the space range SP that is restricted when the “space SP that requires safety-priority control” set at the start of the experiment is too wide to determine whether it is an avoidance trajectory or a rush trajectory. It was.
Therefore, we experimented with several trajectories that can be avoided by taking rush trajectories artificially and taking avoidance actions along the way. As a result, after the movement state of the robot or a part of the human body becomes a state corresponding to a space initially set as “a space SP that requires control giving priority to safety”, It was found that avoidance operation was possible. From the “space SP requiring control giving priority to safety” to the “space FW allowing control giving priority to productivity” from the space that has been found to be prevented from entering the entry prohibition area 14 by the avoidance operation thereafter. The logic to switch to was adopted. That is, a process of learning and expanding the “space FW that allows control giving priority to productivity” was added. By adding this process, a space in which the distance from the reference point 16 of the entry prohibition area 14 to the human hand holding the handle 6 is 25 cm to 15 cm is determined from the “space SP that requires control giving priority to safety”. It was confirmed that it was possible to switch to “space FW allowing control giving priority to productivity”.
After performing the process of expanding the “space FW that allows control giving priority to productivity” by learning, the experiment shown in FIG. 4 is performed again. As a result, safety control is activated while there is an avoidance operation. I was able to reduce the number of times I ended up. Even if the “space FW allowing control giving priority to productivity” was expanded, no dangerous situation occurred.
[0028]
(Warning warning)
When learning the boundary between the space SP that requires control giving priority to safety and the space FW that allows control giving priority to productivity, when entering the space SP that requires control giving priority to safety It is preferable to add a warning for causing a human to take an avoidance action and to take the avoidance action. By encouraging avoidance behavior, the space FW that allows control giving priority to productivity can be greatly expanded. When the space FW that allows control giving priority to productivity is expanded by giving a warning, a process for issuing a warning under the same conditions should be used in the actual operation.
[0029]
(Learning avoidance trajectory group)
Even if the evidence theory is adopted, if it is determined that the rush trajectory D is in the “space FW that allows control giving priority to productivity”, the safety control is activated. Experiments have shown that it may be erroneously determined to be a rush trajectory.
As a result of exploring the cause, it was found that the number of avoidance trajectory teaching data is small. While human movement patterns vary and employ a wide variety of avoidance trajectories S, it is found that the initial number of teachings is insufficient to teach the avoidance trajectory group S corresponding to the variety in advance. It was.
Therefore, a process of adding to the teaching data of the avoidance trajectory group S when a trajectory that is later determined to be the avoidance trajectory S is added.
As a result, trajectories that are erroneously determined as rush trajectories before being added to teaching data and safety control is activated are not erroneously identified after addition of teaching data, and the chances that safety control is activated unnecessarily can be reduced. Was confirmed. The number of teaching data of the avoidance trajectory group S is increased by learning, and then the experiment whose result is shown in FIG. 4 is performed again. As a result, the number of times that the safety control is activated is reduced to almost zero while there is an avoidance operation. I was able to.
[0030]
(Warning warning)
When the teaching data of the avoidance trajectory group S is added, when entering the “space FW where flexible safety control is allowed”, a warning for causing the human to take the avoidance action is added and the avoidance action is taken. It is preferable. The teaching data of various avoidance trajectory groups S can be obtained by prompting the avoidance action. When the avoidance trajectory and the entry trajectory are determined using the teaching data of the avoidance trajectory group S that can be avoided by warning, a process for issuing a warning under the same conditions should be used during actual operation.
[0031]
(Aspects of safety control)
There are various modes of safety control that are activated when it is determined that safety control needs to be activated.
The simplest is to forcibly stop the movement of the robot, and if it is forcibly stopped, the situation in which the robot 2 and the person 8 enter the entry prohibition area 14 can be prevented even if a human error occurs. it can.
It is also possible to force an avoidance trajectory instead of a forced stop. If it can be corrected to the avoidance trajectory without making the human feel uncomfortable, the operation can be continued while correcting the human error when the human error occurs.
Instead of directly restricting the movement of the robot, it is possible to employ a mode in which an error is solved by acting on a human error. A warning by sound, a warning by a physical stimulus such as vibration of the steering wheel, or a warning by weakening the assisting force by the robot and increasing the force required for the operation can be used.
[0032]
(System configuration of control device)
As described above, the process until the skill assist robot of this embodiment is safely controlled has been described.
FIG. 5 shows a system configuration of the skill assist robot 2 and the control device 4. In FIG. 5, the control system for reducing the force required by the human by demonstrating the assist force that reduces the force required for the operation by the skill assist robot 2 is omitted. Yes.
The skill assist robot 2 inputs the output of an encoder incorporated in a motor that drives a joint or the like to the control device 4 (6). The control device 4 is provided with a trajectory calculation unit 8 that calculates the trajectory of the skill assist robot 2 from time to time based on the input encoder output. When the trajectory of the skill assist robot 2 is calculated, the posture, speed, and moving direction of the robot are clarified, and it is possible to calculate where the movement state of the robot is located in the space shown in FIG. . If the human hand holds the handle, the trajectory of the human hand can be obtained by the trajectory of the handle. If the robot 2 includes a sensor for determining where the human hand is touching the robot 2, the position of the human hand can be calculated even if the human hand is away from the handle 6. The position of the human hand can be calculated from the output of the encoder incorporated in the motor and the information on the contact position of the human hand specified by the sensor. The position in the space shown in FIG. 3 is specified by the motion state specifying unit 10. The position in space changes with time.
If the calculated trajectory is an avoidance trajectory, it is stored in the avoidance trajectory group information storage means 17 at the learning end. If it is a rush trajectory, it is stored in the rush trajectory group information storage means 19 at the learning end. The avoidance trajectory that has been found to be an avoidance trajectory during actual operation is additionally stored in the avoidance trajectory group information storage means 17 at the learning end. The learned avoidance trajectory group is modeled to obtain an avoidance trajectory group model (16). The learned rush track group is modeled to obtain a rush track group model (18).
The control device 4 is provided with means 22 for storing a space SP that requires control giving priority to safety and a space FW that allows control giving priority to productivity. When a new avoidance trajectory is found, the boundary between both spaces SP and FW is learned (20), and the space FW that allows control giving priority to productivity is expanded.
The control device 4 is provided with means (program) for determining whether the observed trajectory is an avoidance trajectory or a rush trajectory based on evidence theory (12). In this determination process, the avoidance trajectory group model 16, the rush trajectory group model 18, the space SP that requires control giving priority to the safety stored in the storage means 22, and the space information that allows the flexible safety control FW. Is referenced.
If it is determined by the determination device 12 that it is a rush track, or if it cannot be determined in a space that requires control giving priority to safety (24), the safety control means 26 is activated and skill assist is performed. The behavior of the robot 2 is regulated to ensure safety (28).
[0033]
In this embodiment, safety control is performed so that the handle position held by a human does not enter the entry prohibition area. Instead of this, or in addition to this, for example, safety control may be performed so that the workpiece held by the robot does not enter the entry prohibition area. The present technology can be applied by paying attention to an arbitrary position of the robot or an arbitrary position of the workpiece held by the robot.
[0034]
The present invention has been described technically herein. The research or study lurking behind the technology is explained in detail in the paper previously published by the present inventor, and will not be repeated. The above paper is submitted to the present application as a document certifying that the present application is subject to the exception of loss of novelty, and forms part of the disclosure of the present application. Even if it is not permissible to replace the specification by quoting a document that has not been filed with the application, the above paper is submitted for this application and is included in the examination history type of this application It can be viewed by a third party. It is a document that can be referred to as a document that forms part of the present specification. This specification refers to the paper and discloses all the matters described in the paper.
[Brief description of the drawings]
FIG. 1 shows how a human and a skill assist robot work together.
FIG. 2 is a plan view of the vicinity including an entry prohibition area, and illustrates an observation symbol.
FIG. 3 schematically shows a space that requires control giving priority to safety and a space that allows control giving priority to productivity.
FIG. 4 shows a comparison between non-evidence theory and evidence theory, and shows the number of accidents and the like.
FIG. 5 shows a system configuration of a skill assist robot and a control device.
[Explanation of symbols]
2: Skill assist robot
4: Control device
S: Avoidance trajectory
D: rush track
SP: Space that requires control giving priority to safety
FW: Space that allows control giving priority to productivity

Claims (5)

It is a robot control device that changes the trajectory by human operation,
When a part of a robot is gripped by a human and the robot is moved, the point of interest set on the robot, the person gripping the robot, or the workpiece gripped by the robot does not enter the intrusion prohibited area. To control
A combination of “distance, speed, and direction” enters in a space that has dimensions of distance, speed, and direction to the entry prohibited area that is within the movable range of the point of interest and that must be prohibited from entering. “Space that requires safety control” because it is a combination that is highly likely to enter a prohibited area, and “ First space that requires control that prioritizes safety because no avoidance trajectory to be described later has been found. a ", and means for storing separately the second space" that allows the priority control productivity to "avoidance path has been found,
Means for storing an avoidance trajectory group model describing a learned avoidance trajectory group , which is an avoidance trajectory group of the focus point for avoiding the focus point from entering the entry prohibition area;
Means for storing a rush trajectory group model describing a learned rush trajectory group , which is a rush trajectory group of a focus point where the focus point enters the rush prohibition area;
Each of the measured actual trajectory described by the position and speed and direction of the target point, the avoidance path cloud model and the inrush track group model while human manipulates that adjusts the trajectory of the robot a probability is calculated and the measured actual trajectory was, whether belonging to the avoidance route group, means for determining whether or belongs to rush track group, or not determinable belonging to,
While the position, speed, and direction of the point of interest measured while the human is operating to adjust the trajectory of the robot are in the second space, it is determined that the measured actual trajectory belongs to the rush trajectory group Unless the robot is operated by a human, the robot continues to operate, while the measured position, speed, and direction of the point of interest are in the first space , unless the measured actual trajectory is determined to belong to the avoidance trajectory group. A control device for a robot coexisting with a human having means for restricting the continuation of robot operation by the human. The control apparatus according to claim 1, wherein the rush trajectory group model and the avoidance trajectory group model are stored as a hidden Markov model. It is a robot control method that changes the trajectory by human operation,
When a part of a robot is gripped by a human and the robot is moved, the point of interest set on the robot, the person gripping the robot, or the workpiece gripped by the robot does not enter the intrusion prohibited area. To control
A combination of “distance, speed, and direction” enters in a space that has dimensions of distance, speed, and direction to the entry prohibited area that is within the movable range of the point of interest and that must be prohibited from entering. “Space that requires safety control” because it is a combination that is highly likely to enter a prohibited area, and “ First space that requires control that prioritizes safety because no avoidance trajectory to be described later has been found. And “a second space that allows control to prioritize productivity because an avoidance trajectory has been discovered ” and measuring in advance,
Learning in advance and modeling an avoidance trajectory group of a point of interest that avoids the point of interest from entering the entry prohibited area;
Learning and modeling in advance a rush trajectory group of a point of interest where the point of interest rushes into the rush prohibition region;
Each of the measured actual trajectory described by the position and speed and direction of the target point, the avoidance path cloud model and the inrush track group model while human manipulates that adjusts the trajectory of the robot a probability is calculated, and determines the measured actual trajectory was, whether belonging to the avoidance route group, or belongs to rush track group, or whether it is not determinable process belonging to,
While the position, speed, and direction of the point of interest measured while the human is operating to adjust the trajectory of the robot are in the second space, it is determined that the measured actual trajectory belongs to the rush trajectory group Unless the robot is operated by a human, the robot continues to operate, while the measured position, speed, and direction of the point of interest are in the first space , unless the measured actual trajectory is determined to belong to the avoidance trajectory group. A method for controlling a robot coexisting with a human having a process of restricting the continuation of robot operation by the human. It measured the position and velocity and direction of the target point is avoidance route when in the first space is found, the state space position and the speed and direction of at measured target point that avoidance path is found belongs The control method according to claim 3, further comprising a step of changing the first space to the second space .   5. The control method according to claim 3, further comprising a step of adding a newly discovered avoidance trajectory to a previously learned avoidance trajectory group.

Images