JP7388970B2 - robot system - Google Patents

Description

TECHNICAL FIELD This disclosure relates to a robot system.

A known technique is to teach a robot off-line, then acquire images with a camera attached to the robot while the robot is working, and correct the position and orientation of the robot based on the acquired images (e.g. , see Patent Document 1).
The robot's position taught offline under an ideal environment varies in the actual work environment due to various factors, so by correcting it using image data, the robot can be accurately positioned in relation to the actual workpiece. It can be made to work.

Patent No. 4266946

However, for example, when a camera is attached to the hand of a robot, the camera is photographed from a position close to the workpiece, so that only a partial image of the workpiece can be obtained. In such a situation, when extracting the features of the workpiece surface from image data, depending on the part of the workpiece being photographed, the error in the estimated robot position may become large, and an appropriate amount of correction may not be obtained. There is.

Therefore, there is a need for a robot system that can appropriately correct the position of the robot even if the error in the position of the robot estimated based on sensor detection results fluctuates.

One aspect of the present disclosure includes a robot, a plurality of types of sensors capable of detecting information for calculating coordinates of a tool tip point of the robot, and a processor, and the processor controls the tool tip point according to an operation program. A point group consisting of a plurality of points is generated in the vicinity of a first point and a second point on a motion trajectory of has reachable coordinates, a weight indicating a probability of existence is set for each point of the point cloud based on the information detected by each of the sensors, and the point cloud is divided into groups. , extract the group with the highest probability, calculate the centroid position of the extracted group from the coordinates and weights of the points belonging to the group, and calculate the calculated point among the points belonging to the extracted group. The robot system replaces the coordinates of the center of gravity as new coordinates of the second point when the maximum distance from the center of gravity to the point located farthest from the center of gravity is less than or equal to a predetermined threshold.

1 is a perspective view showing a robot system according to an embodiment of the present disclosure. FIG. 2 is a block diagram showing the robot system of FIG. 1. FIG. 2 is a flowchart illustrating a tool tip point correction process performed by the robot system of FIG. 1. FIG. 4 is a flowchart illustrating a point group generation routine in the correction process of FIG. 3. FIG. 4 is a flowchart illustrating a weighting routine in the correction process of FIG. 3. FIG. 4 is a flowchart illustrating a point group narrowing down routine in the correction process of FIG. 3. FIG.

A robot system 1 according to an embodiment of the present disclosure will be described below with reference to the drawings.
As shown in FIG. 1, a robot system 1 according to the present embodiment includes a six-axis articulated robot 2, a camera (sensor) 3 attached to the wrist of the robot 2, and a control device that controls the robot 2. (processor, memory) 4.

The robot 2 has six operating axes driven by servo motors (not shown), and the rotation angle of each operating axis is detected by an encoder (sensor) 5 provided on the servo motor.
Therefore, the position of the tool tip point TCP of the robot 2 can be calculated based on the rotation angle of each motion axis detected by the encoder 5, and can also be calculated based on the image captured by the camera 3. can.

The position of the tool tip point TCP based on the image can be determined, for example, by photographing the workpiece with the camera 3, processing the acquired image, and extracting feature points using a known method. is calculated geometrically from the calculated position of the camera 3.

The control device 4 reads the operation program taught in the offline operation, operates each axis of the robot 2 according to the operation program, and changes the position of the tool tip point TCP at predetermined time intervals, for example, every 0.1 seconds. Estimate and memorize. As a result, when the operation program is executed next time, the operation trajectory is corrected so that the tool tip point TCP passes through the position estimated at each time point.

More specifically, the control device 4 stores in its memory a first table that is associated with the coordinates of the tool tip point TCP of the robot 2. The first table stores a second table in which point groups are associated with velocity vectors on the motion locus of the tool tip point TCP. The first table and the second table are three-dimensional tables.

The point group includes a plurality of points randomly set in a range that the tool tip point TCP can reach while moving from the first point on the operation trajectory of the tool tip point TCP to the second point after a predetermined time. There is. The coordinates of each point are stored in the second table.

Therefore, as shown in FIG. 3, when the operating program is executed, a point cloud is generated at each point in time during the execution (step S1). In generating the point cloud, as shown in FIG. 4, the coordinates of the current point (second point) placed on the motion trajectory of the tool tip point TCP in the motion program and the coordinates of the first point a predetermined time ago are used. is calculated (step S11).

Next, a second table corresponding to the calculated coordinates of the second point is read from the first table (step S12). Further, a velocity vector directed from the first point to the second point is calculated (step S13), and a point group corresponding to the velocity vector is read out from the read second table (step S14). This generates a point cloud.

Next, a weight is given to each point of the generated point group (step S2).
In step S2 of weighting, as shown in FIG. 5, first, one of the sensors 3 and 5 is selected (step S21). For example, if the encoder 5 is selected as the sensor, the coordinates of the tool tip point TCP are calculated from the rotation angles of the respective operating axes detected by the selected encoder 5 (step S22).

Next, one point to which no weight is given is selected (step S23), and the distance L between the coordinates of the selected point and the coordinates of the tool tip point TCP is calculated (step S24). The weight given to each point is inversely proportional to the distance L between the coordinates of the tool tip point TCP and the coordinates of each point, which is calculated from the rotation angle of each motion axis detected by the encoder 5.

Therefore, the weight W is calculated by dividing the constant A by the distance L (step S25). The calculated weight W is stored in association with the coordinates of each point (step S26).

When the calculated coordinates of the tool tip point TCP include a large error, a smaller constant A is given than when the error is small. The constant A may be set to the same value for all the sensors 3 and 5, or may be set individually in advance by conducting an experiment to measure the error for each sensor 3 and 5. Further, the constant A may be automatically adjusted by machine learning.

It is determined whether weights W have been assigned to all points (step S27), and if there are points to which no weights W have been assigned, the steps from step S23 are repeated.
If weights W have been assigned to all points, it is determined whether weights W have been calculated for all sensors 3 and 5 (step S28).

If there are other sensors 3, 5 whose weights W have not been calculated, the other sensors 3, 5 are selected (step S29), and the steps from step S22 are repeated.
When the camera 3 is selected as another sensor, the coordinates of the tool tip point TCP are calculated from the image acquired by the camera 3 (step S22), and a weight W is given to all points (step S22). S23 to S27).

When two sensors, a camera 3 and an encoder 5, are provided, two weights W calculated for each sensor 3 and 5 are generated as a weight for each point.
The two generated weights W are both stored as weights for each point in association with the coordinates of each point.

Next, the points given weights W are divided into groups, and the point group is narrowed down (step S3). Grouping of points is performed as shown in FIG.

First, an arbitrary point not assigned a group number is selected (step S31), and a group number is assigned to the selected point (step S32). Next, it is determined whether or not there are other points to which no group number is assigned within a predetermined distance X from the point to which a group number is assigned (step S33), and all points existing within the range are determined. The same group number is assigned to (step S34).

Then, the steps from step S33 are repeated. In step S33, if it is determined that there are no other points to which group numbers are assigned within a predetermined distance X from all the points to which group numbers are assigned, there is a point to which no group numbers are assigned. It is determined whether or not to do so (step S35).

If there is a point to which no group number has been assigned, the group number is changed (step S36), and the steps from step S31 are repeated. If all points have been assigned group numbers, the weights of all the points belonging to each group are summed for each group (step S37).

In this case, if a point has multiple weights, all weights are summed. Then, leaving the point group with the largest total weight, the other points are deleted (step S38). This narrows down the point cloud.

Next, as shown in FIG. 1, the coordinates of the center of gravity of the narrowed down point group are calculated based on the coordinates and weights of each point (step S4). Then, the distance (maximum distance) from the calculated center of gravity position to the farthest point is calculated (step S5), and it is determined whether the calculated maximum distance is less than or equal to a predetermined threshold (step S6).

If the maximum distance is greater than a predetermined threshold, all points are deleted and a point group is randomly generated inside a sphere centered at the center of gravity and having the maximum distance as a radius (step S7). Then, the process from step S2 is repeated.
In step S6, if the calculated maximum distance is less than or equal to the predetermined threshold, the coordinates of the center of gravity are stored as the coordinates of the second point to correct the coordinates (step S8).

In this way, by executing the operation program taught offline, the coordinates of the passing point located on the operation locus of the ideal tool tip point TCP are determined based on the information detected by the plurality of sensors 3 and 5. The correction is made based on the coordinates of the tool tip point TCP calculated by

According to the robot system 1 according to the present embodiment, the coordinates calculated based on the information detected by the plurality of sensors 3 and 5 each include errors, but these errors are balanced by weighting. This has the advantage that appropriate correction can be made with careful consideration.

In the present embodiment, the encoder 5 and camera 3 provided in the servo motor are exemplified as sensors capable of detecting information for calculating the coordinates of the tool tip point TCP, but the sensor is not limited to this. do not have. For example, if the robot 2 is equipped with an acceleration sensor, the coordinates of the tool tip point TCP can be calculated by integrating the detected acceleration twice.
In addition, if one or more laser sensors are provided to detect the shape of the workpiece, the coordinates of the tool tip point TCP are calculated based on the distance information between the laser sensor and the workpiece detected by each laser sensor. can do.

1 Robot system 2 Robot 3 Camera (sensor)
4 Control device (processor, memory)
5 Encoder (sensor)
X Predetermined distance W Weight TCP Tool tip point

Claims (6)

robot and
multiple types of sensors capable of detecting information for calculating the coordinates of the tool tip point of the robot;
Equipped with a processor,
The processor is
generating a point group consisting of a plurality of points near a first point and a second point on the motion trajectory of the tool tip point according to the motion program;
each point of the point cloud has coordinates reachable by the tool tip point from the first point to the second point;
setting a weight indicating the probability that each point of the point cloud exists based on the information detected by each of the sensors;
dividing the point cloud into groups and extracting a group with the highest probability;
Calculating the centroid position of the extracted group from the coordinates and weights of the points belonging to the group,
Among the points belonging to the extracted group, if the maximum distance to the point located at the farthest position from the calculated center of gravity is less than or equal to a predetermined threshold, the coordinates of the center of gravity are The robot system replaces the new coordinates of the second point. The processor includes:
2. If the maximum distance is greater than the threshold, a point group is generated inside a sphere having the center of gravity as the center and the maximum distance as the radius, and the steps from setting the weight are repeated. robot system. comprising a memory that stores a first table that is stored in association with the coordinates of the second point;
the first table stores a plurality of second tables storing coordinates of the point group in association with velocity vectors heading from the first point to the second point;
The processor,
reading the second table corresponding to the coordinates of the second point from the first table stored in the memory, and reading the coordinates of the point group corresponding to the velocity vector from the read second table; The robot system according to claim 1, wherein the point cloud is generated by: The processor,
4. Any one of claims 1 to 3, wherein for each coordinate of the tool tip point detected by each of the sensors, a value inversely proportional to the distance from the coordinate to each of the points is set as a weight of each of the points. Robotic system described. The processor,
The robot system according to claim 4, wherein the weight is determined for each sensor by machine learning. The processor,
selecting any one of the points in the point group;
give a group number to the selected point;
Repeating assigning the same group number to the points whose distance from each of the points assigned the group number is less than or equal to a predetermined distance until there are no more points assigned the same group number;
If there is a point to which the group number is not given, select another arbitrary point to which the group number is not given;
Change the group number given,
6. The robot system according to claim 1, wherein the point group is divided into groups by repeating the step of assigning a group number to the selected point.

Images