JPH0774970B2 - Robot work planning device - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION [Industrial field of application] The present invention relates to a work planning apparatus for a robot that carries out work planning such as grasping and moving a target object.

[Prior Art] Conventionally, a robot that automatically determines an action plan of a robot for grasping, moving, and placing (pick and place) the target object based on an initial state of the target object and a target state after assembly. Work planning devices are known.

Here, one of the robot work plans is a target object grasp plan. In this grasping plan, considering the approach to grasping posture and the ease of movement after grasping,
It is considered good to grasp the object from the direction in which there is a wide space as far as possible from the surrounding obstacles.

Therefore, the present inventors have proposed a grasp planning method capable of grasping a target object from a wide space direction both during picking and placing during pick-and-place work by using a space description near the grasping position (No. 5th Annual Conference of the Robotics Society of Japan (November 1987).

In this example, a spatial description near the target object in the initial state and the target state of the target object is performed. This spatial description is
The space near the target object is divided into triangular pyramids (space cones) centered on the grasp position, and the interference between the surrounding objects and all the space cones is examined. That is, the length of the part from the vertex of the space cone (grasping position) to the first intersection with the surrounding object is more than the distance at which the robot's grasping part (hand) can exist and the approaching motion, that is, the open space cone. Ask.

Next, this open space cone is obtained in both the initial state and the target state, and the space cone common to both is obtained, that is, the common open space cone. Then, in order to investigate the distribution of the common open space cones, distance conversion is performed for all common open space cones. This distance conversion is performed by finding the shortest distance from the space cone that did not become the common open space cone for each common open space cone, and the direction of the open space cone with the large distance conversion value is the space not occupied by the object. Indicates that the area is wide.

Therefore, in consideration of stable grasping of an object with two parallel fingers, by selecting one with a large distance conversion value among the intersections of the grasping center plane formed in the center of two parallel planes in the object and the common open space cone, The attitude to grasp is determined.

According to the above-mentioned conventional example, the target object can be grasped from the direction in which a wide space as far away as possible from the surrounding obstacle exists, and the work plan of the pick-and-place work with the fewest problems can be determined.

[Problems to be Solved by the Invention] However, in the conventional work plan determination method, when selecting the grasping posture of the object, the relation between the object and the hand of the robot is grasped only in the central plane. Absent. Therefore, after determining the grasping posture as described above, it is determined whether or not the grasping posture can be actually grasped in consideration of the placement positions and shapes of the robot hand and the target object. Therefore, there is a problem that the method of avoiding the interference between the object and the hand becomes a trial and error, which makes the decision of the plan inefficient.

In addition, in some work plans, the target state cannot be directly transferred to the target state, and the target object must be placed once, and the holding state must be changed before moving to the target state.
In the conventional method, in this case, a common open space cone has to be created for all combinations of the states of the target object, and there is a problem that the efficiency is low.

The present invention has been made to solve the above problems, and an object of the present invention is to provide a robot work planning apparatus that can efficiently select a grasping posture that can be actually grasped.

[Means for Solving the Problem] A work planning apparatus for a robot according to the present invention provides a position for a predetermined grasping position of an object based on the shape of the object and the shape and moving direction of an object grasping part of the robot. Grasping posture candidate calculation means for calculating grasping posture candidates that are postures capable of grasping an object, and dividing the space near the object into space cones of a predetermined size with the grasping position as a vertex and From the open space cone selection means that selects an open space cone that does not interfere with surrounding objects, distance conversion means that obtains the distance from the surrounding object for each of the obtained open space cones, and each obtained by the distance conversion means. And a grasping posture determining means for determining a grasping posture based on both the distance in the open space cone and the grasping posture candidate obtained by the grasping posture candidate calculating means.

[Operation] The robot work planning apparatus according to the present invention has the above-described configuration and obtains grasping posture candidates in advance. Then, only those that match the grasping posture candidate are selected from the open space cones. Therefore, the selected space cone can always be grasped, and the grasping direction can be efficiently selected. In addition, the pick and place grasping posture is determined only from the open space cones that match the grasping posture candidate without obtaining the common open space cone from the two open space cones. It becomes possible to do.

[Examples] Examples of the present invention will be described below with reference to the drawings.

FIG. 1 is a block diagram showing the overall configuration of a work planning device according to the present invention.

Shape of target object, position and orientation of initial state and target state,
Conditions such as the state of surrounding obstacles are described in the three-dimensional model 1 of the environment. Then, using the data of the three-dimensional model 1 of this environment, various calculations are performed to create a work plan. In this embodiment, a grasping posture candidate calculation for obtaining a posture that can be grasped from the shape of the target object is calculated. Means 2, stable posture calculating means 3 for obtaining a state in which the target object can be stably placed on the table, rotational symmetry calculating means 4 for calculating the rotational symmetry of the target object, grasping for creating a geodesic dome near the grasping position Near-position space description creating means 5, grasping position near-space description analysis means 6 for analyzing distance conversion and the like from this geodesic dome, change-over place determining means for deciding a change-over place to a wide space on the table when a change-over is required Have 7.

Then, the data obtained by these analyzes is fed back to the three-dimensional model 1 of the environment, whereby data for work planning is created.

A work plan creation means 8 is connected to the environmental three-dimensional model 1, and the work plan creation means 8 creates a work plan from the initial state to the target state.

Further, the work plan creating means 8 has a state transition determining means 9 therein, and the state transition determining means 9 determines the transition for each individual state.

Here, FIG. 2 shows an overall flow chart of the operation of the robot work planning apparatus according to the present embodiment.

In this way, first, the three-dimensional model 1 of the environment is created from the data of the initial state, the target state, the surrounding environment, etc. (S1). Next, from the shape of the target object in the three-dimensional model 1, the grasping candidate posture calculating means 2 calculates a grasping candidate posture (S2), the stable posture calculating means 3 calculates a stable posture (S3), and the rotational symmetry calculating means 4 calculates. The rotational symmetry calculation (S4) is performed, and the result is described in the three-dimensional model 1.

Next, the work plan creation means 8 creates a work plan using the data of the three-dimensional model 1 (S5). At this time, the state transition determination means 9 activates the grasped position neighborhood space description creation means 5, Data for discrimination is described in the three-dimensional model 1. Then, while determining the state transition, the solution of the work plan from the initial state to the target state is obtained by searching (S8).
If the work plan requires a change of holding pattern, the holding position determining means 7 determines the holding position (S6).

The contents of each step will be described below in order.

Calculation of grasping posture candidate The operation of the grasping posture candidate calculating means 2 will be described with reference to FIG. For example, consider a grasping posture candidate for an L-shaped object as shown in FIG.

First, the grasp position is decided, but since the grasp position can be set as many as necessary, the distance from the peripheral side of the object,
It is determined by presetting the interval between the grasping positions to a predetermined value. In this example, as shown in FIG. 3 (A), four points (indicated by X marks in the figure) are determined.

Next, a direction that can be grasped by the robot hand is represented by a straight line centering on this grasping position. That is, with respect to the above-mentioned four grasping positions, FIG.
The grasping posture candidate of (E) is calculated. In this example, there are a total of 138 grasping posture candidates.

The distance between the straight lines is a predetermined distance. In this way, the grasping posture candidate can be obtained from the shape of the hand, the shape of the target object, and the like.

Stable posture calculation When the target object is changed, the target object must first be placed on a horizontal table. Therefore, the stable posture calculating means 3 obtains a stable posture that can be stably placed for this holding change.

First, consider a convex hull that encloses a target object without forming a concave part, and consider the perpendiculars from the center of the object on each surface of this convex hull. And if the foot of this perpendicular is in the surface of the convex hull, it can be placed stably with the surface and the table surface in contact with each other. For example, if it is an L-shaped object shown in FIG.
Six stable postures as shown in FIGS. (A) to (F) are required. Then, serial numbers are assigned to the obtained stable postures.

Calculation of rotational symmetry Here, when a symmetric object has a rotationally symmetric shape, a model that does not recognize this rotational symmetry treats the same grasping posture as a different one.

That is, when the target object is recognized, it is usually recognized by three-dimensional coordinates of X, Y, and Z. Therefore, 3 in FIGS. 5 (A) to 5 (C)
In the three states, if the axis indicated by the arrow is a specific axis (for example, the X axis), all three states will be recognized as different.

Thus, when not considering the symmetry of the target object,
Depending on how the object coordinate axes of the recognition system are taken, even the same state is recognized as a different state. Therefore, as shown in FIGS. 6A to 6C, even if the grasping postures are substantially the same, they are recognized as different. Therefore, depending on how the coordinate axes are picked and picked up, it may be determined that a practical grasping plan is impossible.

Further, since the substantially same stable postures are recognized as different ones as shown in FIGS. 5 (A), (B) and (C), the number of stable postures increases and the number of routes in the work plan increases. There will be many and unnecessary searches will be performed.

In this embodiment, the rotational symmetry of the object is derived and the relationship between the grasping postures is described. Then, since the identity between the grasping postures is judged by this description, it is also possible to carry out a plan of re-holding a symmetrical object having rotational symmetry. Further, since the identity based on the rotational symmetry of the stable posture is determined, useless search can be omitted.

Rotational symmetry calculation algorithm A specific algorithm for describing the rotational symmetry is shown below.

(1) For one surface of the target object, a conversion that rotates about the center of gravity of the target object is performed, and a conversion that overlaps the surface of the target object (including the surface that is itself) is obtained.

That is, the following four points are checked as an initial check.

(A) The area is the same.

(B) The perimeter is the same.

(C) The distance from the center of gravity of the target object to the center of gravity of the surface is the same.

(D) The angle formed by the vector from the center of gravity of the object to the center of gravity of the surface and the normal vector of the surface is the same.

And about the thing which satisfy | fills these (a)-(c),
The conversion of overlapping surfaces is obtained as follows.

If the angle between the vector from the center of gravity of the object to the center of gravity of the surface and the normal vector of the surface is other than 0 °, there is a possibility that the surfaces may overlap due to the rotation conversion centered on the center of gravity of the object. There is only one. Therefore, the surface to be overlapped is obtained by this rotation transformation, and the symmetric transformation is obtained.

On the other hand, when the angle formed by the vector from the center of gravity of the object to the center of gravity of the surface and the normal vector of the surface is 0 °, the degree of freedom of rotation around the vector from the center of gravity of the object to the center of gravity of the surface is set. Remain. For this reason, whether or not the surfaces overlap with each other is determined by the length and angle of the edges of the surfaces, and the overlap conversion by the rotation about this axis is obtained.

(2) Check whether the transformation obtained in (1) is actually a symmetric transformation. That is, the transformation is applied to all the vertices of the object, it is checked whether they all overlap the vertices of the object before the transformation, and it is checked whether or not the transformation is symmetric. Then it returns a list of overlapping transformations.

(3) Delete duplicate conversions in the list obtained in (2).

In this way, it is possible to describe postures that have the same rotational symmetry.

Therefore, in the determination between the states, these can be treated as the same, and the grasping plan can be made appropriate.

Creation of space description near grasping position In order to determine the posture of grasping the target object from a wide space in the working environment, a space description near the target object is created by the grasping position near space description creating means 5, and
The created spatial description is analyzed by the grasp position proximity spatial description analysis means 6 and is described in the three-dimensional model 1 of the environment. First, a geodesic dome is created by dividing the space into triangular pyramids (space cones) with the grasping position as the apex. As shown in FIG. 7, this geodetic dome has different numbers of space cones depending on the level, and 20 levels at level 0 and 80 levels at level 1,
There are 320 faces in level 2 and 1280 faces in level 3.

Then, with respect to the geodesic dome created in this way, a space cone (open space cone) whose distance from the apex (grasping position) of the space cone to the interference with the surrounding object is not less than a certain value is obtained. That is, only the geodesic domes created for one grasped position are selected so as not to interfere with surrounding objects. When this process is performed for the state shown in FIG. 3 (A), the space cone as shown in FIG. 8 (A) is selected as the open space cone.

Next, on the surface of the geodesic dome, distance conversion is performed for each open space cone, and the distribution state of the open space cone is investigated. That is, regarding the open space cone shown in FIG. 8 (A), the peripheral space cone adjacent to the non-open space cone (space cone not shown in FIG. 8 (A)) is set to the distance 1. And this distance 1
A space cone adjacent to the space cone of is set to a distance 2. In this way, the shortest distance is obtained for all open space cones.
With this, the shortest distance (distance conversion value) from all open space cones to neighboring non-open space cones can be obtained.

This distance conversion value is an index for the size of the space viewed from the center of the measurement dome. That is, the open space cone having the largest distance conversion value is the center of the direction in which the locally widest space exists. Therefore, by selecting an open space cone having a large distance conversion value, it is possible to determine the optimum grasping direction in one state. In this example, the 8th
The 16 open space cones shown in FIG. 6B can be selected as the one having the largest distance conversion value.

State Transition Determination Next, the state transition determination means 9 determines whether or not the target object can transit (pick and place) from one state to another state.

That is, it is determined whether or not the state transition determination means 9 has a grasping posture that can grasp the same place in two states of the target object when transitioning the target object in one state to another state. If there is a grasping posture capable of grasping the same place, transition is possible, and if not, transition is impossible.

Here, this determination will be specifically described.

First, a grasping posture candidate that is highly likely to be grasped in each state is obtained. That is, the grasped near space description analysis means 6
The distance transform value for the open space cone obtained in
Both the grasping candidates calculated by the grasping posture candidate calculating means 2 are combined to describe the distance conversion value corresponding to each grasping posture candidate. As a result, the distance conversion value is described only for the grasping posture that is highly likely to be grasped.

At this time, in the case of an object having symmetry, since the identity of the grasping posture due to the symmetry is recognized as described above, the grasping posture in consideration of this can be obtained.

Then, the description of the distance conversion value for the grasping posture with a high grasping possibility is performed for both of the two target states, and these are compared.

That is, since the grasped postures to which the distance conversion value is given are highly likely to transit between the two states, all of them are obtained.

Here, the obtained grasping posture has a larger conversion value both at the time of picking and at the time of placing.

Therefore, a smaller one is selected from the distance conversion values in the two states for each common graspable posture, and the selected ones are arranged in descending order. And in this order,
Check the operating range and interference check of the manipulator, and select the one that does not have a problem. As a result, the grasping posture in the transition from one state to the next state is determined. The determined grasping posture can be actually grasped and is from the widest space.

In this way, the grasping posture can be determined for the transition from one state to another one state.

However, if it is not possible to directly transit from the initial state to the target state, the holding change must be performed. Therefore, a state transition sequence is created in consideration of holding patterns.

A known obstacle avoidance operation plan or the like is applied to the operation path of the robot.

State Transition Route Plan In the present embodiment, the grasping plan performed in the process of creating the state transition sequence and the plan of the holding location are separated. That is, in the process of creating the state transition sequence, the state of the target object at the time of holding the hand is represented only by the stable posture. Therefore, the state space of the target object is composed of only the initial state, the target state, and a plurality of stable states.

At this time, since the above-mentioned rotational symmetry is also taken into consideration, the state space for an object having symmetry becomes small, and unnecessary search can be omitted.

In the case of the above-mentioned L-shaped symmetric object, the state space for this is only eight states, as shown in FIG.

Then, in such a state space, the pick-and-place work is repeated from the initial state to obtain the state transition series up to the target state. In this case, the state transitions for all combinations are not evaluated. This search is performed by a method based on the A * algorithm of the state space search method.

Search Algorithm This search algorithm will be described using graph symbols.

This graph consists of a set of nodes. And
It is assumed that the nodes are connected by directed branches. If the branch is oriented from node n to node m, node m is said to be a continuation node of contact n. Further, the node n is called a parent node of the node m.

And when displaying the state space using a graph,
State descriptions are entered at the nodes of the graph, and pick-and-place operators are added at the branches. Applying an operator to a node n to obtain its continuation node is called expanding the node n.

In the search in the problem of holding patterns using the state space as described above, there is a possibility that the target state and the state of a stable posture for holding patterns become continuous nodes due to the expansion of the nodes.
Then, one by one is created from continuation nodes with a high possibility, and the possibility of creating subsequent continuation nodes is left behind in the search. That is, at the very beginning, the state transition determination means 9 attempts to transition to the goal node (target state).

After that, the expansion is tried one by one in the order of stable posture numbers, and the path with the smallest cost is selected. The number of expansions is defined for each contact as the number of times expansion is attempted to a continuous node in order to determine the next expansion state.

Here, in order to realize the A * algorithm, the cost value g that has been taken up to that point in an arbitrary state, the predicted value h of the minimum cost related to the target state, and the cost evaluation value f = g + h that is the sum of them Is defined and the cost of the actual shortest path from that to the target state is h ₀ , it is necessary to satisfy h ₀ ≧ h.

Here, the cost value is the number of pick and place operations. Therefore, h is 1 when there is a possibility of extension to the goal node, and 2 when h has already tried to extend to the goal node and failed. This gives h ₀
Since the condition of ≧ h is satisfied, it is understood that the route with the shortest distance can always be derived by the A * algorithm.

Also, since all contact points may be adjacent to the goal node, if there are multiple nodes with the same f, the goal contact may be a continuous node, that is, h is 1 Expand nodes preferentially. Furthermore, if there are several nodes with the same f and g, there is a high possibility that a holding posture that has not been expanded until now exists on the path that is the solution. I try to expand it.

Next, a specific procedure will be described.

(1) Add the start node to the list called OPEN.

(2) If OPEN is empty, it ends with failure. On the other hand, if OPEN is not empty, proceed to the next.

(3) Take out the one with the smallest f from OPEN. If there are a plurality of these, the one with h = 1 is taken out. If there is more than one of this h, the one with the largest number of expansions is taken out of them (the order of expansion of one contact is determined in advance, so the holding posture has not been expanded until now. Will be an extension of).

The node selected here is called n.

(4) If n is a goal node, follow the pointer to guide the route that has become a solution, and end with success. If it is not the goal node, proceed to the next.

(5) Attempt expansion by creating continuous nodes according to the number of expansions.
At this time, the number of expansions of the value continuation node is set to 0. If the number of expansions of n is 0, the expansion to the goal node is attempted, and otherwise, the expansion to a stable posture equal to the number of expansions is attempted. Increase the number of expansions of node n by one. If the number of expansions of the node n reaches the number of stable postures + 1, it means that the expansion to all stable postures is completed, so take out from OPEN and CLOSED
Add to the list called. If the expansion cannot be performed, the process returns to the above (2).

(7) If the continuous node is already on OPEN or CLOSED,
Put it in OPEN and attach a pointer that returns from the continuation node to n.

(8) If the continuous node is already on OPEN, the evaluation value f is compared and the smaller one is left on OPEN.

(9) If the continuous node is already on CLOSED, the evaluation value f is compared. If the new continuous node is smaller, it is taken out from CLOSED and the new continuous node is put in OPEN.

(10) Return to (2).

Here, the value of g at a certain node is obtained from the number of nodes traced so far by tracing the pointer. The value of f is
If the number of expansions of the node is 0, it may change to the goal node in the next expansion, so it is 1. If the number of expansions is 1 or more, it means that the player will aim for the goal node after going through a stable posture. Become.

In this way, the transition state is determined by selecting the transition state while selecting the one having the smaller evaluation value f. Therefore, when reaching the goal, it is guaranteed that the route is the shortest.

Determining holding pattern place After obtaining the state transition series, the holding pattern decision means 7 selects a holding pattern place.

As for the holding place, the two-dimensional space is analyzed horizontally within the robot work range on the table plane, and candidates for the holding place are set in order from the wide space.

This will be described with reference to FIG. Figure 9 (A)
Is an example of a two-dimensional space to be analyzed. The range where the robot can take any grasping posture on the table is indicated by dots.

Then, if a space in which no obstacle exists is selected, a space (dot) as shown in FIG. 9B is selected. Next, the dots thus obtained are subjected to distance conversion from the periphery, and the dot having the largest distance conversion value is selected. This is because a large distance conversion value means that there is the widest space around. Therefore, two dots are selected as shown in FIG. 9 (C), and this point is a suitable place for holding.

Next, a specific algorithm for calculating the holding location will be described.

(1) The work space is divided into small squares (1.5 cm here) on the table, and divided into a square area (obstacle space) where an object other than the grasped object exists and a square area (free space) that does not exist.

(2) An appropriate natural number N is given as a step of the angle θ.

(3) A list L in which distances from the obstacle space are converted for squares in free space on the table and the center positions (X, Y) of these squares are arranged in descending order of distance conversion value
To create.

(4) If list L is empty, the process ends with a failure. On the other hand, if the list L is not empty, the first element is taken out from the list L and assigned to X and Y. Further, θ is set to 0 as an initial value.

(5) θ is updated with θ = θ + 360 / N.

(6) Place the grasped object in a stable posture for holding and place it at the angle X and Y at the angle of θ, and check whether it is possible to realize both the grasped posture for picking and the grasped posture for place already decided. .

(7) If it is feasible, the process ends.

(8) If realization is not possible, determine whether θ exceeds 360 °. If θ is less than 360 °, return to (5), update θ and check again. On the other hand, if it exceeds 360 °, the procedure returns to (4) and the next element is extracted from the list L.

In this way, the holding place is determined.

Example of Reholding Work Plan A reholding operation plan for the work of inserting the L-shaped object in FIG. 10 (A) into the hole as shown in FIG. 10 (F) was performed by the method of the present invention. Here, candidates for the grasping posture of the target L-shaped object are shown in FIG. All the grasping posture candidates
There are 138.

According to the plan created by the apparatus of the present invention, the state transits to the states of FIGS. 10 (A) to 10 (F), but the object of FIG. 10 (A) is near the obstacle. The posture that can be grasped is limited. The grasping posture is limited to one by the above-described grasping posture determining method. Next, the state transition path is followed by the state shown in FIG. 10 (B). The selection of the holding position is set to the place where the influence of the obstacle is least, as described above. Then, the target object is changed over according to the route determined by the state transition route (Fig. 10 (C)),
As shown in FIG. 10 (D), the stable posture (B) of FIG. 4 is changed. Then, from this state, the holding pattern shown in FIG. 10 (E) is changed to the target state shown in FIG. 10 (F).

As described above, this example is an example in which the work cannot be executed unless the holding pattern is changed twice, and the states shown in FIGS. 4 (G) and 4 (B) are selected as the stable states for changing the holding pattern.

[Effects of the Invention] As described above, according to the robot work planning apparatus of the present invention, the distance in each open space cone obtained by the distance conversion means and the grasping candidate obtained by the grasping posture candidate calculating means. Since the grasping posture is determined based on both, an efficient grasping posture can be determined.

[Brief description of drawings]

1 is a block diagram showing an embodiment of a robot work planning apparatus according to the present invention, FIG. 2 is an overall flowchart for explaining the operation of the embodiment, FIG. 3 is an explanatory view showing grasping posture candidates, FIG. 4 is an explanatory view showing a stable posture, FIG. 5 is an explanatory view showing rotational symmetry of an object, FIG. 6 is an explanatory view for grasping an object having rotational symmetry, and FIG. 7 is a geodesic dome. FIG. 8 is an explanatory view showing a space description near a grasped position, FIG. 9 is an explanatory view for explaining selection of a holding position, and FIG. 10 is an explanatory view showing a work plan example. 1 ... Three-dimensional model of environment 2 ... Grasping posture candidate calculation means 3 ... Stable posture calculation means 4 ... Rotational symmetry calculation means 5 ... Grasping position neighborhood space description creating means 6 ... Grasping position neighborhood space description analysis Means 7: holding place determining means 8: work plan creating means 9: state transition determining means

Claims (1)

[Claims] 1. A robot work planning apparatus for planning a grasping operation of an object by a robot from a three-dimensional environment model including an initial state of the grasping object and a state of surrounding objects, the shape of the object, Based on the shape of the object grasping part of the robot, grasping posture candidate calculating means for calculating grasping posture candidates that are postures capable of grasping the object at a predetermined grasping position of the object, and a space near the object are calculated. An open space cone selection means that divides the space cone into a predetermined size with the grasping position as a vertex and selects an open space cone that does not interfere with surrounding objects from this space cone, and the obtained open space cone. Distance conversion means for obtaining the distance from the surrounding object for each of, and the distance in each open space cone obtained by the distance conversion means,
A work planning apparatus for a robot, comprising: a grasping posture determining means that determines a grasping posture based on both grasping posture candidates obtained by the grasping posture candidate calculating means.