JP7462827B2 - CONTROL DEVICE, ROBOT SYSTEM, LEARNING DEVICE, TRAJECTORY MODEL, CONTROL METHOD, AND PROGRAM - Google Patents

Description

The present disclosure relates to a control device, a robot system, a learning device, a trajectory model, a control method, and a program.

Industrial robots are equipped with multiple links and multiple joints connecting the multiple links to enable various movements. A control device controls multiple servo motors that drive the multiple joints of the robot, allowing the robot to perform the desired movement. For example, if there is a difference in the delay of each servo motor, the actual trajectory of the tip link, which is the link located at the tip of the multiple links, may deviate from the target trajectory. An example of a technique for minimizing the error between the actual trajectory and the target trajectory of a robot is disclosed in Patent Document 1.

The robot position teaching device disclosed in Patent Document 1 measures the actual trajectory of the robot and calculates potential teaching points to minimize the error based on the trajectory error, which is the distance between the actual trajectory and the target trajectory. By controlling the robot based on the potential teaching points, the trajectory error is reduced.

Japanese Patent Application Laid-Open No. 11-048176

The robot position teaching device disclosed in Patent Document 1 repeatedly calculates potential teaching points indicating multiple positions in the robot's motion path and controls the robot based on the potential teaching points until the trajectory error falls below an allowable value. If the target trajectory becomes complex, the number of potential teaching points increases, and it takes longer for the trajectory error to converge below the allowable value, making it difficult to bring the robot's actual trajectory close to the target trajectory. As a result, the control accuracy of the robot decreases.

The present disclosure has been made in consideration of the above-mentioned circumstances, and aims to provide a control device, robot system, learning device, trajectory model, control method, and program that provide high control accuracy for robots.

In order to achieve the above object, the control device according to the present disclosure is a control device for controlling a robot, and includes a drive condition storage unit, a reference trajectory determination unit, an actual trajectory acquisition unit, a learning unit, a target trajectory determination unit, and a control unit. The drive condition storage unit stores a drive condition that specifies at least the motion start point and the motion end point of the robot in association with each of a plurality of reference intermediate points. The reference trajectory determination unit acquires one of a plurality of reference intermediate points from the drive condition storage unit, and determines a reference trajectory that passes through the reference intermediate point based on the drive condition for each acquired reference intermediate point. The actual trajectory acquisition unit acquires an actual trajectory of the robot. The learning unit obtains a trajectory error indicating a deviation from an ideal trajectory based on the drive condition of the actual trajectory acquired by the actual trajectory acquisition unit when the robot is controlled according to the reference trajectory for each of the acquired reference intermediate points, learns the correspondence between the reference intermediate points and the trajectory error, and generates a trajectory model indicating a target intermediate point that minimizes the trajectory error according to the drive condition. The target trajectory determination unit determines a target trajectory that passes through the target intermediate point obtained from the trajectory model. The control unit controls the robot according to the reference trajectory or the target trajectory.

The control device according to the present disclosure learns the correspondence between reference waypoints and trajectory errors, generates a trajectory model that indicates a target waypoint that minimizes the trajectory error according to the drive conditions, and controls the robot according to a target trajectory that passes through the target waypoint. Therefore, the control device according to the present disclosure provides high control accuracy for the robot.

FIG. 1 is a perspective view of a robot according to a first embodiment; FIG. 1 is a side view of a robot according to a first embodiment; Block diagram of a robot system according to a first embodiment FIG. 1 is a block diagram showing a hardware configuration of a control device according to a first embodiment. A flowchart showing an example of a learning process performed by the control device according to the first embodiment. FIG. 1 is a diagram showing an example of a working condition table stored in a working condition storage unit according to the first embodiment; FIG. 1 is a diagram showing an example of a reference midpoint in the first embodiment; FIG. 1 shows an example of a reference trajectory and an actual trajectory in the first embodiment. FIG. 1 shows an example of a reference trajectory and an actual trajectory in the first embodiment. FIG. 1 is a diagram showing an example of an error table stored in a learned data storage unit according to the first embodiment; FIG. 1 is a diagram showing an example of a target waypoint table stored in a learned data storage unit according to the first embodiment; 1 is a flowchart showing an example of an operation process performed by a control device according to a first embodiment. FIG. 1 is a diagram showing an example of a target trajectory and an actual trajectory in the first embodiment; FIG. 13 is a perspective view of a robot according to a second embodiment; Block diagram of a robot system according to a second embodiment. A flowchart showing an example of a learning process performed by a control device according to a second embodiment. FIG. 13 is a diagram showing an example of an error table stored in a learned data storage unit according to the second embodiment; Block diagram of a robot system according to a third embodiment. A flowchart showing an example of a learning process performed by a control device according to a third embodiment. FIG. 13 is a diagram showing an example of a working condition table stored in a working condition storage unit according to the third embodiment; FIG. 13 is a diagram showing an example of a working condition table stored in a working condition storage unit according to the third embodiment; Block diagram of a robot system according to a fourth embodiment A flowchart showing an example of a learning process performed by a control device according to a fourth embodiment. Block diagram of a robot system according to a fifth embodiment FIG. 13 is a diagram showing a modified example of a target waypoint table stored in a learned data storage unit according to an embodiment; FIG. 13 is a diagram showing a first modified example of an error table stored in the learned data storage unit according to the embodiment; FIG. 13 is a diagram showing a second modified example of the error table stored in the learned data storage unit according to the embodiment; FIG. 13 is a diagram showing a third modified example of an error table stored in the learned data storage unit according to the embodiment; FIG. 1 is a block diagram of a modified example of a robot system according to an embodiment. FIG. 13 is a perspective view of a first modified example of the robot according to the embodiment; FIG. 13 is a perspective view of a second modified example of the robot according to the embodiment;

Hereinafter, a control device, a robot system, a learning device, a trajectory model, a control method, and a program according to embodiments of the present disclosure will be described in detail with reference to the drawings. In the drawings, the same or equivalent parts are denoted by the same reference numerals.

(Embodiment 1)
Taking a robot having multiple links and multiple joints as an example, a robot system including a robot and a control device for controlling the robot will be described in embodiment 1. The robot has multiple links, multiple joints connecting the multiple links to each other, and multiple motors associated with the multiple joints and driving the corresponding joints.

As shown in Figures 1 and 2, the robot 10 is a vertical articulated robot. In detail, the robot 10 includes a first arm 110, a second arm 120, a third arm 130, a fourth arm 140, a fifth arm 150, a flange 160, and a base 170 as a plurality of links. The flange 160 corresponds to a tip link that is the link located at the tip of the plurality of links. The robot 10 further includes a fixing part 180 that is fixed to a plane and supports the base 170. In the first embodiment, the fixing part 180 is a plate-like member whose main surface is shaped like a square.

1 and 2, the X-axis and Y-axis are set as axes included in a plane parallel to the main surface of the fixed part 180 and perpendicular to each other, and the Z-axis is set as an axis perpendicular to each of the X-axis and Y-axis. The X-axis and Y-axis each extend along the end face of the fixed part 180. If the fixed part 180 is fixed to a horizontal surface, the Z-axis extends in the vertical direction. In the first embodiment, the trajectory of the tip link, specifically, the trajectory of the center of gravity of the tip surface 160a of the flange 160 is expressed in an XYZ orthogonal coordinate system. The origin of the XYZ orthogonal coordinate system is, for example, the position of the center of gravity of the tip surface 160a when the robot 10 is in the initial position. When the robot 10 is in the initial position, it refers to a time when the rotation positions of the motors M1, M2, M3, M4, M5, and M6 described later are in the initial positions.

The first arm 110 is attached to the base 170 in a state in which it can rotate around a rotation axis AX1 parallel to the Z axis. The second arm 120 is connected to the first arm 110 and can rotate around a rotation axis AX2. The third arm 130 is connected to the second arm 120 and can rotate around a rotation axis AX3. The fourth arm 140 is connected to the third arm 130 and can rotate around a rotation axis AX4. The fifth arm 150 is connected to the fourth arm 140 and can rotate around a rotation axis AX5. The flange 160 is connected to the fifth arm 150 and can rotate around a rotation axis AX6. A machining tool can be attached to the flange 160. In other words, the robot 10 has six joints corresponding to the rotation axes AX1, AX2, AX3, AX4, AX5, and AX6.

As shown in FIG. 3, the robot 10 includes motors M1, M2, M3, M4, M5, and M6 corresponding to the rotation axes AX1, AX2, AX3, AX4, AX5, and AX6. The motors M1, M2, M3, M4, M5, and M6 are servo motors. The control device 1 controls the motors M1, M2, M3, M4, M5, and M6 according to the drive conditions, so that the motors M1, M2, M3, M4, M5, and M6 rotate, and the first arm 110, the second arm 120, the third arm 130, the fourth arm 140, the fifth arm 150, and the flange 160 rotate. As a result, the trajectory of the flange 160, which is the tip link of the robot 10, becomes a trajectory according to the drive conditions. The robot 10 further includes encoders E1, E2, E3, E4, E5, and E6 that detect the rotation positions of the motors M1, M2, M3, M4, M5, and M6.

The robot system 100 includes a robot 10 having the above-described configuration, and a control device 1 that controls the robot 10. The control device 1 controls the trajectory of the robot 10, specifically the trajectory of the flange 160, to approach an ideal trajectory according to driving conditions that specify at least the motion start point and motion end point of the robot 10. The motion start point of the robot 10 means the motion start point of the movable part of the robot 10, specifically the motion start point of the flange 160 provided on the robot 10. The motion end point of the robot 10 means the motion end point of the movable part of the robot 10, specifically the motion end point of the flange 160 provided on the robot 10.

In the first embodiment, the control device 1 controls the motors M1, M2, M3, M4, M5, and M6 of the robot 10 in accordance with drive conditions that specify at least the start and end points of motion of the flange 160 of the robot 10. The motors M1, M2, M3, M4, M5, and M6 operate with a delay in response to commands from the control device 1. If there is a difference in the delay in the respective motions of the motors M1, M2, M3, M4, M5, and M6, the actual trajectory that indicates the actual trajectory of the robot 10 deviates from the ideal trajectory based on the drive conditions.

For each reference midpoint corresponding to the drive condition, the control device 1 determines a trajectory error indicating the deviation of the actual trajectory from the ideal trajectory when the robot 10 is controlled based on a reference trajectory passing through the reference midpoint, and learns the correspondence between the reference midpoint and the trajectory error. The reference midpoint indicates a predetermined position that the tip link can pass through from the motion start point to the motion end point. The ideal trajectory is, for example, the shortest trajectory from the motion start point indicated by the drive condition to the motion end point indicated by the drive condition.

After learning the correspondence between the reference waypoints and the trajectory error as described above, the control device 1 generates a trajectory model that indicates the target waypoint that minimizes the trajectory error for each drive condition, and controls the robot 10 according to the target trajectory that passes through the target waypoint. By controlling the robot 10 according to the target trajectory, it is possible to improve the control accuracy of the robot 10.

Each part of the control device 1 will be described below.
The control device 1 includes a drive condition storage unit 11 that stores drive conditions and a plurality of reference waypoints in association with each other, a reference trajectory determination unit 12 that determines a reference trajectory that passes through the reference waypoints, and an actual trajectory acquisition unit 13 that acquires an actual trajectory of the robot 10, specifically, an actual movement trajectory of a flange 160 that is a tip link. The control device 1 further includes a learning unit 14 that determines a trajectory error that indicates a deviation of the actual trajectory from an ideal trajectory for each of the reference waypoints, learns the correspondence between the reference waypoints and the trajectory error, and generates a trajectory model that indicates a target waypoint that minimizes the trajectory error according to the drive conditions, and a learned data storage unit 15 that stores the learning result of the learning unit 14, specifically, the correspondence between the reference waypoints and the trajectory error. The control device 1 further includes a target trajectory determination unit 16 that determines a target trajectory passing through a target waypoint in accordance with the driving conditions, based on a driving command including a trajectory model and the driving conditions, and a control unit 17 that controls the robot 10, specifically, the motors M1, M2, M3, M4, M5, and M6, in accordance with the reference trajectory or the target trajectory.

As shown in FIG. 4, the control device 1 has a hardware configuration including a processor 31, a memory 32, and an interface 33. The processor 31, the memory 32, and the interface 33 are connected to each other via a bus 34. Each function of the control device 1 is realized by the processor 31 executing a program stored in the memory 32. The interface 33 connects the control device 1 to an external device, enabling communication with the external device. In detail, the control device 1 is connected to motors M1, M2, M3, M4, M5, and M6 and encoders E1, E2, E3, E4, E5, and E6 via the interface 33. The interface 33 has multiple types of interface modules as necessary.

The control device 1 shown in FIG. 4 has one processor 31 and one memory 32, but the control device 1 may have multiple processors 31 and multiple memories 32. In this case, each function of the control device 1 can be realized by the multiple processors 31 and multiple memories 32 working together.

The control device 1 having the above configuration repeats controlling the robot 10 according to a reference trajectory passing through the reference waypoint for different reference waypoints and different driving conditions, and performs a learning process to generate a trajectory model indicating a target waypoint that minimizes the trajectory error for each driving condition, and an operation process to control the robot 10 according to a target trajectory passing through the target waypoint. Minimizing the trajectory error means bringing the trajectory error closer to a minimum value or a value near a minimum value, or a local minimum value or a value near a local minimum value.

An overview of the learning process performed by the control device 1 will be described with reference to Figure 5. For example, after the robot 10 is installed, when the control device 1 starts operating for the first time or when a new operating condition is added, the control device 1 starts the learning process in Figure 5.

The reference trajectory determination unit 12 acquires a driving condition and one of a number of reference intermediate points corresponding to the driving condition from the driving condition storage unit 11, which stores the driving condition and a number of reference intermediate points in association with each other (step S11). The reference trajectory determination unit 12 determines a reference trajectory that passes through the reference intermediate point based on the driving condition and the reference intermediate point acquired in step S11 (step S12). The control unit 17 controls the motors M1, M2, M3, M4, M5, and M6 of the robot 10 according to the reference trajectory determined in step S12, and the actual trajectory acquisition unit 13 acquires the actual trajectory of the robot 10 (step S13).

In detail, the control unit 17 determines the target positions of the motors M1, M2, M3, M4, M5, and M6 to move the flange 160 according to the reference trajectory, and controls the motors M1, M2, M3, M4, M5, and M6 to move to the target positions. The actual trajectory acquisition unit 13 acquires the actual trajectory of the robot 10 based on the rotation positions of the motors M1, M2, M3, M4, M5, and M6 acquired from the encoders E1, E2, E3, E4, E5, and E6. The learning unit 14 calculates a trajectory error indicating the deviation of the trajectory acquired in step S13 from the ideal trajectory based on the driving conditions acquired in step S11 (step S14). While the calculation of the trajectory error has not been completed for all the reference intermediate points of all the driving conditions stored in the driving condition storage unit 11 (step S15; No), the above-mentioned processes from steps S11 to S14 are repeated.

When the calculation of the trajectory error for all reference intermediate points for all driving conditions stored in the driving condition storage unit 11 is completed (step S15; Yes), the learning unit 14 determines the target intermediate point that minimizes the trajectory error and generates a trajectory model that indicates the target intermediate point according to the driving condition (step S16). When the processing of step S16 is completed, the control device 1 ends the learning process.

The learning process performed by the control device 1 will be described in detail below.
As shown in Fig. 6, the drive condition storage unit 11 holds a drive condition table in which drive conditions are associated with a plurality of reference intermediate points. The drive conditions specify at least the motion start point and motion end point of the flange 160, which is the tip link. In the first embodiment, the drive conditions include the motion start point, motion end point, speed and posture of the tip link. The motion start point and motion end point indicate positions in the XYZ orthogonal coordinate system shown in Figs. 1 and 2. The speed indicates the target speed during the motion of the tip link. The posture indicates the orientation of the tip link, specifically, the angle formed between the tip surface 160a of the flange 160 and the plane to which the robot 10 is fixed.

Figure 6 shows an example of multiple reference intermediate points stored corresponding to one drive condition. For example, the drive condition included in the record in the first row of the drive condition table shown in Figure 6 indicates that the flange 160 is moved from the operation start point (0,0,0) to the operation end point (100,0,0) at a target speed V1 with the tip surface 160a of the flange 160 parallel to the plane on which the robot 10 is fixed. In this case, the ideal trajectory is represented by a line segment extending from the operation start point SP1 (0,0,0) to the operation end point EP1 (100,0,0) and located on the X-axis.

In the first embodiment, multiple reference intermediate points are defined for each drive condition. In the example of Fig. 6, multiple reference intermediate points are defined for a drive condition in which the flange 160 is moved from the motion start point (0,0,0) to the motion end point (100,0,0) at a target speed V1 with the tip surface 160a of the flange 160 parallel to the plane on which the robot 10 is fixed.

The multiple reference intermediate points are arranged three-dimensionally in the space from the motion start point to the motion end point. For example, for a drive condition that instructs movement from the motion start point SP1 to the motion end point EP1, multiple reference intermediate points RP are defined that are arranged three-dimensionally in the space S1 from the motion start point SP1 to the motion end point EP1, as shown by the dotted line in FIG. 7. For example, if the distance from the motion start point SP1 to the motion end point EP1 is 100 millimeters, the reference intermediate points RP are defined that are arranged three-dimensionally in the space S1 at intervals of 5 millimeters. The size of the space S1 may be determined according to the possible size of the trajectory error depending on the accuracy of the robot 10.

The reference trajectory determination unit 12 acquires the associated drive conditions and reference intermediate points from the drive condition table shown in FIG. 6. The reference trajectory determination unit 12 then determines a reference trajectory that passes through the acquired reference intermediate points based on the drive conditions. In detail, the reference trajectory determination unit 12 performs spline interpolation based on the operation start point, reference intermediate point, and operation end point indicated by the drive conditions to calculate a reference trajectory that passes through the reference intermediate point from the operation start point indicated by the drive conditions to the operation end point indicated by the drive conditions. The reference trajectory indicates, for example, the position of the tip link in an XYZ orthogonal coordinate system for each control cycle. The control cycle is determined, for example, according to the calculation processing capacity of the control device 1.

For example, the reference trajectory determination unit 12 obtains the drive conditions and the reference intermediate point indicated by the record in the first row of the drive condition table shown in FIG. 6. The motion start point (0,0,0) and motion end point (100,0,0) included in the drive conditions are shown as the motion start point SP1 and motion end point EP1, respectively, in FIG. 8, and the reference intermediate point (50,0,0) is shown as the reference intermediate point RP1 in FIG. 8. The reference trajectory determination unit 12 determines the reference trajectory RT1 that passes through the reference intermediate point RP1 (50,0,0) based on the drive conditions that instruct the motion from the motion start point SP1 (0,0,0) to the motion end point EP1 (100,0,0). In this case, the reference trajectory RT1 shown by the dotted line extends from the motion start point SP1 (0,0,0) to the motion end point EP1 (100,0,0) and is indicated by a line segment located on the X-axis. When the reference trajectory is determined as described above, the reference trajectory determination section 12 sends the drive conditions and the determined reference trajectory to the learning section 14 and the control section 17 .

The control unit 17 shown in FIG. 3 generates control commands for the motors M1, M2, M3, M4, M5, and M6 according to the drive conditions and the reference trajectory acquired from the reference trajectory determination unit 12, and transmits the control commands to the motors M1, M2, M3, M4, M5, and M6. In detail, the control unit 17 generates control commands for the motors M1, M2, M3, M4, M5, and M6 according to the reference position, which is the position of the tip link indicated by the reference trajectory, and transmits the control commands to the motors M1, M2, M3, M4, M5, and M6 for each control period. As a result, the motors M1, M2, M3, M4, M5, and M6 rotate according to the control commands, and the first arm 110, the second arm 120, the third arm 130, the fourth arm 140, the fifth arm 150, and the flange 160 rotate. As a result, the flange 160, which is the tip link, moves.

It is preferable that the control unit 17 obtains the rotational positions of motors M1, M2, M3, M4, M5, and M6 from encoders E1, E2, E3, E4, E5, and E6, and feedback controls motors M1, M2, M3, M4, M5, and M6.

The actual trajectory acquisition unit 13 acquires the rotational positions of the motors M1, M2, M3, M4, M5, and M6 from the encoders E1, E2, E3, E4, E5, and E6, and acquires the actual position of the tip link from the rotational positions of the motors M1, M2, M3, M4, M5, and M6. The actual trajectory acquisition unit 13 performs the above-mentioned process for each control cycle to determine an actual trajectory that indicates the actual position, which is the actual position of the tip link in the XYZ Cartesian coordinate system for each control cycle.

As an example, the actual trajectory AT1 acquired by the actual trajectory acquisition unit 13 when the motors M1, M2, M3, M4, M5, and M6 are controlled according to the drive conditions indicated by the record in the first row of the drive condition table shown in Fig. 6 and the reference trajectory RT1 based on the reference midpoint is shown by a solid line in Fig. 8. The actual trajectory acquisition unit 13 sends the actual trajectory acquired as described above to the learning unit 14.

When the learning unit 14 acquires the actual trajectory from the actual trajectory acquisition unit 13, it calculates a trajectory error indicating the deviation of the actual trajectory acquired by the actual trajectory acquisition unit 13 from the ideal trajectory when the robot 10 is controlled according to the reference trajectory. In detail, the learning unit 14 determines an ideal trajectory, which is the shortest trajectory from the operation start point to the operation end point based on the driving conditions, and determines the position of the tip link on the ideal trajectory for each control cycle. Then, the learning unit 14 calculates the distance between the position of the robot 10 for each control cycle indicated by the ideal trajectory, specifically, the ideal position which is the position of the tip link of the robot 10, and the position of the robot 10 for each control cycle indicated by the actual trajectory, specifically, the actual position which is the position of the tip link. The learning unit 14 uses the deviation of the actual position from the ideal position when this distance is maximum as the trajectory error.

For example, when the learning unit 14 acquires the drive conditions indicated by the record in the first row of the drive condition table shown in FIG. 6 and the reference trajectory corresponding to the drive conditions from the reference trajectory determination unit 12, it determines an ideal trajectory, which is the shortest trajectory from the motion start point SP1 (0,0,0) to the motion end point EP1 (100,0,0). In this case, the ideal trajectory is shown by a line segment extending from the motion start point SP1 (0,0,0) to the motion end point EP1 (100,0,0) shown in FIG. 8 and located on the X-axis. The learning unit 14 calculates the distance between the ideal position of the tip link for each control cycle on the ideal trajectory and the actual position of the tip link for each control cycle indicated by the actual trajectory AT1 acquired by the actual trajectory acquisition unit 13, and uses the deviation of the actual position from the ideal position when the distance is maximum as the trajectory error, as shown by the arrow ER1 in FIG. 8. In the example of FIG. 8, the actual position is deviated from the ideal position on the XY plane in the positive direction of the Y-axis.

The control device 1 performs the above-mentioned process for all combinations of drive conditions and reference midpoints stored in the drive condition table shown in FIG. 6. For example, after the above-mentioned process for the drive conditions and reference midpoints indicated by the record in the first row of the drive condition table shown in FIG. 6 is completed, the reference trajectory determination unit 12 acquires the drive conditions and reference midpoints indicated by the record in the second row of the drive condition table shown in FIG. 6. The drive conditions indicated by the record in the second row are the same as the drive conditions indicated by the record in the first row. The reference midpoint (50, -5, 0) indicated by the record in the second row is shown as reference midpoint RP2 in FIG. 9. The reference trajectory determination unit 12 determines a reference trajectory RT2 that passes through the reference midpoint RP2 (50, -5, 0) based on the drive conditions that instruct the operation from the operation start point SP1 (0, 0, 0) to the operation end point EP1 (100, 0, 0). For example, as shown by the dotted line in FIG. 9, the reference trajectory RT2 is a smooth curve that extends from the motion start point SP1 through the reference intermediate point RP2 to the motion end point EP1, and is shown as a curve that protrudes in the negative Y-axis direction on the XY plane.

The control unit 17 shown in Figure 3 generates control commands for motors M1, M2, M3, M4, M5, and M6 based on the driving conditions and reference trajectory RT2 obtained from the reference trajectory determination unit 12, and transmits the control commands to motors M1, M2, M3, M4, M5, and M6.

The actual trajectory acquisition unit 13 acquires the rotational positions of the motors M1, M2, M3, M4, M5, and M6 from the encoders E1, E2, E3, E4, E5, and E6, and acquires the actual position of the tip link from the rotational positions of the motors M1, M2, M3, M4, M5, and M6. As an example, the actual trajectory AT2 acquired by the actual trajectory acquisition unit 13 when the motors M1, M2, M3, M4, M5, and M6 are controlled according to the drive conditions indicated by the record in the second row of the drive condition table shown in FIG. 6 and the reference trajectory RT2 based on the reference midpoint is shown by a solid line in FIG. The actual trajectory acquisition unit 13 sends the actual trajectory acquired as described above to the learning unit 14.

The learning unit 14 calculates the distance between the ideal position of the tip link for each control cycle on the ideal trajectory and the actual position of the tip link for each control cycle indicated by the actual trajectory AT2 determined by the actual trajectory acquisition unit 13, and uses the deviation of the actual position from the ideal position when the distance between the ideal position and the actual position is at its maximum value as the trajectory error, as shown by the arrow ER2 in Figure 9.

As described above, the learning unit 14 calculates the trajectory error for each reference midpoint associated with each drive condition, and learns the correspondence between the reference midpoint and the trajectory error. The learning unit 14 then generates an error table as shown in FIG. 10 and stores it in the learned data storage unit 15. In the example of FIG. 10, for multiple reference midpoints corresponding to one drive condition, the trajectory error is stored for each reference midpoint. The trajectory error is stored as a vector component. Specifically, the trajectory error is expressed as a vector component that starts from the ideal position and indicates the deviation from the actual position.

When the calculation of the trajectory error for all the reference midpoints for all the driving conditions is completed, the learning unit 14 obtains a target midpoint that minimizes the trajectory error from the trajectory error corresponding to each of the reference midpoints for each driving condition, and generates a trajectory model that indicates the target midpoint according to the driving condition. In detail, the learning unit 14 learns the change in trajectory error according to the change in the reference midpoint for each driving condition, obtains the target midpoint at which the trajectory error is minimized, and generates a trajectory model that indicates the target midpoint according to the driving condition. In the first embodiment, the learning unit 14 generates a target midpoint table shown in FIG. 11 as a trajectory model and stores it in the learned data storage unit 15. Specifically, the learning unit 14 performs a regression analysis for each driving condition to calculate a target midpoint that minimizes the trajectory error for each driving condition. In the regression analysis, the XYZ coordinate components of the reference midpoint are used as independent variables, and the trajectory error is used as the objective function. The value of the independent variable when the trajectory error is minimized corresponds to the target midpoint.

An overview of the operation process, which is the control process of the robot 10 that is performed after the above-mentioned learning process is completed, will be described with reference to Fig. 12. For example, after the learning process is completed, the control device 1 starts the operation process in Fig. 12 when a drive command including the drive conditions of the robot 10 is input by operation from an operation unit (not shown).

When the target trajectory determination unit 16 acquires a drive command including a drive condition, it acquires a target intermediate point according to the drive condition indicated by the drive command from the learned data storage unit 15 (step S21). Then, the target trajectory determination unit 16 determines a target trajectory passing through the target intermediate point acquired in step S21 based on the drive condition (step S22). The control unit 17 controls the motors M1, M2, M3, M4, M5, and M6 of the robot 10 according to the target trajectory determined in step S22 (step S23).

The operation process performed by the control device 1 will be described in detail below.
When the target trajectory determination unit 16 acquires a drive command, it acquires a target intermediate point according to the drive condition indicated by the drive command from a target intermediate point table stored in the learned data storage unit 15. Then, the target trajectory determination unit 16 determines a target trajectory indicating a trajectory from the motion start point indicated by the drive condition through the target intermediate point to the motion end point indicated by the drive condition. In detail, the target trajectory determination unit 16 performs spline interpolation based on the motion start point, target intermediate point, and motion end point indicated by the drive condition to determine a target trajectory from the motion start point indicated by the drive condition through the target intermediate point to the motion end point indicated by the drive condition. The target trajectory indicates a target position of the tip link for each control cycle.

For example, the target trajectory determination unit 16 acquires a drive command indicating the drive conditions for moving the flange 160 from the motion start point (0,0,0) to the motion end point (100,0,0) at a target speed V1 while the tip surface 160a of the flange 160 is parallel to the plane on which the robot 10 is fixed. The target trajectory determination unit 16 acquires the target intermediate point (50,-2.5,0) indicated by the record in the first row of the target intermediate point table shown in FIG. 11, which corresponds to the acquired drive conditions. This target intermediate point (50,-2.5,0) is shown as the target intermediate point TP1 in FIG. 13. The target trajectory determination unit 16 then determines the target trajectory TT1 that runs from the motion start point SP1 (0,0,0) through the target intermediate point TP1 (50,-2.5,0) to the motion end point EP1 (100,0,0). 13, the target trajectory TT1 is a smooth curve extending from the motion start point SP1 through the target intermediate point TP1 to the motion end point EP1, and is indicated by a curve protruding in the negative direction of the Y axis on the XY plane. When the target trajectory is determined as described above, the target trajectory determination unit 16 sends the drive conditions and the determined target trajectory to the control unit 17.

The control unit 17 shown in FIG. 3 generates control commands for the motors M1, M2, M3, M4, M5, and M6 according to the target trajectory acquired from the target trajectory determination unit 16, and transmits the control commands to the motors M1, M2, M3, M4, M5, and M6. As a result, the motors M1, M2, M3, M4, M5, and M6 rotate according to the control commands, and the first arm 110, the second arm 120, the third arm 130, the fourth arm 140, the fifth arm 150, and the flange 160 rotate. This causes the tip link to move. By controlling the motors M1, M2, M3, M4, M5, and M6 according to the target trajectory that minimizes the trajectory error, the actual trajectory AT1' of the tip link approaches the ideal trajectory, as shown in FIG. 13.

As described above, the control device 1 according to the first embodiment learns the correspondence between the reference waypoint and the trajectory error, finds the target waypoint that minimizes the trajectory error, generates a trajectory model that indicates the target waypoint according to the drive conditions, and controls the robot 10 according to the target trajectory that passes through the target waypoint. As a result, the actual trajectory approaches the ideal trajectory, and the control accuracy of the robot 10 by the control device 1 is improved.

(Embodiment 2)
When a workpiece is machined using a machining tool attached to a robot, for example, a machining tool attached to a tip link, the robot receives a force from the workpiece during machining. The robot is affected by a reaction force, which is a force that the robot receives from the workpiece, and the trajectory of the tip link may deviate from an ideal trajectory. In the second embodiment, a robot system 101 including a robot 30 and a control device 2 that controls the robot 30 in accordance with the reaction force received by the robot 30 during machining will be described, focusing on the differences from the robot system 100 according to the first embodiment.

The robot 30 shown in Fig. 14 includes a force sensor 190 in addition to the configuration of the robot 10 according to embodiment 1. A machining tool 200 is attached to the flange 160. By controlling the robot 30 to which the machining tool 200 is attached, it becomes possible to perform processing such as deburring the workpiece, which is the object to be processed, and attaching parts.

The force sensor 190 outputs a sensor signal corresponding to the reaction force, which is the force that the flange 160 receives when the workpiece is machined by the machining tool 200.

The control device 2 provided in the robot system 101 shown in Fig. 15 includes a reaction force acquisition unit 19 that acquires a sensor signal from a force sensor 190 provided in the robot 30 and calculates the reaction force received by the robot 30 from the sensor signal, and a learning unit 18 that learns the trajectory error and the reaction force. The reaction force acquisition unit 19 detects the value of the sensor signal acquired from the force sensor 190 for each control cycle, calculates the reaction force received by the robot 30 from the value detected for each control cycle, and sends the calculated reaction force to the learning unit 18. The learning unit 18 acquires the reaction force calculated by the reaction force acquisition unit 19 and determines the reaction force corresponding to the trajectory error, i.e., the reaction force when the distance between the ideal position of the tip link for each control cycle and the actual position of the tip link for each control cycle indicated by the actual trajectory is maximum.

The hardware configuration of the control device 2 is the same as that of the control device 1 shown in Fig. 4. The control device 2 is connected to the motors M1, M2, M3, M4, M5, and M6, the encoders E1, E2, E3, E4, E5, and E6, and the force sensor 190 via the interface 33.

The learning process performed by the control device 2 having the above configuration will be described with reference to FIG. 16. For example, when the robot 30 performs machining for the first time after the robot 30 is installed, the control device 2 starts the learning process of FIG. 16. The processes from step S11 to S12 are the same as the processes performed by the control device 1 shown in FIG. 5. The control unit 17 controls the motors M1, M2, M3, M4, M5, and M6 of the robot 10 according to the reference trajectory determined in step S12, the actual trajectory acquisition unit 13 acquires the actual trajectory of the robot 10, and the reaction force acquisition unit 19 calculates the reaction force (step S17). In detail, in step S17, the motors M1, M2, M3, M4, M5, and M6 are controlled based on the reference trajectory, and machining is performed by the robot 30. While motors M1, M2, M3, M4, M5, and M6 are being controlled based on the reference trajectory, an actual trajectory acquisition unit 13 acquires the actual trajectory of the tip link of the robot 30, and a reaction force acquisition unit 19 calculates the reaction force from the sensor signal output by a force sensor 190.

The learning unit 18 calculates a trajectory error indicating the deviation of the actual trajectory obtained in step S17 from the ideal trajectory based on the driving conditions obtained in step S11, and matches the trajectory error with the reaction force when the distance between the ideal position and the actual position is maximum (step S18).

As in the first embodiment, while the correspondence between the trajectory error and the reaction force has not been completed for all reference midpoints of all driving conditions stored in the driving condition storage unit 11 (step S19; No), the above-mentioned steps S11, S12, S17, and S18 are repeated. In other words, the learning unit 18 calculates the trajectory error for each of the reference midpoints associated with each driving condition to learn the correspondence between the driving condition, the reference midpoint, the trajectory error, and the reaction force. Then, the learning unit 18 generates an error table shown in FIG. 17 and stores it in the learned data storage unit 15. In the example of FIG. 17, for multiple reference midpoints corresponding to one driving condition, the trajectory error and the reaction force are stored for each reference midpoint. The reaction force is represented by a force Fx in the X-axis direction, a force Fy in the Y-axis direction, and a force Fz in the Z-axis direction. In the example of FIG. 17, the reaction force represented by (Fx, Fy, Fz) is stored in the error table.

When the correspondence between the trajectory error and the reaction force is completed for all reference midpoints under all driving conditions (step S19; Yes), the learning unit 18 obtains a target midpoint that minimizes the trajectory error and generates a trajectory model that indicates the target midpoint according to the driving condition (step S20). In detail, the learning unit 18 calculates the target midpoint that minimizes the trajectory error for each driving condition using a multiple regression model for each driving condition. In the multiple regression model, the XYZ coordinate components of the reference midpoint and the reaction force are used as independent variables, and the trajectory error is used as the objective function. The reference midpoint when the trajectory error is minimized corresponds to the target midpoint. The learning unit 18 generates the target midpoint table shown in FIG. 11, as in the first embodiment. When the processing of step S20 is completed, the control device 2 ends the learning process.

The operation process performed by the control device 2 after the above-mentioned learning process is completed is the same as that in embodiment 1. The control device 2 performs the operation process shown in FIG. 12 to control the robot 30.

As described above, the control device 2 according to the second embodiment determines a target midpoint that minimizes the trajectory error from the trajectory error and reaction force corresponding to each reference midpoint for each drive condition, generates a trajectory model indicating the target midpoint according to the drive condition, and controls the robot 30 according to the target trajectory that passes through the target midpoint. Since the target midpoint is determined according to the reaction force, which is a force that the robot 30 receives during machining, the control device 2 achieves high control accuracy of the robot 30.

(Embodiment 3)
In the first and second embodiments, all of the reference intermediate points corresponding to each drive condition are stored in advance in the drive condition table, but the reference intermediate points may be estimated. In the third embodiment, a robot system 102 including a robot 10 and a control device 3 that estimates at least some of the reference intermediate points and performs a learning process will be described, focusing on the differences from the robot system 100 according to the first embodiment.

18 includes a control device 3 that includes the configuration of the control device 1 according to embodiment 1, and further includes a midpoint estimation unit 20. The midpoint estimation unit 20 estimates a reference midpoint when the drive condition storage unit 11 does not store multiple reference midpoints corresponding to the drive conditions, or when the number of multiple reference midpoints stored in the drive condition storage unit 11 corresponding to the drive conditions is insufficient.

In detail, the midpoint estimation unit 20 estimates a plurality of reference midpoints corresponding to other driving conditions from the target midpoints according to the driving conditions indicated by the trajectory model generated by the learning unit 14. For example, the midpoint estimation unit 20 estimates a plurality of reference midpoints for a driving condition for which a corresponding plurality of reference midpoints are not stored in the driving condition storage unit 11 from the driving conditions and the target midpoints corresponding to the driving conditions stored in the learned data storage unit 15. Then, the midpoint estimation unit 20 stores the driving conditions and the estimated plurality of reference midpoints in the driving condition storage unit 11 in association with each other.

The hardware configuration of the control device 3 is similar to the hardware configuration of the control device 1 shown in Figure 4.

The learning process performed by the control device 3 having the above configuration will be described with reference to Figure 19. As with the control device 1, for example, when the control device 3 starts operating for the first time after the robot 10 is installed, the control device 3 starts the learning process of Figure 19. The processes from steps S11 to S16 are similar to the processes performed by the control device 1 shown in Figure 5. In other words, based on the drive conditions and multiple reference waypoints previously stored in the drive condition memory unit 11, the trajectory error is calculated and a trajectory model indicating the target waypoint is generated as in embodiment 1.

When the process of step S16 is completed, the control device 3 performs the process from step S31 onward. The midpoint estimation unit 20 determines whether or not it is necessary to estimate the reference midpoint corresponding to the driving condition stored in the driving condition storage unit 11 (step S31). In detail, the midpoint estimation unit 20 determines whether or not it is necessary to estimate the reference midpoint based on whether or not the number of multiple reference midpoints associated with the driving conditions stored in the driving condition storage unit 11 is equal to or greater than the number required for generating the trajectory model in the learning unit 14. As an example, the midpoint estimation unit 20 determines that it is necessary to estimate the reference midpoint when the reference midpoint is not associated with the driving condition stored in the driving condition storage unit 11. As another example, the midpoint estimation unit 20 determines that it is necessary to estimate the reference midpoint when the number of reference midpoints corresponding to each driving condition stored in the driving condition storage unit 11 is less than 50.

For example, in the drive condition table shown in FIG. 20, the reference midpoint is not associated with the drive condition in which the flange 160 is driven at a target speed V1 from the motion start point (0,0,0) to the motion end point (-100,0,0) with the tip surface 160a of the flange 160 parallel to the plane on which the robot 10 is fixed. Similarly, the reference midpoint is not associated with the drive condition in which the flange 160 is driven at a target speed V1 from the motion start point (0,0,0) to the motion end point (0,100,0) with the tip surface 160a of the flange 160 parallel to the plane on which the robot 10 is fixed. For this reason, the midpoint estimation unit 20 determines that estimation of the reference midpoint is necessary for these drive conditions in the process of step S31 in FIG. 19 (step S31; Yes).

If estimation of the reference midpoint is necessary (step S31; Yes), the midpoint estimation unit 20 estimates the reference midpoint from the target midpoint according to the driving conditions indicated by the trajectory model (step S32).

In detail, the midpoint estimation unit 20 estimates a plurality of reference midpoints for a driving condition that requires estimation of a reference midpoint from the driving conditions and the target midpoints corresponding to the driving conditions stored in the learned data storage unit 15. For example, the target midpoint for each driving condition is obtained from the target midpoint table shown in FIG. 11, and a regression analysis is performed for each driving condition that requires estimation of a reference midpoint to estimate a target midpoint for a driving condition that does not correspond to a reference midpoint. Then, the midpoint estimation unit 20 estimates a plurality of reference midpoints arranged three-dimensionally as shown in FIG. 7 in a space centered on the estimated target midpoint. It is preferable that the space centered on the estimated target midpoint is narrower than the space in which the plurality of reference midpoints stored in advance in the driving condition storage unit 11 are located, for example, the space S1 in FIG. 7. This shortens the time required to calculate the trajectory error, and as a result, it becomes possible to efficiently obtain the target midpoint.

19, when the midpoint estimation unit 20 estimates multiple reference midpoints for each driving condition that requires estimation of the reference midpoint as described above, the midpoint estimation unit 20 associates each driving condition with the estimated multiple reference midpoints and stores them in the driving condition storage unit 11 (step S33). As a result, as shown in FIG. 21, a record that associates the driving condition with the estimated reference midpoint is added to the driving condition table stored in the driving condition storage unit 11.

When the processing of step S33 in FIG. 19 is completed, the control device 3 repeats the above-described processing from the processing of step S11. In step S33, a record associating the driving condition with the estimated reference midpoint has been added to the driving condition table stored in the driving condition storage unit 11, so that a learning process is performed for the driving condition and the estimated reference midpoint in the same manner as in embodiment 1. There is no need to repeat the learning process for driving conditions that have already been the subject of the learning process and for which a record associated with the estimated reference midpoint has not been added.

In step S33, a record associating the driving condition with the estimated reference midpoint is added to the driving condition table stored in the driving condition storage unit 11, and the learning process described above is performed. Then, the midpoint estimation unit 20 determines that it is not necessary to estimate the reference midpoint (step S31; No). If it is not necessary to estimate the reference midpoint (step S31; No), the control device 3 ends the learning process.
The operational process performed by the control device 3 after the above-mentioned learning process is completed is the same as that in the first embodiment.

As described above, the control device 3 according to the third embodiment estimates reference midpoints corresponding to other driving conditions from target midpoints according to driving conditions indicated by the trajectory model. The control device 3 estimates the reference midpoints based on the trajectory model, calculates the trajectory error based on the estimated reference midpoints, and determines the target midpoint. This results in higher learning efficiency for the control device 3 compared to a case in which trajectory errors are calculated for all predetermined reference midpoints. By reducing the number of reference midpoints to be estimated below the number of reference midpoints previously stored in the driving condition memory unit 11, it becomes possible for the control device 3 to learn more efficiently.

(Embodiment 4)
In the first and second embodiments, all reference midpoints corresponding to each drive condition are stored in advance in the drive condition table, but the reference midpoints may be estimated. Furthermore, the drive conditions may also be estimated. In the fourth embodiment, a robot system 103 including a robot 10 and a control device 4 that estimates at least some drive conditions and at least some reference midpoints and performs a learning process will be described, focusing on the differences from the robot system 100 according to the first embodiment.

22 includes a control device 4 that is provided in the robot system 103, and includes the configuration of the control device 1 according to the first embodiment, and further includes a drive condition estimation unit 21. The drive condition estimation unit 21 estimates a reference midpoint when the number of reference midpoints stored in the drive condition storage unit 11 corresponding to the drive condition is insufficient, and estimates a drive condition and a plurality of reference midpoints corresponding to the drive condition when the number of drive conditions stored in the drive condition storage unit 11 is insufficient.

In detail, the driving condition estimation unit 21 estimates a new reference intermediate point associated with the driving condition stored in the driving condition storage unit 11, as well as a new driving condition not stored in the driving condition storage unit 11 and a plurality of reference intermediate points associated with the new driving condition, from the trajectory error calculated by the learning unit 14 according to the driving condition and the reference intermediate point stored in the driving condition storage unit 11. The driving condition estimation unit 21 then associates the driving condition stored in the driving condition storage unit 11 with the new reference intermediate point estimated based on the driving condition, and stores them in the driving condition storage unit 11. The driving condition estimation unit 21 further associates the estimated driving condition with each of the plurality of estimated reference intermediate points, and stores them in the driving condition storage unit 11.

The hardware configuration of the control device 4 is similar to the hardware configuration of the control device 1 shown in Figure 4.

The learning process performed by the control device 4 having the above configuration will be described with reference to Figure 23. As with the control device 1, for example, when the control device 4 starts operating for the first time after the robot 10 is installed, the control device 4 starts the learning process of Figure 23. The processes from steps S11 to S15 are similar to the processes performed by the control device 1 shown in Figure 5. In other words, the trajectory error calculation process is performed similarly to embodiment 1 based on the drive conditions and reference midpoints previously stored in the drive condition memory unit 11.

When the processing of step S15 is completed, the control device 4 performs the processing from step S41 onward. The driving condition estimation unit 21 determines whether or not it is necessary to estimate the reference intermediate point corresponding to the driving condition stored in the driving condition storage unit 11 (step S41). In detail, the driving condition estimation unit 21 determines whether or not it is necessary to estimate the reference intermediate point based on whether or not the number of reference intermediate points corresponding to the driving condition stored in the driving condition storage unit 11 is equal to or greater than the number required for generating the trajectory model in the learning unit 14. For example, if the number of reference intermediate points corresponding to each driving condition stored in the driving condition storage unit 11 is less than 50, the driving condition estimation unit 21 determines that it is necessary to estimate the reference intermediate point.

For example, if only the first row of records shown in FIG. 6 is stored in the drive condition table stored in the drive condition memory unit 11 at the time the control device 4 starts the learning process, the drive condition estimation unit 21 determines in the processing of step S41 in FIG. 23 that it is necessary to estimate the reference midpoint for the drive condition for driving the flange 160 from the operation start point (0,0,0) to the operation end point (100,0,0) at the target speed V1 with the tip surface 160a of the flange 160 parallel to the plane to which the robot 10 is fixed (step S41; Yes).

If it is necessary to estimate the reference midpoint (step S41; Yes), the drive condition estimation unit 21 estimates the reference midpoint by a regression algorithm based on the error table stored in the learned data storage unit 15 (step S42). For example, the drive condition estimation unit 21 estimates the reference midpoint by a Gaussian process regression algorithm for a drive condition in which the flange 160 is driven from the operation start point (0,0,0) to the operation end point (100,0,0) at a target speed V1 with the tip surface 160a of the flange 160 parallel to the plane on which the robot 10 is fixed. The objective function of the Gaussian process regression algorithm is the trajectory error.

The driving condition estimation unit 21 stores the reference midpoint estimated in step S42 in the driving condition storage unit 11 in association with the driving condition (step S43). As a result, for example, the records from the second row onward shown in FIG. 6 are generated. When the processing of step S43 is completed, the control device 4 repeats the above-mentioned processing from step S11. In step S43, a record associating the driving condition with the estimated reference midpoint has been added to the driving condition table stored in the driving condition storage unit 11, so that the trajectory error calculation processing is performed for the driving condition and the estimated reference midpoint in the same manner as in embodiment 1. There is no need to repeat the learning processing for the driving conditions that have already been the subject of the calculation processing.

When a sufficient number of reference midpoints are associated with each driving condition and stored in the driving condition storage unit 11 as a result of repeating the above-mentioned process, the driving condition estimation unit 21 determines that estimation of the reference midpoint is unnecessary (step S41; No). If estimation of the reference midpoint is unnecessary (step S41; No), the driving condition estimation unit 21 determines whether or not estimation of the driving condition is necessary (step S44).

In detail, the drive condition estimation unit 21 determines whether or not it is necessary to estimate the drive conditions based on whether or not the drive conditions stored in the drive condition storage unit 11 are sufficient for the types of control patterns of the robot 10. For example, the drive condition estimation unit 21 determines that it is necessary to estimate the drive conditions if the types of drive conditions stored in the drive condition storage unit 11 are fewer than the types of predetermined control patterns of the robot 10. The drive condition estimation unit 21 only needs to store information about the types of control patterns of the robot 10 in a memory (not shown).

For example, when the processing of step S43 is completed, if the only driving condition stored in the driving condition memory unit 11 is a driving condition in which the tip surface 160a of the flange 160 is parallel to the plane on which the robot 10 is fixed and the flange 160 is driven from the operation start point (0,0,0) to the operation end point (100,0,0) at the target speed V1, the driving condition estimation unit 21 determines that estimation of the driving conditions is necessary (step S44; Yes).

If it is necessary to estimate the driving conditions (step S44; Yes), the driving condition estimation unit 21 estimates other driving conditions and multiple reference intermediate points corresponding to the other driving conditions by a regression algorithm based on the error table stored in the learned data storage unit 15 (step S45). The other driving conditions are driving conditions different from the driving conditions stored in the driving condition storage unit 11. For example, the driving condition estimation unit 21 estimates driving conditions that drive the flange 160 from the operation start point (0,0,0) to the operation end point (-100,0,0) at a target speed V1 by a Gaussian process regression algorithm with the tip surface 160a of the flange 160 parallel to the plane on which the robot 10 is fixed, and estimates multiple reference intermediate points corresponding to the estimated driving conditions. The estimation of multiple reference intermediate points corresponding to the estimated driving conditions is similar to the processing of step S42.

The drive condition estimation unit 21 stores the drive conditions estimated in step S45 in association with the multiple reference intermediate points in the drive condition storage unit 11 (step S46). As a result, for example, as shown in FIG. 21, multiple records are generated for drive conditions in which the tip surface 160a of the flange 160 is parallel to the plane on which the robot 10 is fixed, and the flange 160 is driven at a target speed V1 from the operation start point (0,0,0) to the operation end point (-100,0,0). When the process of step S46 is completed, the control device 4 repeats the above-mentioned process from step S11. In step S46, a record that associates the estimated drive conditions with each of the estimated multiple reference intermediate points is added to the drive condition table stored in the drive condition storage unit 11, so that the trajectory error calculation process is performed for the estimated drive conditions and the estimated multiple reference intermediate points in the same manner as in embodiment 1. There is no need to repeat the calculation process for drive conditions that have already been subject to the calculation process.

When a sufficient number of driving conditions and a sufficient number of reference intermediate points corresponding to each driving condition are stored in the driving condition storage unit 11 as a result of repeating the above-mentioned process, the driving condition estimation unit 21 determines that estimation of the reference intermediate point is not necessary (step S41; No) and that estimation of the driving condition is not necessary (step S44; No). If estimation of the reference intermediate point is not necessary (step S41; No) and estimation of the driving condition is not necessary (step S44; No), the learning unit 14 generates a trajectory model indicating the target intermediate point that minimizes the trajectory error for each driving condition, as in the first embodiment (step S16). When the process of step S16 is completed, the control device 4 ends the learning process.
The operational process performed by the control device 4 after the above-mentioned learning process is completed is the same as that in the first embodiment.

As described above, the control device 4 according to the fourth embodiment estimates a reference midpoint corresponding to the drive condition stored in the drive condition storage unit 11, calculates a trajectory error based on the estimated reference midpoint, and obtains a target midpoint. Furthermore, the control device 4 estimates a drive condition not stored in the drive condition storage unit 11, estimates a plurality of reference midpoints corresponding to the estimated drive condition, calculates a trajectory error based on the estimated drive condition and a plurality of reference midpoints, and obtains a target midpoint. Therefore, even if a large number of drive conditions and a large number of reference midpoints are not stored in the drive condition storage unit 11 in advance, it is possible to calculate a trajectory error for various drive conditions for controlling the robot 10 and bring the actual trajectory closer to the ideal trajectory.

(Embodiment 5)
The method of acquiring the trajectory of the robot 10 is not limited to the above example. A robot system 104 including the robot 10 and the control device 5 that acquires the trajectory of the robot 10 by a method different from that of the first embodiment will be described, focusing on the differences from the robot system 100 according to the first embodiment.

The control device 5 shown in Fig. 24 includes an actual trajectory acquisition unit 22 that acquires the trajectory of the robot 10 regardless of the rotational positions of the motors M1, M2, M3, M4, M5, and M6. The actual trajectory acquisition unit 22 has a measuring device, for example a three-dimensional measuring device, that measures the trajectory of the robot 10, and acquires the trajectory of the robot 10 according to the measurement values of the measuring device. The three-dimensional measuring device acquires the position of the tip link of the robot 10 by irradiating a laser beam within the movement range of the tip link of the robot 10 and receiving the laser beam reflected by the tip link of the robot 10.

In the fifth embodiment, the actual trajectory acquisition unit 22 performs the above-mentioned process for each control cycle to calculate an actual trajectory indicating the actual position, which is the actual position of the tip link in the XYZ Cartesian coordinate system for each control cycle. Then, the actual trajectory acquisition unit 22 sends the actual trajectory to the learning unit 14.

The hardware configuration of the control device 5 is similar to that of the control device 1 shown in FIG.
The learning process performed by the control device 5 having the above configuration is the same as that of the control device 1, except for the method of acquiring the trajectory of the end link. The operation process performed by the control device 5 is the same as that of the control device 1.

As described above, the control device 5 of embodiment 5 acquires the trajectory of the robot 10 regardless of the rotational positions of the motors M1, M2, M3, M4, M5, and M6, so that the acquired actual trajectory does not include errors caused by the mechanical rigidity of the robot 10, improving the accuracy of the actual trajectory.

The present disclosure is not limited to the above-described embodiments. Any combination of two or more of the above-described embodiments is possible. As an example, the control device 3-5 may calculate a reaction force from the measurement value of the force sensor 190, similar to the control device 2, and determine a target intermediate point according to the trajectory error and the reaction force corresponding to the trajectory error.

The driving conditions are not limited to the above examples. As an example, the driving conditions may further include information about the workpiece. Specifically, the driving conditions may include the motion start point, motion end point, speed, and posture of the tip link, as well as the weight of the workpiece. The learning unit 14 may acquire information about the weight of the workpiece, which is the workpiece, from an external device, for example, a weight sensor that measures the weight of the workpiece. In detail, the control device 1-5 acquires measurements from the weight sensor, controls the robots 10 and 30 according to the driving conditions stored in the driving condition storage unit 11, and repeatedly calculates the trajectory error according to the driving conditions and the measured values of the weight sensor to generate a trajectory model. As a result, the learning unit 14 generates a target intermediate point table that associates the driving conditions, including the motion start point, motion end point, speed, and posture of the tip link, as well as the weight of the workpiece, with the target intermediate point, as shown in FIG. 25.

During operation, when the target trajectory determination unit 16 receives a drive command, it obtains from the learned data storage unit 15 the motion start point, motion end point, speed, and posture of the tip link indicated by the drive command, and a target intermediate point corresponding to the measured value of the weight sensor. The target trajectory determination unit 16 then determines a target trajectory passing through the target intermediate point based on the motion start point, motion end point, speed, and posture of the tip link indicated by the drive command. The control accuracy of the robots 10, 30 is improved by controlling the robots 10, 30 based on the target trajectory corresponding to the weight of the workpiece.

Information about the workpiece is not limited to the weight of the workpiece, but may include at least one of the weight, shape, and dimensions of the workpiece.

As another example, the drive conditions may include the motion start point, motion end point, time required for the tip link to move from the motion start point to the motion end point, and posture.
In the above embodiment, the drive condition specifies linear movement on a plane, but the drive condition may specify movement in three dimensions, curvilinear movement, or the like.

In the above embodiment, one reference midpoint is stored in one record of the drive condition table stored in the drive condition storage unit 11, but multiple reference midpoints may be stored in one record of the drive condition table. In this case, the reference trajectory determination unit 12 may determine a reference trajectory that passes through the multiple reference midpoints stored in the same record.

The drive condition storage unit 11 and the learned data storage unit 15 may be provided outside the control device 1-5. As an example, the drive condition storage unit 11 and the learned data storage unit 15 may be realized as functions of a storage device on a network.

The method of determining the reference trajectory, ideal trajectory, and target trajectory is not limited to the above example. In detail, the reference trajectory RT2 determined by the reference trajectory determination unit 12 in the first embodiment is a smooth curve extending from the motion start point SP1 through the reference intermediate point RP2 to the motion end point EP1, and is shown as a curve protruding in the negative Y-axis direction on the XY plane, but the reference trajectory RT2 is not limited to such a curve and may be shown as a curve protruding in multiple directions. As an example, the reference trajectory RT2 may be shown as a curve protruding in the negative Y-axis direction from the motion start point SP1 to the motion end point EP1, and protruding in the positive Y-axis direction on the XY plane near the motion start point SP1 and the motion end point EP1.

As another example, the reference trajectory determination unit 12 may determine the reference trajectory by linearly interpolating the motion start point and the reference intermediate point indicated by the drive conditions, and linearly interpolating the reference intermediate point and the motion end point indicated by the drive conditions. Even when the reference trajectory is determined by linear interpolation, the control unit 17 controls the motors M1, M2, M3, M4, M5, and M6 to rotate smoothly, so that the tip link moves smoothly and the actual trajectory becomes a smooth curve.

As another example, the reference trajectory determination unit 12 may determine the reference trajectory using a table (not shown) in which the operation start point, reference intermediate point, and operation end point are associated with the reference position of the tip link for each control cycle.

The ideal trajectory and target trajectory can also be determined using a method similar to that used to determine the reference trajectory described above.

In the first embodiment, the ideal trajectory is a straight line because the robot 10 is moved linearly, but the ideal trajectory need only be determined according to how the robot 10 is moved, and is not limited to a straight line. Specifically, the ideal trajectory may be an arc, a curve, or a combination of a straight line, an arc, and a curve.

In the first embodiment, the target trajectory TT1 determined by the target trajectory determination unit 16 is a smooth curve extending from the motion start point SP1 through the target intermediate point TP1 to the motion end point EP1, and is shown as a curve protruding in the negative Y-axis direction on the XY plane, but the target trajectory TT1 is not limited to such a curve. As an example, the target trajectory TT1 may be shown as a curve protruding in the negative Y-axis direction from the motion start point SP1 to the motion end point EP1, and protruding in the positive Y-axis direction on the XY plane near the motion start point SP1 and the motion end point EP1.

As another example, the target trajectory determination unit 16 may determine the target trajectory by linearly interpolating the motion start point and the target intermediate point indicated by the drive conditions, and linearly interpolating the target intermediate point and the motion end point indicated by the drive conditions. Even when the target trajectory is determined by linear interpolation, the control unit 17 controls the motors M1, M2, M3, M4, M5, and M6 to rotate smoothly, so that the tip link moves smoothly and the actual trajectory becomes a smooth curve.

The method of determining the target intermediate point is not limited to the above example. As an example, the learning units 14 and 18 may determine the target intermediate point using linear regression, curvilinear regression, polynomial regression, multidimensional function fitting, or the like. For example, the learning units 14 and 18 may determine the target intermediate point that minimizes the trajectory error by performing multidimensional function fitting with the XYZ coordinate components of the reference intermediate point as independent variables and the trajectory error as an objective function.

The method of generating the trajectory model is not limited to the above example. As an example, the learning units 14 and 18 may perform supervised learning according to a neural network model to generate a trajectory model. Supervised learning means that a large amount of data sets of input and results are provided to a learning device, so that the learning device learns features in the large amount of data sets and generates a model that estimates a result from the input. For example, the learning units 14 and 18 generate a trajectory model, which is a neural network model, using the operation start point and operation end point indicated by the driving conditions as input data and the target intermediate point as result data. Using the generated trajectory model, it is possible to obtain a target intermediate point that minimizes the trajectory error according to the driving conditions.

In machine learning, it is known that overlearning reduces the estimation accuracy, so it is preferable to suppress overlearning regardless of which learning model is used. Methods for suppressing overlearning include, for example, normalizing the data to be learned and reducing the number of parameters in the learning model. As an example, when performing multidimensional function fitting, it is preferable to limit the number of dimensions of the function, for example, to 5th degree or less. As another example, in a neural network model, it is preferable to limit the number of intermediate layers, for example, to 10 layers or less. In addition, in a neural network model, the output of a specific layer of the learning model may be randomly set to 0 during learning. Furthermore, overlearning may be suppressed by combining the above-mentioned methods.

The learning unit 14, 18 may perform the learning process according to the control of the robot 10, 30 performed in the operation process immediately after. As an example, the operation of the learning unit 14 in a case where it is predetermined to move the robot 10, 30 from the operation start point (0,0,0) to the operation end point (100,0,0) in the operation process will be described. In this case, the learning unit 14 may learn the trajectory error and generate a trajectory model based on the driving conditions including the operation start point (0,0,0) and the operation end point (100,0,0) and a plurality of reference intermediate points associated with the driving conditions. By performing the learning process based on the driving conditions according to the control of the robot 10, 30 performed in the operation process and a plurality of reference intermediate points associated with the driving conditions, the learning efficiency is improved.

As another example, the operation of the learning unit 14 will be described using an example in which the robot 30 is moved to process a workpiece weighing 4 kg while the tip surface 160a of the flange 160 is positioned horizontally from the operation start point (0,0,0) to the operation end point (100,0,0) in the operational processing. In this case, the learning unit 14 learns the trajectory error and generates a trajectory model based on the drive condition that includes the operation start point (0,0,0) and the operation end point (100,0,0), shows a posture of 0 degrees, and shows a weight of 4 kg, and multiple reference intermediate points associated with the drive condition.

In this case, the learning unit 14, 18 may generate a trajectory model by calculating a trajectory error based on not only the driving conditions that completely match the driving conditions in the operation process, but also the driving conditions that partially match the driving conditions in the operation process. The operation of the learning unit 14 will be described using an example in which the robot 30 is moved to process a workpiece weighing 4 kg while the tip surface 160a of the flange 160 is positioned horizontally from the operation start point (0,0,0) to the operation end point (100,0,0) in the operation process as described above. The learning unit 14 may generate a trajectory model by calculating a trajectory error based on a driving condition that indicates a posture of 0 degrees and a weight of 4 kg and a plurality of reference intermediate points associated with the driving condition. Alternatively, the learning unit 14 may generate a trajectory model by calculating a trajectory error based on a driving condition that includes the operation start point (0,0,0) and the operation end point (100,0,0) and indicates a weight of 4 kg and a plurality of reference intermediate points associated with the driving condition.

The learning unit 14, 18 may calculate the distance between the ideal position and the actual position of the tip link for each control cycle, and use the deviation of the actual position from the ideal position when the distance is at its maximum value as the trajectory error. In this case, the learning unit 14, 18 may store the deviation of the actual position from the ideal position when the distance is at its maximum value as the trajectory error in an error table, as shown in FIG. 26. Then, the learning unit 14, 18 may determine the target intermediate point based on the trajectory error.

The learning unit 14, 18 may store in the error table the actual position of the tip link corresponding to the trajectory error. For example, in the case where the learning unit 14, 18 uses the deviation of the actual position from the ideal position when the distance between the ideal position and the actual position of the tip link for each control cycle becomes a maximum value as the trajectory error, the learning unit 14, 18 may store in the error table the actual position of the tip link when the distance becomes a maximum value, as shown in FIG. 27. Then, the learning unit 14, 18 may determine the target intermediate point based on the trajectory error and the actual position of the tip link corresponding to the trajectory error.

As one example, as shown in Fig. 28, the learning unit 18 may store in the error table a reaction force calculated from a sensor signal output by the force sensor 190 when the distance is at a maximum value. The learning unit 18 may then determine the target intermediate point based on the trajectory error and the reaction force corresponding to the trajectory error. As another example, the learning unit 18 may store in the error table a maximum value of the reaction force and an average value of the reaction force.

The learning units 14 and 18 may perform the learning process again in parallel with the operation process performed after the learning process is completed. In detail, the learning units 14 and 18 calculate a trajectory error indicating the deviation of the actual trajectory acquired by the actual trajectory acquisition unit 13 from the ideal trajectory according to the drive conditions indicated by the drive command acquired by the target trajectory determination unit 16 during the operation process, and obtain a target intermediate point. By performing the learning process in parallel with the operation process, the accuracy of the target intermediate point that minimizes the trajectory error is improved, and as a result, the control accuracy of the robot 10 and 30 is improved.

The learning units 14 and 18 may determine the target intermediate point using the absolute value, average value, and median value of the deviation of the actual position of the tip link from the ideal position as the trajectory error.

When determining the target intermediate point, the learning unit 14, 18 may estimate the trajectory error when the robot 10, 30 is controlled according to the target trajectory passing through the target intermediate point, and generate a trajectory model indicating the estimated values of the target intermediate point and the trajectory error. In this case, the midpoint estimation unit 20 may estimate a plurality of reference intermediate points arranged at intervals according to the estimated value of the trajectory error indicated by the trajectory model. For example, the midpoint estimation unit 20 may widen the intervals between the plurality of reference intermediate points to be estimated as the estimated value of the trajectory error becomes smaller. As an example, the midpoint estimation unit 20 may estimate a plurality of reference intermediate points arranged in three dimensions at intervals of twice the length of the maximum value of the trajectory error when the robot 10 is controlled under other driving conditions indicated by the trajectory model generated by the learning unit 14.

The drive condition estimation unit 21 may estimate the drive condition within a range of predetermined drive conditions. For example, the distance from the motion start point to the motion end point and the maximum speed are predetermined according to the performance of the robot 10, 30, and the drive condition estimation unit 21 may estimate the drive condition within a range in which the distance from the motion start point to the motion end point is less than the predetermined maximum value and the speed is less than the predetermined maximum value.

The learning process performed by the control device 1-5 may be a function of another device. As an example, FIG. 29 shows an example in which the functions of the control device 1 according to embodiment 1 are realized by a control device 6 that controls the robot 10 and a learning device 7 that generates a trajectory model. The robot system 105 shown in FIG. 29 comprises the robot 10, the control device 6 that controls the robot 10, and a learning device 7 that learns a trajectory model for controlling the robot 10.

The learning device 7 includes a drive condition storage unit 11, a reference trajectory determination unit 12, an actual trajectory acquisition unit 13, a learning unit 14, and a learned data storage unit 15. The functions of each unit of the learning device 7 are the same as the functions of the corresponding units of the control device 1. The control device 6 includes a target trajectory determination unit 16 and a control unit 17. The functions of each unit of the control device 6 are the same as the functions of the corresponding units of the control device 1. The target trajectory determination unit 16 of the control device 6 may acquire a trajectory model from the learned data storage unit 15 of the learning device 7. The control unit 17 of the control device 6 controls the robot 10 according to the reference trajectory acquired from the reference trajectory determination unit 12 of the learning device 7 or the target trajectory determined by the target trajectory determination unit 16. The learned data storage unit 15 may be an independent storage device accessible from the control device 6 and the learning device 7. Similarly, the functions of the control devices 2-5 can be realized by the control device 6 and the learning device 7.

As another example, in the robot system 105 shown in FIG. 29, the control device 6 may be provided with the process of generating a trajectory model, which is one of the functions of the learning unit 14 provided in the learning device 7. In this case, the learning device 7 may calculate the trajectory error for each of a plurality of reference midpoints for each drive condition, and store the trajectory error in the learned data storage unit 15. The control device 6 may acquire and retain data previously learned by the learning device 7, specifically, the trajectory error stored in the learned data storage unit 15. The control device 6 may then generate a trajectory model indicating the target midpoint that minimizes the trajectory error for each drive condition.

The trajectory of the robot 10, 30, which is the object of control by the control device 1-6, is not limited to the trajectory of the tip link, but is the trajectory of any part of the robot 10, 30 or the machining tool 200 attached to the robot 30.

The method of acquiring the trajectory of the tip link by the actual trajectory acquisition unit 22 is not limited to the above example. Due to the influence of external factors such as gravity, centrifugal force, processing reaction force, etc., mechanical deflection occurs in at least one of the first arm 110, the second arm 120, the third arm 130, the fourth arm 140, the fifth arm 150, and the flange 160, and the actual trajectory may deviate from the ideal trajectory. In order to acquire the actual trajectory with high accuracy even if mechanical deflection occurs due to external factors, for example, the actual trajectory acquisition unit 22 may have multiple cameras as a measuring device. The actual trajectory acquisition unit 22 may capture the movement range of the tip link of the robot 10 with multiple cameras and determine the position of the tip link of the robot 10 from the images captured by the multiple cameras.

As another example, the actual trajectory acquisition unit 22 has an acceleration sensor attached to the tip link of the robot 10 as a measuring device. The actual trajectory acquisition unit 22 may determine the position of the tip link of the robot 10 based on the measurement value of the acceleration sensor.

The actual trajectory acquisition unit 22 may measure the position of the tip link of the robot 10 for each acquisition cycle that is independent of the control cycle. In this case, the learning unit 14 may interpolate the position of the tip link of the robot 10 for each acquisition cycle, determine the position of the tip link of the robot 10 for each control cycle, and compare it with the ideal position. In detail, the learning unit 14 performs the interpolation process using linear interpolation, polynomial interpolation, etc.

The control unit 17 may feedback control the motors M1, M2, M3, M4, M5, and M6 depending on the position of the tip link of the robot 10 acquired by the above-mentioned actual trajectory acquisition unit 22.

In addition, the above-mentioned hardware configuration and flowchart are merely examples and can be changed and modified as desired. As one example, the control device 1-6 and the learning device 7 may start the learning process in response to the operation of an operation unit (not shown). As another example, the learning unit 14 provided in the control device 1 may generate a trajectory model when learning of the trajectory error for all reference waypoints associated with each driving condition is completed. In other words, the learning unit 14 may calculate the trajectory error for all associated reference waypoints and generate a trajectory model, repeatedly for each driving condition.

In the above-described embodiment, the robots 10 and 30 are actually operated to calculate the trajectory error, but the learning process may also be performed by simulation.

A computer program for performing the above-mentioned operations may be stored on a computer-readable recording medium such as a flexible disk, a CD-ROM (Compact Disc - Read Only Memory), or a DVD-ROM (Digital Versatile Disc - Read Only Memory) and distributed, and the control device 1-6 and learning device 7 that perform the above-mentioned operations may be realized by installing the computer program on the computer. Alternatively, the control device 1-6 and learning device 7 that perform the above-mentioned operations may be realized by a dedicated system. The computer program may be provided via a communication network by being superimposed on a carrier wave.

The robots 10 and 30 are not limited to the above examples and are arbitrary. As an example, the control device 1-6 may control the robot 40, which is a horizontal articulated robot shown in FIG. 30. The robot 40 includes a first arm 410, a second arm 420, a third arm 430, and a base 440. The first arm 410 is attached to the base 440 in a state in which it can rotate around a rotation axis AX1 parallel to the Z axis. The second arm 420 is connected to the first arm 410 and can rotate around a rotation axis AX2 parallel to the Z axis. The third arm 430 is connected to the second arm 420 and can rotate around a rotation axis AX3 parallel to the Z axis. Furthermore, the third arm 430 can expand and contract in a direction along the rotation axis AX3. The base 440 is fixed to a plane and supports the first arm 410. A machining tool 200 can be attached to the third arm 430. The control device 1 - 6 may control a motor (not shown) provided in the robot 40 to control the position of a tip link that is the tip of the third arm 430 .

As another example, the control device 1-6 may control a robot 50 having multiple axes as shown in FIG. 31. The robot 50 includes a first arm 510, a second arm 520, and a base 530. The first arm 510 is attached to the base 530 in a state in which it can move in the X-axis direction. The second arm 520 is connected to the first arm 510 and can move in the Y-axis direction. The base 530 is fixed to a flat surface and supports the first arm 510. A processing tool 200 is attached to the second arm 520. In the example of FIG. 31, the second arm 520 is a print head, and an electrode is formed on a substrate on a stage 540 installed on the base 530 by the processing tool 200 having a liquid discharge unit and a discharge nozzle. In detail, the tip of the processing tool 200 can move in the X-axis direction and the Y-axis direction on the stage 540. The control device 1-6 may control the position of the tip link, which is the tip of the second arm 520, by controlling a motor (not shown) provided in the robot 50.

In the above examples, the positions of the tip links of robots 10, 30, 40, and 50 were defined in an XYZ Cartesian coordinate system, but may also be defined in a cylindrical coordinate system.

Various embodiments and modifications of this disclosure are possible without departing from the broad spirit and scope of this disclosure. Furthermore, the above-described embodiments are intended to explain this disclosure and do not limit the scope of this disclosure. In other words, the scope of this disclosure is indicated by the claims, not the embodiments. Various modifications made within the scope of the claims and within the scope of the disclosure equivalent thereto are deemed to be within the scope of this disclosure.

This application is based on Japanese Patent Application No. 2021-23847, filed on February 18, 2021. The entire specification, claims, and drawings of Japanese Patent Application No. 2021-23847 are incorporated herein by reference.

1, 2, 3, 4, 5, 6 Control device, 7 Learning device, 10, 30, 40, 50 Robot, 11 Drive condition memory unit, 12 Reference trajectory determination unit, 13, 22 Actual trajectory acquisition unit, 14, 18 Learning unit, 15 Learned data memory unit, 16 Target trajectory determination unit, 17 Control unit, 19 Reaction force acquisition unit, 20 Midpoint estimation unit, 21 Drive condition estimation unit, 31 Processor, 32 Memory, 33 Interface, 34 Bus, 100, 101, 102, 103, 104, 105 Robot system, 110, 410, 510 First arm, 120, 420, 520 Second arm, 130, 430 Third arm, 140 Fourth arm, 150 Fifth arm, 160 Flange, 160a Tip surface, 170, 440, 530 Base, 180 fixed part, 190 force sensor, 200 machining tool, 540 stage, AT1, AT1', AT2 actual trajectory, AX1, AX2, AX3, AX4, AX5, AX6 rotation axis, E1, E2, E3, E4, E5, E6 encoder, EP1 operation end point, ER1, ER2 trajectory error, M1, M2, M3, M4, M5, M6 motor, RP, RP1, RP2 reference intermediate point, RT1, RT2 reference trajectory, S1 space, SP1 operation start point, TP1 target intermediate point, TT1 target trajectory.

Claims (22)

A control device for controlling a robot,
a drive condition storage unit that stores drive conditions that specify at least a motion start point and a motion end point of the robot in association with a plurality of reference intermediate points;
a reference trajectory determination unit that acquires any one of the plurality of reference midpoints from the drive condition storage unit, and determines, for each of the acquired reference midpoints, a reference trajectory that passes through the reference midpoint based on the drive condition;
an actual trajectory acquisition unit for acquiring an actual trajectory of the robot;
a learning unit that calculates, for each of the acquired reference midpoints, a trajectory error that indicates a deviation of an actual trajectory acquired by the actual trajectory acquisition unit from an ideal trajectory based on the driving conditions when the robot is controlled according to the reference trajectory, learns a correspondence between the reference midpoints and the trajectory error, and generates a trajectory model that indicates a target midpoint that minimizes the trajectory error according to the driving conditions;
a target trajectory determination unit that determines a target trajectory that passes through the target waypoint obtained from the trajectory model;
a control unit that controls the robot in accordance with the reference trajectory or the target trajectory;
A control device comprising: a middle point estimation unit that estimates the plurality of reference middle points corresponding to the other driving conditions from the target middle points according to the driving conditions indicated by the trajectory model, and associates the other driving conditions with the plurality of estimated reference middle points and stores them in the driving condition storage unit,
The control device according to claim 1 . the learning unit estimates the trajectory error when the robot is controlled according to the target trajectory, and generates the trajectory model indicating the target waypoint and the estimated trajectory error according to the drive conditions;
the midpoint estimating unit estimates the plurality of reference midpoints arranged at intervals according to a magnitude of the estimated trajectory error indicated by the trajectory model;
The control device according to claim 2. a drive condition estimation unit that estimates another reference midpoint corresponding to the drive condition from the trajectory error, which is a deviation of the actual trajectory from the ideal trajectory when the robot is controlled according to the reference trajectory that passes through the reference midpoint predetermined for the drive condition, and that associates the estimated reference midpoint with the drive condition and stores it in the drive condition storage unit.
The control device according to any one of claims 1 to 3. the driving condition estimation unit estimates another driving condition and a plurality of the reference intermediate points corresponding to the other driving condition from the driving condition, the reference intermediate point corresponding to the driving condition, and the trajectory error, and associates the estimated other driving condition with the estimated plurality of reference intermediate points and stores them in the driving condition storage unit.
The control device according to claim 4. the driving condition estimation unit estimates the other reference intermediate point corresponding to the driving condition by a regression algorithm based on the trajectory error;
The control device according to claim 4 or 5. the driving condition estimation unit estimates the other driving condition and the plurality of reference intermediate points corresponding to the other driving condition by a regression algorithm based on the driving condition, the reference intermediate point corresponding to the driving condition, and the trajectory error;
The control device according to claim 5. the actual trajectory acquisition unit has a measuring device that measures a trajectory of the robot, and acquires the trajectory of the robot according to a measurement value of the measuring device;
A control device according to any one of claims 1 to 7. A control device for controlling a robot,
a learning unit that learns a correspondence between a trajectory error, which indicates a deviation of an actual trajectory of the robot controlled according to a reference trajectory that passes through the reference intermediate points, from an ideal trajectory based on the driving conditions, and the reference intermediate points, and that determines a trajectory model that indicates a target intermediate point that minimizes the trajectory error, in accordance with the driving conditions; and
a target trajectory determination unit that determines a target trajectory that passes through the target waypoint obtained from the trajectory model;
A control unit that controls the robot in accordance with the target trajectory;
A control device comprising: the learning unit learns a change in the trajectory error corresponding to a change in the reference waypoint, and generates the trajectory model indicating the target waypoint that minimizes the trajectory error corresponding to the driving conditions.
A control device according to any one of claims 1 to 9. a reaction force acquisition unit that acquires a reaction force, which is a force that the robot receives from a workpiece when the workpiece is processed by a processing tool attached to the robot,
the learning unit learns the reference midpoint, the trajectory error, and a correspondence between the reaction force and the trajectory error when the trajectory error occurs, and generates the trajectory model.
A control device according to any one of claims 1 to 10. the trajectory error indicates a deviation of the actual position from the ideal position when a distance between an ideal position, which is the position of the robot indicated by the ideal trajectory, and an actual position, which is the position of the robot indicated by the actual trajectory, becomes a maximum value;
A control device according to any one of claims 1 to 11. The maximum value is a maximum value of the distance between the ideal position and the actual position.
The control device according to claim 12. The drive conditions further include information about an object to be processed by a processing tool attached to the robot.
A control device according to any one of claims 1 to 13. The information about the object to be processed includes at least one of a weight and a dimension of the object to be processed.
The control device according to claim 14. A control device for controlling a robot,
a target trajectory determination unit that obtains a trajectory model indicating target waypoints that minimize a trajectory error indicating a deviation of an actual trajectory of the robot from an ideal trajectory based on the driving conditions in accordance with the driving conditions that specify at least a motion start point and a motion end point of the robot, and determines a target trajectory that passes through the target waypoints obtained from the trajectory model;
A control unit that controls the robot in accordance with the target trajectory;
A control device comprising: a trajectory model that calculates, for each of reference intermediate points indicating predetermined positions that the tip link can pass through from a motion start point to a motion end point in accordance with a driving condition of a robot having a tip link, a trajectory error corresponding to a deviation of an actual trajectory of the robot from an ideal trajectory based on the driving condition when the robot is controlled based on a reference trajectory that passes through the reference intermediate points, learns the correspondence between the reference intermediate points and the trajectory error, and generates a trajectory model that outputs a target intermediate point that minimizes the trajectory error for each driving condition when a driving command including the driving condition of the robot is input;
Control device. Robots and
A control device according to any one of claims 1 to 17, which controls the robot according to a drive condition that specifies at least a motion start point and a motion end point of the robot;
A robot system comprising: A learning device that learns a trajectory model for controlling a robot, comprising:
a drive condition storage unit that stores drive conditions that specify at least a motion start point and a motion end point of the robot in association with a plurality of reference intermediate points;
a reference trajectory determination unit that acquires any one of the plurality of reference midpoints from the drive condition storage unit, and determines, for each of the acquired reference midpoints, a reference trajectory that passes through the reference midpoint based on the drive condition;
an actual trajectory acquisition unit for acquiring an actual trajectory of the robot;
a learning unit that calculates, for each of the acquired reference midpoints, a trajectory error that indicates a deviation of the actual trajectory acquired by the actual trajectory acquisition unit from an ideal trajectory based on the driving condition when the robot is controlled according to the reference trajectory, for each of the acquired reference midpoints, learns a correspondence between the reference midpoints and the trajectory error, and generates a trajectory model that indicates a target midpoint that minimizes the trajectory error according to the driving condition;
A learning device comprising: for each of reference intermediate points indicating predetermined positions which the tip link may pass through from a motion start point to a motion end point in accordance with a drive condition of a robot having a tip link, a trajectory error corresponding to a deviation of an actual trajectory of the robot from an ideal trajectory based on the drive condition when the robot is controlled based on a reference trajectory passing through the reference intermediate points is obtained by learning the correspondence between the reference intermediate points and the trajectory error,
When a drive command including the drive conditions of the robot is input, a target intermediate point that minimizes the trajectory error for each of the drive conditions is output.
A trajectory model for making computers function. A control method for controlling a robot, comprising:
determining a reference trajectory that passes through any one of a plurality of reference intermediate points associated with a drive condition that specifies at least a motion start point and a motion end point of the robot;
determining, for each of the reference waypoints, a trajectory error indicating a deviation of an actual trajectory of the robot when the robot is controlled according to the reference trajectory from an ideal trajectory based on the driving conditions, learning a correspondence between the reference waypoints and the trajectory error, and generating a trajectory model indicating a target waypoint that minimizes the trajectory error according to the driving conditions;
determining a desired trajectory through the desired waypoint obtained from the trajectory model;
controlling the robot according to the reference trajectory or the target trajectory;
Control methods. The computer that controls the robot
a drive condition storage unit that stores drive conditions that specify at least a motion start point and a motion end point of the robot in association with a plurality of reference intermediate points;
a reference trajectory determination unit that acquires any one of the plurality of reference midpoints from the drive condition storage unit, and determines, for each of the acquired reference midpoints, a reference trajectory that passes through the reference midpoint based on the drive condition;
an actual trajectory acquisition unit for acquiring an actual trajectory of the robot;
a learning unit that calculates, for each of the acquired reference midpoints, a trajectory error that indicates a deviation of the actual trajectory acquired by the actual trajectory acquisition unit from an ideal trajectory based on the driving conditions when the robot is controlled according to the reference trajectory, learns the correspondence between the reference midpoints and the trajectory error, and generates a trajectory model that indicates a target midpoint that minimizes the trajectory error according to the driving conditions;
a target trajectory determination unit that determines a target trajectory passing through the target waypoint obtained from the trajectory model; and
a control unit for controlling the robot in accordance with the reference trajectory or the target trajectory;
A program to function as a

Images