Collision Detection Device for Robot

Abstract

A robot collision detection device (1) comprises: a driving torque calculation unit (3) for calculating an estimated value of the driving torque of the robot; a torque estimation error model learning unit (5) for learning a difference between a driving torque calculated based on a motor current for driving the robot and an estimated value calculated by the driving torque calculation unit (3) and fluctuations in the difference; a threshold value calculation unit (8) for calculating a threshold value based on a torque estimation error model learned by the torque estimation error model learning unit (5); and a collision determination unit (9) for determining a collision between the robot and an object by comparing the difference between the driving torque calculated based on a motor current for driving the robot and the estimated value calculated by the driving torque calculation unit (3) with a threshold value calculated by the threshold value calculation unit (8).

Description

Technical Field

[0001] The present invention relates to a collision detection device for a robot, which detects a collision between the robot and an object in the robot's surroundings. Background Art

[0002] In the past, when a robot made contact or collided with a device around the robot or an operator, the robot would detect the contact or collision and stop its movement, thereby preventing damage to the device around the robot, the operator, or the robot itself. In the past, the robot controller calculated the torque required to perform the action performed by the robot and derived a torque measurement value based on the actual motor current value or the measurement value of the torque sensor installed in the drive mechanism. The robot controller compared the required torque value with the torque measurement value, and if the difference between the required torque value and the torque measurement value exceeded a threshold, it determined that the robot had collided with a device around the robot or the operator. In order to detect collisions between the robot and devices around the robot or the operator with high sensitivity, a technology that can reduce the threshold as much as possible without causing false detection is required.

[0003] Patent Document 1 discloses a technique for estimating parameters related to the inertia and friction of a robot and using the estimated parameters to improve the accuracy of calculating or estimating the torque required for the robot's movements. Patent Document 1 also discloses a technique for reducing the influence of factors not modeled by the calculation unit that calculates the required torque by using a high-pass filter.

[0004] Patent Document 1: Japanese Patent Application No. 2016-511699 Summary of the Invention

[0005] While existing technologies address fluctuations in robot dynamics parameters, such as the effect of changes in the workpiece gripped by the robot's hand attached to its fingertips, they do not consider the effects of torque fluctuations when the robot is operating under the same conditions. Specifically, in existing technologies, during operations with large torque fluctuations, the robot may falsely detect a collision even if no collision has occurred with peripheral devices or the operator. Furthermore, existing technologies use high-pass filters to remove the effects of unmodeled factors, such as the effects of elasticity caused by the transmission mechanism or the nonlinearity of friction. However, in the event of a collision, such as when the robot slowly contacts peripheral devices or the operator, only considering errors in the high-frequency region may miss collisions that are difficult to detect.

[0006] The present invention is proposed in view of the above situation, and its purpose is to obtain a robot collision detection device that takes into account the influence of unmodeled factors, prevents false detection, and detects the collision between the robot and the objects around the robot with relatively high sensitivity when the robot collides with the objects.

[0007] In order to solve the above-mentioned problems and achieve the purpose, the collision detection device of the robot involved in the present invention includes: a driving torque calculation unit, which calculates an estimated value of the driving torque of the robot; a torque estimation error model learning unit, which learns the difference between the driving torque calculated based on the motor current used to drive the robot or the driving torque derived from the torque sensor provided in the driving part and the estimated value calculated by the driving torque calculation unit, and the fluctuation of the difference; a threshold calculation unit, which calculates a threshold based on the torque estimation error model learned by the torque estimation error model learning unit; and a collision judgment unit, which judges the collision between the robot and the object by comparing the difference between the driving torque calculated based on the motor current used to drive the robot and the estimated value calculated by the driving torque calculation unit with the threshold calculated by the threshold calculation unit.

[0008] Effects of the Invention

[0009] The robot collision detection device involved in the present invention has the following effects, namely, taking into account the influence of unmodeled factors including their fluctuations, and preventing false detection while being able to detect the collision between the robot and the object around the robot with relatively high sensitivity when the robot collides with the object. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] Figure 1 This is a diagram showing the configuration of a robot collision detection device according to the first embodiment.

[0011] Figure 2 This is a diagram showing the configuration of a robot collision detection device according to a second embodiment.

[0012] Figure 3 This is a diagram showing the configuration of a robot collision detection device according to a fifth embodiment.

[0013] Figure 4 This is a diagram showing the configuration of a robot collision detection device according to a sixth embodiment.

[0014] Figure 5 This is a diagram showing the configuration of a robot collision detection device according to an eighth embodiment.

[0015] Figure 6This is a diagram showing the configuration of a robot collision detection device according to a ninth embodiment.

[0016] Figure 7 This is a diagram showing a processor in a case where at least a portion of the drive torque calculation unit, action state calculation unit, torque estimation error model learning unit, conversion unit, error calculation unit, threshold calculation unit, and collision determination unit of the collision detection device of the robot involved in embodiment 1 is implemented by a processor.

[0017] Figure 8 This is a diagram showing a processing circuit in a case where at least a portion of the drive torque calculation unit, action state calculation unit, torque estimation error model learning unit, conversion unit, error calculation unit, threshold calculation unit, and collision determination unit of the collision detection device of the robot involved in embodiment 1 is implemented by the processing circuit. DETAILED DESCRIPTION

[0018] Hereinafter, a robot collision detection device according to an embodiment will be described in detail with reference to the drawings.

[0019] Implementation method 1.

[0020] Figure 1 This figure shows the structure of a robot collision detection device 1 according to Embodiment 1. Hereinafter, the robot collision detection device 1 may be referred to as “collision detection device 1.” The collision detection device 1 includes a robot control device 2 that controls the robot.

[0021] exist Figure 1 No robot is shown.

[0022] The robot control device 2 includes a drive torque calculation unit 3 that receives information indicating the position of the motors of the respective axes driving the robot and calculates an estimated value of the drive torque of the robot. The drive torque calculation unit 3 calculates the motor speed and motor acceleration of each axis by taking a difference with respect to the motor position. The drive torque calculation unit 3 calculates the drive torque of the robot based on the motor position, motor speed, and motor acceleration, and the following robot motion equation (1).

[0023] τ=M(q)a+h(q,v)+g(q)+f(v) · · · (1)

[0024] In equation (1), "τ" is a vector consisting of the driving torque of each axis of the robot, "q" is a vector consisting of the position of each axis of the robot after converting the motor position of each axis of the robot to the output position of the transmission mechanism. "a" is a vector consisting of the acceleration of each axis of the robot after converting the motor acceleration to the output acceleration of the transmission mechanism of each axis of the robot, and "v" is a vector consisting of the speed of each axis of the robot after converting the motor speed to the output speed of the transmission mechanism of each axis of the robot.

[0025] M(q)a is the inertial force of each axis of the robot, h(q, v) is the centrifugal Coriolis force, g(q) is gravity, and f(v) is friction, all of which contribute to the driving torque. Friction is the sum of Coulomb friction, which is determined by the direction of velocity, and viscous friction, which is determined by the direction and magnitude of velocity. The simplest viscous friction model is a velocity-proportional model, which is used in Implementation 1. The viscous friction model can be a velocity polynomial model or a velocity power exponential model.

[0026] The collision detection device 1 further includes an action state calculation unit 4, which is located outside the robot control device 2. The action state calculation unit 4 calculates state quantities related to the action state of the robot. Specifically, the action state calculation unit 4 calculates state quantities including information related to motor speed, information related to motor acceleration, and all or part of a portion of the elements of the robot's driving torque. A portion of the elements of the robot's driving torque is any one of the inertial force M(q)a, the centrifugal Coriolis force h(q, v), the gravity g(q), the friction force f(v), or the sum of the driving torque elements such as M(q)a+h(q, v). In the first embodiment, the action state calculation unit 4 receives information indicating the motor position, calculates the motor speed by differentiating the motor position, and calculates the action state quantity indicating the speed of each axis of the robot after converting the motor speed into the output speed of the transmission mechanism of each axis of the robot.

[0027] The collision detection device 1 further includes a torque estimation error model learning unit 5, which is located outside the robot controller 2. The torque estimation error model learning unit 5 learns the difference between the drive torque calculated based on the motor current used to drive the robot and the estimated value of the drive torque calculated by the drive torque calculation unit 3, as well as the fluctuations in this difference. The motion state calculation unit 4 outputs information representing the motion state quantity to the torque estimation error model learning unit 5. The torque estimation error model learning unit 5 learns the state quantity calculated by the motion state calculation unit 4 as an input signal for the correction function. One or both of the motion state calculation unit 4 and the torque estimation error model learning unit 5 may be located within the robot controller 2.

[0028] The robot control device 2 also includes a conversion unit 6 that converts the motor current according to the torque constant of the motor and the transmission ratio of the transmission mechanism to calculate the measured value of the driving torque of each axis of the robot, that is, the measured torque. The collision detection device 1 also includes an error calculation unit 7, which is located outside the robot control device 2. The driving torque calculation unit 3 outputs information representing the estimated value of the driving torque, that is, the estimated torque, to the error calculation unit 7. The conversion unit 6 outputs information representing the measured torque to the error calculation unit 7. The error calculation unit 7 subtracts the estimated torque calculated by the driving torque calculation unit 3 from the measured torque obtained by the conversion unit 6 to calculate the difference between the measured torque and the estimated torque, that is, the torque estimation error. The error calculation unit 7 outputs information representing the torque estimation error to the torque estimation error model learning unit 5.

[0029] During learning, the torque estimation error model learning unit 5 receives information indicating the motion state quantity output from the motion state calculation unit 4 and information indicating the torque estimation error output from the error calculation unit 7. The torque estimation error model learning unit 5 includes a learning unit 5a for each axis that uses a nonparametric method. The learning unit 5a employs Gaussian process regression. The nonparametric learning unit 5a can employ kernel density estimation or the K-nearest neighbor method.

[0030] When the torque estimation error model learning unit 5 performs learning, the output of the operating state calculation unit 4 is set to x1, x2, ..., xn, and the difference between the measured torque corresponding to xi and the estimated torque calculated by the drive torque calculation unit 3 is set to yi. n is an integer greater than or equal to 2, and i is an integer greater than or equal to 1 and less than or equal to n. The torque estimation error model learning unit 5 uses n pairs D = {(x1, y1), (x2, y2), ..., (xn, yn)} consisting of input x and output y as learning data to learn the hyperparameters of the kernel function used in Gaussian process regression. Gaussian kernels and radial basis functions (RBF) kernels are well known as kernel functions for Gaussian process regression, and the torque estimation error model learning unit 5 uses RBF as the kernel function. The torque estimation error model learning unit 5 can also use kernel functions other than RBF, such as exponential kernels or periodic kernels.

[0031] After completing the learning, the torque estimation error model learning unit 5 obtains a torque estimation error model for calculating the predicted distribution of y* for the new input x*. The following equation (2) shows an example of the torque estimation error model.

[0032] P(y*|x*, D)=N(k* T K －1 y, k**-k* T K －1 k*)··· (2)

[0034] k* in the formula (2) is represented by the following formula (3), and k** in the formula (2) is represented by the following formula (4).

[0035] k*=(k(x*,x1),k(x*,x2),···,k(x*,xn)) T ···(3)

[0036] k**=k(x*,x*) · ·· (4)

[0037] k() is the kernel function. K is a matrix called the kernel matrix, and the ijth component of the matrix is k(xi, xj). j is an integer greater than or equal to 1 and less than or equal to n. N(b, σ 2 ) has mean b and variance σ 2 The probability density function of the Gaussian distribution.

[0038] The robot controller 2 further includes a threshold calculation unit 8 that receives information indicating the motor position. The torque estimation error model learning unit 5 outputs information indicating the torque estimation error model to the threshold calculation unit 8. The threshold calculation unit 8 calculates a threshold value based on the torque estimation error model learned by the torque estimation error model learning unit 5. Specifically, the threshold calculation unit 8 performs the same calculation as that performed by the motion state calculation unit 4. If the motion state calculation unit 4 calculates the motor speed, the threshold calculation unit 8 calculates the motor speed. If the motion state calculation unit 4 calculates the motor acceleration, the threshold calculation unit 8 calculates the motor acceleration. If the motion state calculation unit 4 calculates a portion of the driving torque element, such as the inertial force M(q)a, the threshold calculation unit 8 calculates the inertial force M(q)a. In the first embodiment, the threshold calculation unit 8 calculates the motor speed by differentiating the motor position and converts the motor speed into the output speed of the transmission mechanism of each axis of the robot to calculate the speed of each axis of the robot. The threshold calculation unit 8 performs the calculation of equation (2) using the speed of each axis as x* in equation (2).

[0039] The robot control device 2 also includes a collision determination unit 9. After performing the calculation of formula (2), the threshold value calculation unit 8 uses the upper limit and lower limit of the range of ±2σ as threshold values and outputs information indicating the threshold values to the collision determination unit 9. When performing the calculation of formula (2), the threshold value calculation unit 8 can perform calculations in accordance with the definition, for example, a method such as the auxiliary variable method that reduces the amount of calculation of Gaussian process regression can be used. The collision determination unit 9 compares the difference between the driving torque calculated based on the motor current used to drive the robot and the estimated value calculated by the driving torque calculation unit 3 with the threshold value calculated by the threshold value calculation unit 8, thereby determining whether the robot and the object have collided.

[0040] The driving torque calculation unit 3 outputs information indicating the calculated estimated torque to the collision determination unit 9. The conversion unit 6 outputs information indicating the measured torque to the collision determination unit 9. The collision determination unit 9 receives information output from the driving torque calculation unit 3, the conversion unit 6, and the threshold calculation unit 8. If the difference between the measured torque obtained by the conversion unit 6 and the estimated torque calculated by the driving torque calculation unit 3 is greater than or equal to an upper limit or less than or equal to a lower limit, the collision determination unit 9 determines that a collision between the robot and the object is about to occur or has occurred, and stops the robot.

[0041] The robot collision detection device 1 according to Embodiment 1 compares the difference between the drive torque calculated based on the motor current driving the robot and the estimated value calculated by the drive torque calculation unit 3 with a threshold value calculated by the threshold calculation unit 8 to determine whether the robot has collided with an object. The threshold calculation unit 8 calculates the threshold value based on the torque estimation error model learned by the torque estimation error model learning unit 5. Therefore, the robot collision detection device 1 can set a threshold value that takes into account the effects of torque estimation errors, such as nonlinearities in the elasticity or friction of the transmission mechanism that are not modeled by the drive torque calculation unit 3, and fluctuations in the torque estimation error. As a result, the robot collision detection device 1 can prevent false detections while determining with relatively high sensitivity that a collision has occurred between the robot and the object.

[0042] Furthermore, the robot collision detection device 1 includes an operation state calculation unit 4 that calculates state quantities related to the robot's operation state; a torque estimation error model learning unit 5 that learns the state quantities calculated by the operation state calculation unit 4 and uses them as input signals for a correction function; and a threshold value calculation unit 8 that calculates a threshold value based on the torque estimation error model learned by the torque estimation error model learning unit 5. Therefore, the robot collision detection device 1 can set a threshold value that takes into account state quantities related to the robot's operation state, the presence of correlation, the influence of torque estimation errors not modeled by the drive torque calculation unit 3, and fluctuations in the torque estimation error. As a result, the robot collision detection device 1 can detect collisions between the robot and surrounding objects with relatively high sensitivity while taking into account the influence of unmodeled factors and preventing false detections.

[0043] Furthermore, in the first embodiment, the motion state calculation unit 4 and the threshold calculation unit 8 calculate the speed of each axis after converting the motor speed to the output speed of the transmission mechanism. The robot collision detection device 1 may use the motor speed as x1, x2, ..., xn, and x*, rather than the speed of each axis after converting it to the output speed of the transmission mechanism. The robot collision detection device 1 may use the norm of the motor speed of each axis, or the norm of the speed of each axis after converting the motor speed to the output speed of the transmission mechanism, as x1, x2, ..., xn, and x*.

[0044] Implementation method 2.

[0045] Figure 2This figure shows the configuration of a robot collision detection device 1A according to a second embodiment. The robot collision detection device 1A includes all the components of the robot collision detection device 1 according to the first embodiment, excluding the conversion unit 6. The robot collision detection device 1A includes a robot control unit 2A, which includes a drive torque calculation unit 3, a threshold calculation unit 8, and a collision determination unit 9, but does not include the conversion unit 6. In the second embodiment, the differences from the first embodiment will be mainly described.

[0046] In the second embodiment, the measured torque, which is the measured value of the driving torque of each axis of the robot, is measured by a torque sensor provided in the driving part of the robot. Figure 2 The drive unit and torque sensor are not shown. Information indicating the measured torque is received by the error calculation unit 7 and the collision determination unit 9. The drive torque calculation unit 3 calculates an estimated value of the drive torque, excluding the friction torque of the drive unit and the acceleration and deceleration torque of the motor itself, based on the following equation (5).

[0047] τ=M L (q)a+h(q,v)+g(q) · · · (5)

[0048] In formula (5), M L (q) is obtained by removing the inertia of each axis on the motor side from the robot's inertia matrix using the torque sensor. The above inertia includes the inertia of the motor itself.

[0049] Torque estimation error model learning unit 5 learns the difference between the driving torque measured by the torque sensor and the estimated driving torque calculated by driving torque calculation unit 3, as well as the fluctuation in this difference. Collision determination unit 9 determines a collision between the robot and an object by comparing the difference between the driving torque measured by the torque sensor and the estimated driving torque calculated by driving torque calculation unit 3 with a threshold value calculated by threshold calculation unit 8.

[0050] The robot collision detection device 1A according to the second embodiment can set a threshold value for torque estimation error fluctuations, taking into account the influence of factors not modeled by the drive torque calculation unit 3, such as the elasticity of the transmission mechanism and the tension from the cables attached to the robot, based on the measured torque obtained by the torque sensor, which is unaffected by the motor and transmission mechanism. Therefore, the robot collision detection device 1A can determine with relatively high sensitivity that a collision has occurred between the robot and an object, while taking into account the influence of factors not modeled and preventing false detections.

[0051] Implementation method 3.

[0052] Figure 1 This figure also shows the configuration of a robot collision detection device according to a third embodiment. In the first embodiment, the motion state calculation unit 4 receives information indicating the motor position, calculates the motor speed by taking a difference in the motor position, and then calculates the speed of each axis by converting the motor speed into the output speed of the transmission mechanism as the motion state quantity. In the third embodiment, the motion state calculation unit 4 receives information indicating the motor position, calculates the motor acceleration by taking a difference in the motor position twice, and then calculates the acceleration of each axis by converting the motor acceleration into the output acceleration of the transmission mechanism as the motion state quantity.

[0053] Upon receiving information indicating the motor position, the threshold calculation unit 8 performs the same calculation as that performed by the operating state calculation unit 4. Specifically, the threshold calculation unit 8 calculates the motor acceleration by performing a second difference with respect to the motor position, and then calculates the acceleration of each axis after converting the motor acceleration into the output acceleration of the transmission mechanism. Next, the threshold calculation unit 8 performs the calculation of equation (2) using the acceleration of each axis derived for each axis as x* in equation (2). Matters other than the above are the same as those in embodiment 1, and therefore, description of these matters will be omitted.

[0054] The robot collision detection device according to the third embodiment can correct the influence of the torque estimation error of each axis that is correlated with the acceleration or acceleration norm of each axis with relatively high accuracy.

[0055] Furthermore, the operating state calculation unit 4 and the threshold calculation unit 8 calculate the acceleration of each axis by converting the motor acceleration into the output acceleration of the transmission mechanism. The motor acceleration is used in the calculation of equation (2) or in the learning of the torque estimation error model learning unit 5. Instead of the motor acceleration of each axis, the norm of the motor acceleration of each axis or the norm of the acceleration of each axis after converting the motor acceleration into the output acceleration of the transmission mechanism may be used as x1, x2, ..., xn, and x*.

[0056] Implementation method 4.

[0057] Figure 1This figure also shows the structure of the collision detection device of the robot involved in the fourth embodiment. In the first embodiment, the motion state calculation unit 4 receives information indicating the motor position, calculates the motor speed by taking a difference with respect to the motor position, and then calculates the speed of each axis after converting the motor speed to the output speed of the transmission mechanism as the motion state quantity. In the fourth embodiment, the motion state calculation unit 4 receives information indicating the motor position, calculates the motor speed and motor acceleration by taking a difference with respect to the motor position, and then calculates the speed of each axis after converting the motor speed to the output speed of the transmission mechanism, and calculates the acceleration of each axis after converting the motor acceleration to the output acceleration of the transmission mechanism, and calculates the motion state quantity τ based on the following formula (6). The calculated motion state quantity τ is used in the learning of the torque estimation error model learning unit 5.

[0058] τ=M(q)a+h(q,v)+g(q) · · · (6)

[0059] Upon receiving information indicating the motor position, the threshold value calculation unit 8 calculates the motor speed and motor acceleration in the same manner as the motion state calculation unit 4. The threshold value calculation unit 8 calculates the speed of each axis after converting the motor speed to the output speed of the transmission mechanism, and the acceleration of each axis after converting the motor acceleration to the output acceleration of the transmission mechanism. The calculation of equation (2) is performed using the element of each axis of the motion state quantity τ calculated based on equation (6) as x* in equation (2). Matters other than the above are the same as those in embodiment 1, and therefore, description of these matters will be omitted.

[0060] The robot collision detection device according to the fourth embodiment can correct the influence of the driving torque of each axis or the torque estimation error of each axis correlated with the driving torque with relatively high accuracy.

[0061] In the above description, the operating state calculation unit 4 uses the sum of the elements M(q)a, h(q, v), and g(q) in equation (6) as the operating state quantity. M(q)a, h(q, v), and g(q) are all part of the elements of the driving torque. Since the friction force f(v) is not included, the sum of M(q)a, h(q, v), and g(q) is also an example of a part of the elements of the driving torque. However, the operating state calculation unit 4 can also use any one of M(q)a, h(q, v), and g(q) as the operating state quantity. This case is also an example of using a part of the elements of the driving torque. The threshold calculation unit 8 can perform the calculation of equation (6) or use the calculation result of the driving torque calculation unit 3. Moreover, instead of the sum of the elements M(q)a, h(q, v), and g(q) in equation (6), the calculation result equivalent to the calculation result of equation (1) added up to the friction force f(v) can be used as the operating state quantity. When the same calculation result as that of the equation (1) is used as the operation state quantity, the threshold value calculation unit 8 may perform the calculation of the equation (1) or use the calculation result of the driving torque calculation unit 3 .

[0062] Implementation method 5.

[0063] Figure 3 This is a diagram showing the structure of a robot collision detection device 1B according to a fifth embodiment. The robot collision detection device 1B includes all the structural elements of the robot collision detection device 1 according to the first embodiment, except for the drive torque calculation unit 3. The robot collision detection device 1B includes a drive torque calculation unit 3B instead of the drive torque calculation unit 3. The robot collision detection device 1B includes a robot control device 2B, which includes a drive torque calculation unit 3B, a conversion unit 6, a threshold calculation unit 8, and a collision determination unit 9. The robot collision detection device 1B also includes a parameter identification unit 10, which is located outside the robot control device 2B. The parameter identification unit 10 may be located inside the robot control device 2B. In the fifth embodiment, the differences from the first embodiment are mainly described.

[0064] The parameter identification unit 10 identifies the values of the parameters of the motion equation used by the drive torque calculation unit 3B based on pre-measured data. That is, the parameter identification unit 10 identifies all or part of the parameters of the motion equation of the robot of formula (1). When the parameter identification unit 10 identifies all the parameters of the motion equation of the robot of formula (1), the parameter identification unit 10 uses the vector p formed by treating the calculation results obtained by using the parameters such as mass, center of gravity position and friction coefficient themselves or the mass × center of gravity position as two parameters greater than or equal to the new parameters, and transforms the formula (1) into the following formula (7). In addition, the parameter identification unit 10 is based on the vector Y derived from the position, velocity and acceleration at each moment. p and derive vector Y p The driving torque τ at the time is calculated using the least squares method for parameter p. The calculated parameter p is used in the calculation performed by the driving torque calculation unit 3B using equation (1). In other words, the driving torque calculation unit 3B calculates the estimated value of the driving torque using the parameter values determined by the parameter determination unit 10.

[0065] τ=M(q)a+h(q,v)+g(q)+f(v)=Y p T p··· (7)

[0066] When the parameter identification unit 10 identifies a part of the parameters of the robot's motion equation of formula (1), the parameter values are set to known, the inertia matrix, centrifugal force, gravity and friction force calculated based on the parameters deviating from the identified object are set to M0(q), h0(q, v), g0(q), f0(v) respectively, the vector composed of the parameters of the identified object is set to p1, and the motion equation of formula (1) is transformed into the following formula (8).

[0067] τ=M0(q)a+h0(q,v)+g0(q)+f0(v)+Y p1 T p1···(8)

[0068] τ0 is defined by the following formula (9).

[0069] τ0=M0(q)a+h0(q,v)+g0(q)+f0(v) ··· (9)

[0070] The parameter determination unit 10 calculates τ1=τ-τ0 based on τ0 derived from the position, velocity and acceleration at each moment and the driving torque τ when τ0 is derived, and compares it with Y derived from the position, velocity and acceleration at each moment. p1Accordingly, the parameter p1 is calculated using the least squares method. The calculated parameter p1 is used in the calculation performed by the driving torque calculation unit 3B using equation (1). The driving torque calculation unit 3B differs from the driving torque calculation unit 3 in that the driving torque calculation unit 3B uses the calculated parameter p or the calculated parameter p1.

[0071] The robot collision detection device 1B according to the fifth embodiment can improve the accuracy of the parameter values of the model related to the dynamic characteristics of the robot, thereby preventing false detections and detecting a collision between the robot and an object with relatively high sensitivity. Furthermore, even if the parameter values are unknown, the robot collision detection device 1B can prevent false detections and detect a collision between the robot and an object with relatively high sensitivity.

[0072] Implementation method 6.

[0073] Figure 4 1 is a diagram showing the structure of a robot collision detection device 1C according to a sixth embodiment. The robot collision detection device 1C includes all the structural elements of the robot collision detection device 1 according to the first embodiment, except for the drive torque calculation unit 3. The robot collision detection device 1C includes a drive torque calculation unit 3C instead of the drive torque calculation unit 3. The robot collision detection device 1C also includes an online parameter identification unit 11. The robot collision detection device 1C includes a robot control device 2C, which includes a drive torque calculation unit 3C, a conversion unit 6, a threshold calculation unit 8, a collision determination unit 9, and an online parameter identification unit 11. In the sixth embodiment, the differences from the first embodiment will be mainly described.

[0074] The online parameter locating unit 11 locates the values of the parameters of the motion equation used by the drive torque calculation unit 3C based on data from the robot's motion. For example, the online parameter locating unit 11 uses an adaptive locating method to locate the values of unknown parameters and parameters whose values vary during the robot's motion. The online parameter locating unit 11 locates all or part of the parameters of the robot's motion equation in equation (1).

[0075] When all parameters of the robot motion equation of equation (1) are aligned, the online parameter identification unit 11 sets τ of equation (7) in the kth identification cycle to τ[k], and sets Y p Set to Y p[k], the fixed value of p in the kth fixed period is set to p[k], the fixed period is set to moit, and fixed based on the following equations (10), (11), and (12). k is an integer greater than or equal to 1 and less than or equal to n.

[0076] R[k]=R[k-1]+moit*(-k1*R[k-1]+Y p [k]Y p

[0077] [k] T )···(10)

[0078] r[k]=r[k-1]+moit*(-k1*r[k-1]+τ[k]*Y p

[0079] [k])···(11)

[0080] p[k]=p[k-1]-moit*G1·(R[k]·p[k-1]-r[k])···(12)

[0081] k1 is a weighting coefficient for adjusting the constant speed, and G1 is a gain matrix for adjusting the constant speed.

[0082] When the online parameter identification unit 11 identifies a part of the parameters of the robot motion equation of equation (1), in equations (10), (11), and (12), τ is replaced by τ1, and Y is replaced by p Replace with Y p1 , parameter p1 is determined instead of p. The online parameter determination unit 11 outputs information indicating the determined parameter value to the driving torque calculation unit 3C. The driving torque calculation unit 3C receives the information output from the online parameter determination unit 11, and uses the parameter value indicated by the received information for the calculation of the motion equation. That is, when the driving torque calculation unit 3C calculates the estimated value of the driving torque, it uses the value determined by the online parameter determination unit 11. The difference between the driving torque calculation unit 3C and the driving torque calculation unit 3 is that when the driving torque calculation unit 3C calculates the estimated value of the driving torque, it uses the value determined by the online parameter determination unit 11. In addition, the online parameter determination unit 11 outputs information indicating the parameter value that has been updated successively to the driving torque calculation unit 3C.

[0083] The collision detection device 1C of the robot involved in embodiment 6 can also improve the accuracy of the values of the parameters of the model related to the dynamic characteristics of the robot when the values of the parameters related to the dynamic characteristics of the robot change during the movement of the robot, thereby preventing false detection and detecting the collision between the robot and the object with relatively high sensitivity when the robot collides with the object.

[0084] Implementation method 7.

[0085] Figure 1 This figure also shows the structure of a robot collision detection device according to a seventh embodiment. In the first embodiment, the motion state calculation unit 4 receives information indicating the motor position, calculates the motor speed by taking a difference in the motor position, and then calculates the speed of each axis by converting the motor speed into the output speed of the transmission mechanism as the motion state quantity. In the seventh embodiment, the motion state calculation unit 4 receives information indicating the motor position, calculates the motor speed and motor acceleration by taking a difference in the motor position, and then calculates the speed of each axis by converting the motor speed into the output speed of the transmission mechanism and the acceleration of each axis by converting the motor acceleration into the output acceleration of the transmission mechanism. The motion state calculation unit 4 calculates a vector consisting of the speed and acceleration of each axis as the motion state quantity.

[0086] In the first embodiment, the input elements of the kernel functions used by the torque estimation error model learning unit 5 and the threshold calculation unit 8 were scalars, but in the seventh embodiment, these elements are vectors. The robot collision detection device according to the seventh embodiment can correct the influence of the torque estimation error of each axis, which is correlated with both the acceleration and velocity of each axis, with relatively high accuracy.

[0087] The acceleration of the vector can be replaced by a value obtained by multiplying the acceleration of each axis by a weighting factor after converting the motor acceleration into the output acceleration of the transmission mechanism. The motion state calculation unit 4 calculates not only the velocity and acceleration of the axis to be learned as the motion state quantity, but also a vector composed of the velocities and accelerations of all axes as the motion state quantity. The motion state calculation unit 4 calculates the vector composed of the positions of all axes, the velocities of all axes, and the result of multiplying the accelerations of all axes by a weighting factor as the motion state quantity.

[0088] Implementation method 8.

[0089] Figure 5This figure shows the structure of a robot collision detection device 1D according to an eighth embodiment. The robot collision detection device 1D includes all the components of the robot collision detection device 1 according to the first embodiment, except for the threshold value calculation unit 8. The robot collision detection device 1D includes a threshold value calculation unit 8D instead of the threshold value calculation unit 8. The robot collision detection device 1D includes a robot control device 2D, which includes a drive torque calculation unit 3, a conversion unit 6, a threshold value calculation unit 8D, and a collision determination unit 9. The robot collision detection device 1D also includes an approximate function learning unit 12, which is located outside the robot control device 2D. In the eighth embodiment, the differences from the first embodiment will be mainly described.

[0090] The approximate function learning unit 12 learns an approximate function based on the torque estimation error model learned by the torque estimation error model learning unit 5. Specifically, the approximate function learning unit 12 receives equation (2) for calculating the predicted distribution and the data used to derive equation (2). In the case of additional data, the approximate function learning unit 12 also receives the output of the action state calculation unit 4. The approximate function learning unit 12 uses the function and parameters built into equation (2), the data used by the torque estimation error model learning unit 5 during learning, and the data newly added from the action state calculation unit 4 to obtain an approximate function that outputs the upper limit and lower limit of the range of ±2σ of the estimated value of the torque estimation error through learning. For example, the approximate function is a feedforward neural network or a recursive neural network.

[0091] After learning, the approximate function learning unit 12 outputs information indicating the learned approximate function to the threshold calculation unit 8D. The threshold calculation unit 8D calculates the threshold using the approximate function learned by the approximate function learning unit 12. That is, the threshold calculation unit 8D calculates the threshold using the approximate function derived by the approximate function learning unit 12. Specifically, upon receiving information indicating the motor position, the threshold calculation unit 8D performs the same calculation as the operation state calculation unit 4, inputs the calculation result into the approximate function obtained from the approximate function learning unit 12, and outputs the calculation result of the approximate function as the positive and negative thresholds to the collision determination unit 9. The positive threshold is the upper limit value described above, and the negative threshold is the lower limit value described above. Matters other than the above are the same as those in the first embodiment, and therefore descriptions of matters other than the above are omitted.

[0092] The robot collision detection device 1D according to the eighth embodiment uses an approximate function instead of a torque estimation error model when calculating the threshold value, thereby reducing the amount of threshold value calculation and thus allowing the threshold value to be calculated in a relatively short time.

[0093] Implementation method 9.

[0094] Figure 6 This figure shows the configuration of a robot collision detection device 1E according to a ninth embodiment. The robot collision detection device 1E includes all the components of the robot collision detection device 1 according to the first embodiment, excluding the torque estimation error model learning unit 5 and the threshold value calculation unit 8. The robot collision detection device 1E includes a torque estimation error model learning unit 5E in place of the torque estimation error model learning unit 5, and a threshold value calculation unit 8E in place of the threshold value calculation unit 8. The robot collision detection device 1E also includes a temperature measurement unit 13 for measuring temperature. For example, the temperature measurement unit 13 is a temperature sensor attached to an encoder that measures the angle of the motor of each axis of the robot. The robot collision detection device 1E includes a robot control unit 2E, which includes a drive torque calculation unit 3, a conversion unit 6, a threshold value calculation unit 8E, a collision determination unit 9, and a temperature measurement unit 13. The ninth embodiment will be described primarily regarding the differences from the first embodiment.

[0095] When the robot's collision detection device 1E is learning, the temperature measurement unit 13 outputs information indicating the measured temperature to the torque estimation error model learning unit 5E. The torque estimation error model learning unit 5E uses the temperature measured by the temperature measurement unit 13 for learning. In the first embodiment, the torque estimation error model learning unit 5 uses the speed of each axis of the robot as input for Gaussian process regression when learning. The torque estimation error model learning unit 5E uses a vector consisting of the speed of each axis and the temperature of each axis as input for the Gaussian process regression of each axis of the robot. The difference between the torque estimation error model learning unit 5E and the torque estimation error model learning unit 5 is that the torque estimation error model learning unit 5E uses a vector consisting of the speed of each axis and the temperature of each axis as input for the Gaussian process regression of each axis of the robot. The temperature of each axis can be replaced by a value obtained by multiplying the temperature of each axis by a weighting coefficient.

[0096] When the robot's collision detection device 1E determines whether the robot has collided with an object during its movement, the temperature measurement unit 13 outputs information indicating the measured temperature to the threshold calculation unit 8E. The threshold calculation unit 8E calculates the threshold value using the temperature measured by the temperature measurement unit 13. In the first embodiment, the kernel function used by the threshold calculation unit 8 takes the velocity of each axis as an input element. The threshold calculation unit 8E uses a vector consisting of the velocity and temperature of each axis. The threshold calculation unit 8E calculates the threshold value using the temperature measured by the temperature measurement unit 13. The difference between the threshold calculation unit 8E and the threshold calculation unit 8 is that the threshold calculation unit 8E calculates the threshold value using the temperature measured by the temperature measurement unit 13. While the torque estimation error model learning unit 5E uses a value obtained by multiplying the temperature of each axis by a weighting coefficient during learning, the threshold calculation unit 8E uses a vector consisting of the velocity of each axis and the value obtained by multiplying the temperature of each axis by a weighting coefficient. All other matters are the same as those in the first embodiment, and therefore their description is omitted.

[0097] The robot collision detection device 1E involved in embodiment 9 can correct the influence of the torque estimation error of each axis that is correlated with the temperature of each axis with relatively high accuracy, and thus can prevent false detection while detecting the collision between the robot and the object with relatively high sensitivity when the robot collides with the object.

[0098] Figure 7 This diagram shows a processor 71 in the case where at least a portion of the drive torque calculation unit 3, the motion state calculation unit 4, the torque estimation error model learning unit 5, the conversion unit 6, the error calculation unit 7, the threshold value calculation unit 8, and the collision determination unit 9 included in the robot collision detection device 1 according to Embodiment 1 is implemented by the processor 71. Specifically, at least a portion of the functions of the drive torque calculation unit 3, the motion state calculation unit 4, the torque estimation error model learning unit 5, the conversion unit 6, the error calculation unit 7, the threshold value calculation unit 8, and the collision determination unit 9 can be implemented by the processor 71 executing a program stored in the memory 72.

[0099] The processor 71 is a CPU (Central Processing Unit), a processing device, an arithmetic device, a microprocessor or a DSP (Digital Signal Processor). Figure 7 A memory 72 is also shown.

[0100] When at least a portion of the functions of the drive torque calculation unit 3, the operating state calculation unit 4, the torque estimation error model learning unit 5, the conversion unit 6, the error calculation unit 7, the threshold value calculation unit 8, and the collision determination unit 9 are implemented by the processor 71, the at least a portion of the functions are implemented by the processor 71 and software, firmware, or a combination of software and firmware. The software or firmware is described as a program and stored in the memory 72. The processor 71 reads and executes the program stored in the memory 72, thereby implementing at least a portion of the functions of the drive torque calculation unit 3, the operating state calculation unit 4, the torque estimation error model learning unit 5, the conversion unit 6, the error calculation unit 7, the threshold value calculation unit 8, and the collision determination unit 9.

[0101] When at least some of the functions of the drive torque calculation unit 3, the motion state calculation unit 4, the torque estimation error model learning unit 5, the conversion unit 6, the error calculation unit 7, the threshold value calculation unit 8, and the collision determination unit 9 are implemented by the processor 71, the robot collision detection device 1 includes a memory 72 for storing a program that ultimately executes at least some of the program steps executed by the drive torque calculation unit 3, the motion state calculation unit 4, the torque estimation error model learning unit 5, the conversion unit 6, the error calculation unit 7, the threshold value calculation unit 8, and the collision determination unit 9. The program stored in the memory 72 can be said to cause a computer to execute at least a portion of the procedures or methods executed by the drive torque calculation unit 3, the motion state calculation unit 4, the torque estimation error model learning unit 5, the conversion unit 6, the error calculation unit 7, the threshold value calculation unit 8, and the collision determination unit 9.

[0102] The memory 72 is, for example, a non-volatile or volatile semiconductor memory such as RAM (Random Access Memory), ROM (Read Only Memory), flash memory, EPROM (Erasable Programmable Read Only Memory), EEPROM (registered trademark) (Electrically Erasable Programmable Read-Only Memory), a magnetic disk, a floppy disk, an optical disk, a compact disk, a mini disk, or a DVD (Digital Versatile Disk).

[0103] Figure 8This diagram shows a processing circuit 81 in the case where at least a portion of the drive torque calculation unit 3, the motion state calculation unit 4, the torque estimation error model learning unit 5, the conversion unit 6, the error calculation unit 7, the threshold value calculation unit 8, and the collision determination unit 9 included in the robot collision detection device 1 according to Embodiment 1 is implemented by the processing circuit 81. Specifically, at least a portion of the drive torque calculation unit 3, the motion state calculation unit 4, the torque estimation error model learning unit 5, the conversion unit 6, the error calculation unit 7, the threshold value calculation unit 8, and the collision determination unit 9 can be implemented by the processing circuit 81.

[0104] The processing circuit 81 is dedicated hardware and may be, for example, a single circuit, a complex circuit, a programmed processor, a parallel programmed processor, an ASIC (Application Specific Integrated Circuit), an FPGA (Field-Programmable Gate Array), or a combination thereof.

[0105] Part of the driving torque calculation unit 3 , the operation state calculation unit 4 , the torque estimation error model learning unit 5 , the conversion unit 6 , the error calculation unit 7 , the threshold calculation unit 8 , and the collision determination unit 9 may be realized by dedicated hardware different from the remaining parts.

[0106] Regarding the multiple functions of the drive torque calculation unit 3, the operation state calculation unit 4, the torque estimation error model learning unit 5, the conversion unit 6, the error calculation unit 7, the threshold calculation unit 8, and the collision determination unit 9, some of these functions can be implemented by software or firmware, while the remaining functions can be implemented by dedicated hardware. As described above, the multiple functions of the drive torque calculation unit 3, the operation state calculation unit 4, the torque estimation error model learning unit 5, the conversion unit 6, the error calculation unit 7, the threshold calculation unit 8, and the collision determination unit 9 can be implemented by hardware, software, firmware, or a combination thereof.

[0107] At least some of the functions of the drive torque calculation unit 3, motion state calculation unit 4, torque estimation error model learning unit 5, error calculation unit 7, threshold calculation unit 8, and collision determination unit 9 included in the robot collision detection device 1A according to Embodiment 2 can be implemented by a processor executing a program stored in a memory. This memory is the same as memory 72, and this processor is the same as processor 71. At least some of the drive torque calculation unit 3, motion state calculation unit 4, torque estimation error model learning unit 5, error calculation unit 7, threshold calculation unit 8, and collision determination unit 9 can also be implemented by a processing circuit. This processing circuit is the same as processing circuit 81.

[0108] At least some of the functions of the drive torque calculation unit 3, motion state calculation unit 4, torque estimation error model learning unit 5, conversion unit 6, error calculation unit 7, threshold calculation unit 8, and collision determination unit 9 included in the robot collision detection devices according to Embodiments 3, 4, and 7 can be implemented by a processor executing a program stored in a memory. The memory is the same as memory 72, and the processor is the same as processor 71. At least some of the functions of the drive torque calculation unit 3, motion state calculation unit 4, torque estimation error model learning unit 5, conversion unit 6, error calculation unit 7, threshold calculation unit 8, and collision determination unit 9 included in the robot collision detection devices according to Embodiments 3, 4, and 7 can also be implemented by a processing circuit. The processing circuit is the same as processing circuit 81.

[0109] At least some of the functions of the driving torque calculation unit 3B, the motion state calculation unit 4, the torque estimation error model learning unit 5, the conversion unit 6, the error calculation unit 7, the threshold calculation unit 8, the collision determination unit 9, and the parameter identification unit 10 included in the robot collision detection device 1B according to the fifth embodiment can be implemented by a processor executing a program stored in a memory. The memory is the same as the memory 72, and the processor is the same as the processor 71. At least some of the driving torque calculation unit 3B, the motion state calculation unit 4, the torque estimation error model learning unit 5, the conversion unit 6, the error calculation unit 7, the threshold calculation unit 8, the collision determination unit 9, and the parameter identification unit 10 can also be implemented by a processing circuit. The processing circuit is the same as the processing circuit 81.

[0110] At least some of the functions of the driving torque calculation unit 3C, motion state calculation unit 4, torque estimation error model learning unit 5, conversion unit 6, error calculation unit 7, threshold calculation unit 8, collision determination unit 9, and online parameter identification unit 11 included in the robot collision detection device 1C according to Embodiment 6 can be implemented by a processor executing a program stored in a memory. The memory is the same as memory 72, and the processor is the same as processor 71. At least some of the aforementioned driving torque calculation unit 3C, motion state calculation unit 4, torque estimation error model learning unit 5, conversion unit 6, error calculation unit 7, threshold calculation unit 8, collision determination unit 9, and online parameter identification unit 11 can also be implemented by a processing circuit. The processing circuit is the same as processing circuit 81.

[0111] At least some of the functions of the driving torque calculation unit 3, motion state calculation unit 4, torque estimation error model learning unit 5, conversion unit 6, error calculation unit 7, threshold calculation unit 8D, collision determination unit 9, and approximate function learning unit 12 included in the robot collision detection device 1D according to Embodiment 8 can be implemented by a processor executing a program stored in a memory. The memory is the same as memory 72, and the processor is the same as processor 71. At least some of the driving torque calculation unit 3, motion state calculation unit 4, torque estimation error model learning unit 5, conversion unit 6, error calculation unit 7, threshold calculation unit 8D, collision determination unit 9, and approximate function learning unit 12 can also be implemented by a processing circuit. The processing circuit is the same as processing circuit 81.

[0112] At least some of the functions of the driving torque calculation unit 3, motion state calculation unit 4, torque estimation error model learning unit 5E, conversion unit 6, error calculation unit 7, threshold calculation unit 8E, collision determination unit 9, and temperature measurement unit 13 included in the robot collision detection device 1E according to Embodiment 9 can be implemented by a processor executing a program stored in a memory. The memory is the same as memory 72, and the processor is the same as processor 71. At least some of the aforementioned driving torque calculation unit 3, motion state calculation unit 4, torque estimation error model learning unit 5E, conversion unit 6, error calculation unit 7, threshold calculation unit 8E, collision determination unit 9, and temperature measurement unit 13 can also be implemented by a processing circuit. The processing circuit is the same as processing circuit 81.

[0113] The configuration shown in the above embodiment is merely an example, and can be combined with other known technologies, and the embodiments can be combined with each other. Part of the configuration can also be omitted or changed without departing from the scope of the invention.

[0114] Description of the label

[0115] 1. 1A, 1B, 1C, 1D, 1E robot collision detection device, 2. 2A, 2B, 2C, 2D, 2E robot control device, 3. 3B, 3C drive torque calculation unit, 4. action state calculation unit, 5. 5E torque estimation error model learning unit, 5a learning unit, 6. conversion unit, 7. error calculation unit, 8. 8D, 8E threshold calculation unit, 9. collision discrimination unit, 10. parameter identification unit, 11. online parameter identification unit, 12. approximate function learning unit, 13. temperature measurement unit, 71. processor, 72. memory, 81. processing circuit.

Claims

1. A robot collision detection device, characterized in that: have: a driving torque calculation unit that calculates an estimated value of the driving torque of the robot; a torque estimation error model learning unit that learns a difference between a driving torque calculated based on a motor current for driving the robot and the estimated value calculated by the driving torque calculation unit, and a fluctuation in the difference; a threshold value calculation unit that calculates a threshold value based on the torque estimation error model learned by the torque estimation error model learning unit; a collision determination unit that determines a collision between the robot and an object by comparing a difference between the drive torque calculated based on the motor current for driving the robot and the estimated value calculated by the drive torque calculation unit with the threshold value calculated by the threshold value calculation unit; and an action state calculation unit for calculating a state quantity related to the action state of the robot; The torque estimation error model learning unit learns the state amount calculated by the operation state calculation unit as an input signal of a correction function.

2. A robot collision detection device, characterized in that: have: a driving torque calculation unit that calculates an estimated value of the driving torque of the robot; a torque estimation error model learning unit that learns a difference between a driving torque measured by a torque sensor provided on a driving portion of the robot and the estimated value calculated by the driving torque calculation unit, and a fluctuation in the difference; a threshold value calculation unit that calculates a threshold value based on the torque estimation error model learned by the torque estimation error model learning unit; a collision determination unit that determines a collision between the robot and an object by comparing a difference between the driving torque measured by the torque sensor and the estimated value calculated by the driving torque calculation unit with the threshold value calculated by the threshold value calculation unit; and an action state calculation unit for calculating a state quantity related to the action state of the robot; The torque estimation error model learning unit learns the state amount calculated by the operation state calculation unit as an input signal of a correction function.

3. The robot collision detection device according to claim 1 or 2, characterized in that: The torque estimation error model learning unit includes a learning unit using a non-parametric method.

4. The robot collision detection device according to claim 3, characterized in that: The learning unit uses Gaussian process regression.

5. The robot collision detection device according to claim 1 or 2, characterized in that: The drive torque calculation unit further includes a parameter determination unit that determines the values of the parameters of the motion equation used by the drive torque calculation unit based on pre-measured data. The driving torque calculation unit calculates the estimated value using the value determined by the parameter determination unit.

6. The robot collision detection device according to claim 1 or 2, characterized in that: The device further comprises an online parameter determination unit for determining the values of parameters of the motion equation used by the drive torque calculation unit based on the data during the movement of the robot. The driving torque calculation unit uses the value identified by the online parameter identification unit when calculating the estimated value.

7. The robot collision detection device according to claim 1 or 2, characterized in that: The operation state calculation unit calculates the state quantity including information related to the motor speed.

8. The robot collision detection device according to claim 1 or 2, characterized in that: The operation state calculation unit calculates the state quantity including information related to motor acceleration.

9. The robot collision detection device according to claim 1 or 2, characterized in that: The operation state calculation unit calculates the state quantity including a part of the elements of the driving torque of the robot.

10. The robot collision detection device according to claim 1 or 2, characterized in that: further comprising an approximate function learning unit that learns an approximate function based on the torque estimation error model learned by the torque estimation error model learning unit, The threshold value calculation unit calculates the threshold value using the approximate function derived by the approximate function learning unit.

11. The robot collision detection device according to claim 1 or 2, characterized in that: It also has a temperature measuring unit, which measures the temperature. The torque estimation error model learning unit performs learning using the temperature measured by the temperature measuring unit. The threshold value calculation unit calculates the threshold value using the temperature measured by the temperature measurement unit.

Images