A human-robot collaboration method based on motion intention estimation and force feedforward compensation

Abstract

This invention discloses a human-machine collaboration method based on motion intention estimation and force feedforward compensation. It mainly includes a human-machine collaboration system, a human motion intention recognition system, a force feedforward compensation system, and a position inner-loop control system. The human-machine collaboration system includes a collaborative robot and a six-dimensional force sensor; the human motion intention recognition system uses maximum likelihood estimation to obtain the human's motion intention and stiffness during human-machine collaboration and integrates it into impedance control; the force feedforward compensation system designs force feedforward compensation based on the obtained motion intention and stiffness to reduce interaction forces and position tracking fluctuations; the position inner-loop control system uses RBF neural network control for position inner-loop control. This invention has superior effects in reducing human interaction forces, improving position response speed, and reducing position tracking fluctuations in human-machine collaboration.

Description

Technical Field

[0001] This invention belongs to the field of robot control, and more specifically, relates to a human-machine collaboration method based on motion intention estimation and force feedforward compensation. Background Technology

[0002] With the rise of robotics, human-robot collaboration has become central to current robotics development. Human-robot cooperation, a crucial component of human-robot interaction, has found widespread application in high-end manufacturing, medical rehabilitation, and social services. How to integrate the high-precision, high-efficiency, and repetitive performance of robots with the advantages of humans in understanding, perception, and decision-making is also a key research focus for many scholars. Impedance control has established a theoretical foundation for compliant human-robot collaboration. Under impedance control, the robot passively follows the human's movement, requiring the human partner to expend more energy to compensate for the dynamics of the robot and the object to achieve the desired movement. The robot becomes an additional load, and the human partner is prone to fatigue.

[0003] The goal of this invention is to enable robots to actively follow human movement trajectories and reduce human energy input. This raises a thorny issue: how to enable robots to know the target location of humans, that is, how to estimate human movement intentions. Summary of the Invention

[0004] To address the aforementioned problems in human-robot collaboration, this invention proposes a neural impedance control method for human-robot collaboration based on motion intention estimation and force feedforward compensation. This method enables highly compliant human-guided robot collaborative operations. This invention mainly comprises four systems: a human-robot collaboration system, a human motion intention recognition system, a force feedforward compensation system, and a position inner-loop control system.

[0005] The human-machine collaboration system includes a collaborative robot and a six-dimensional force sensor. The six-dimensional force sensor is installed at the end of the collaborative robot. A human partner guides the robot to work collaboratively by dragging the six-dimensional force sensor at the end of the robotic arm. In this invention, the position and speed information of the end of the robotic arm and the human-machine interaction force are available.

[0006] The human motion intention recognition system estimates the human motion intention and limb stiffness based on the position, speed and interaction force information of the robotic arm end effector using maximum likelihood estimation, and integrates the motion intention into impedance control.

[0007] The force feedforward compensation system is designed based on the human movement intention and limb stiffness estimated by maximum likelihood, and the stability of the system is proved by Lyapunov function, further reducing the fluctuation of human-computer interaction force and position response.

[0008] The position inner loop control system uses RBFNN for position inner loop control, and the stability of the control is proved by Lyapunov function. Compared with traditional PID control, the position response is faster.

[0009] The human-machine collaboration system described is primarily designed for a simplified human-machine collaboration scenario, involving slow human-machine interaction. In this scenario, a human hand drags a force sensor at the end of a robotic arm, determining the hand's current position (x) and velocity. The interaction force f can be obtained by further calculation using the position information of the robot encoder, and can be measured by a force sensor.

[0010] This invention acquires the human's motion intention and stiffness in human-machine collaboration and integrates them into impedance control. Force feedforward compensation is designed based on the acquired motion intention and stiffness, and RBFNN control is used for position inner loop control.

[0011] The technical solution adopted by the present invention to achieve the above objectives is as follows:

[0012] A human-computer collaboration method based on motion intention estimation and force feedforward compensation, characterized by the following steps:

[0013] Based on the end-effector state information and interaction force information, the motion intention and limb stiffness of the human are estimated through the motion intention recognition system;

[0014] Based on the estimated human movement intention and limb stiffness, the force feedforward compensation system reduces fluctuations in human-computer interaction force and position response;

[0015] Based on the end-effector state information and interaction force information, the human's motion intention, and the force feedforward compensation obtained from the force feedforward compensation system, the target position of the interaction force is obtained through the robot impedance model.

[0016] Based on the target position generated by the interaction force and the position in the end-effector state information, the robot is controlled through the position inner loop control system to achieve human-robot collaboration.

[0017] The step of estimating a person's movement intention and limb stiffness based on the end-effector state information of the robotic arm through a motion intention recognition system includes the following steps:

[0018] Human limb dynamics model by -K h (xx h ) = f means that x h Let x represent the person's intention to move, and K represent the person's current position. h represents the stiffness of a human limb, and f represents the interaction force;

[0019] Based on the position, velocity, and interaction force of the robotic arm's end effector, maximum likelihood estimation is used to estimate the human's movement intention and limb stiffness, and the estimated limb stiffness is set... And the average stiffness and variance From observation data It is estimated that the movement intention is located in the interval The estimated motion intention is then used in the robot impedance model;

[0020] Where T represents a sampling period, f i Let represent the i-th interaction force, and n represent the number of data points; These represent the maximum and minimum values ​​for estimating human limb stiffness, respectively.

[0021] The process of reducing human-computer interaction force and position response fluctuations through a force feedforward compensation system based on estimated human movement intentions and limb stiffness includes the following steps:

[0022] The force feedforward compensation system consists of This indicates that a = -K h If b > 0, the required variable for feedforward force compensation is the human intention x. h Upper limb stiffness K h The intentions of the people are estimated respectively. and the stiffness of the human upper limb express;

[0023] Where Δf represents feedforward force compensation, and a and b are coefficients.

[0024] The step of controlling the robot through a position inner loop control system based on the target position generated by the interaction force and the position in the end-effector state information includes the following steps:

[0025] The position inner loop control system is a robotic arm dynamics construction controller designed for Cartesian space;

[0026] The dynamics of the robotic arm in Cartesian space are as follows:

[0027]

[0028] in, These represent the current position of the person, i.e., the position, velocity, and acceleration of the robotic arm in Cartesian space. It is a symmetric positive definite inertial matrix. The matrix of Coriolis force and centrifugal force. For the gravity matrix, This represents the total control input, and m represents the number of degrees of freedom of the robotic arm. Let h represent the interaction force between the robot and the object, and h denote the dimension of the interaction force.

[0029] The controller is:

[0030] Where u represents the control quantity output by the controller, and K v Represents a positive constant. The input to the neural network is represented by the radial basis function, r represents the error function, and f represents the interaction force. This represents the estimated value of the weight parameter W for the RBF neural network;

[0031]

[0032] T represents a sampling period, and L is a positive definite matrix.

[0033] A human-machine collaboration method based on motion intention estimation and force feedforward compensation guides the robot to work collaboratively by dragging a six-dimensional force sensor at the end of the robotic arm.

[0034] A human-machine collaborative system based on motion intention estimation and force feedforward compensation includes:

[0035] The motion intention recognition system is used to estimate a person's motion intention and limb stiffness based on the end-effector state information and interaction force information of the robotic arm.

[0036] The force feedforward compensation system is used to reduce fluctuations in human-computer interaction force and position response based on the estimated human movement intention and limb stiffness.

[0037] The position inner loop control system is used to control the robot based on the target position generated by the interaction force and the position in the end-effector state information, so as to realize human-robot collaboration. The target position generated by the interaction force is obtained by using the robot impedance model based on the end-effector state information, interaction force information, human movement intention and force feedforward compensation obtained from the force feedforward compensation system.

[0038] A human-machine collaboration system based on motion intention estimation and force feedforward compensation includes a collaborative robot and a six-dimensional force sensor. The six-dimensional force sensor is installed at the end of the collaborative robot, and a human guides the robot to work collaboratively by dragging the six-dimensional force sensor at the end of the robotic arm.

[0039] A human-computer collaboration device based on motion intention estimation and force feedforward compensation includes a memory and a processor; the memory is used to store a computer program; the processor is used to implement the human-computer collaboration method based on motion intention estimation and force feedforward compensation when the computer program is executed.

[0040] A computer-readable storage medium storing a computer program that, when executed by a processor, implements the aforementioned human-computer collaboration method based on motion intention estimation and force feedforward compensation.

[0041] The present invention has the following beneficial effects and advantages:

[0042] In human-robot collaboration, the method of this invention can enable robots to better adapt to human partners. Estimating human intentions can reduce human energy consumption and interaction forces. Feedforward force compensation can reduce interaction forces and position tracking fluctuations. RBF neural networks can improve position response speed. Attached Figure Description

[0043] Figure 1 This is a schematic diagram of a typical human-machine collaboration system considered in this invention;

[0044] Figure 2 This is a simplified schematic diagram of a human-computer collaboration system as conceived in this invention;

[0045] Figure 3 This is a control block diagram of the present invention;

[0046] Figure 4 It is a graph showing the position of the x-axis;

[0047] Figure 5 It is a curve showing the position of the y-axis;

[0048] Figure 6 This is a simulation diagram of the interaction force along the x-axis;

[0049] Figure 7 This is a simulation diagram of the y-axis interactive force. Detailed Implementation

[0050] To enable those skilled in the art to better understand the technical solutions in this application, the technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments in this application, all other embodiments obtained by those of ordinary skill in the art without creative effort should fall within the scope of protection of this application.

[0051] This invention is aimed at such as Figure 1 The typical human-robot collaboration scenario shown involves any robot collaboratively moving a table. In this invention, we consider the table as a point mass, and the robot follows the human partner's movements. Therefore, the interaction force exerted by the human on the table is equal to the interaction force exerted by the robot. This human-robot collaboration scenario can then be approximated by a human hand dragging a force sensor at the end of a robotic arm, such as... Figure 2As shown, the current position x and velocity of the hand are... The interaction force f can be obtained by further calculation using the position information of the robot encoder, and can be measured by a force sensor.

[0052] 1. The first step is to provide the required human-robot collaboration system, including a collaborative robot and a six-dimensional force sensor. The six-dimensional force sensor is installed at the end of the collaborative robot. Consider a simplified human-robot collaboration system in which a human partner guides the robot to work collaboratively by dragging the six-dimensional force sensor at the end of the robotic arm. In this patent, the position and speed information of the end of the robotic arm and the human-robot interaction force are available.

[0053] 2. The second step is to establish a human motion intention recognition system. Based on the position, speed and interaction force information of the robotic arm end effector, the human motion intention and limb stiffness are estimated using maximum likelihood estimation, and the motion intention is integrated into impedance control.

[0054] 3. The third step is to establish a force feedforward compensation system. Based on the human movement intention and limb stiffness estimated by maximum likelihood, a force feedforward compensation is designed, and the stability of the system is proved by Lyapunov function, which further reduces the fluctuation of human-computer interaction force and position response.

[0055] 4. The fourth step is to establish a position inner-loop control system, using RBFNN for position inner-loop control, and to prove the stability of the control through Lyapunov functions. Compared with traditional PID control, the position response is faster.

[0056] 5. Further, the human motion intention recognition system in step two is implemented as follows: This paper considers a slow human-computer interaction scenario, and the human limb dynamics model is represented by equation (1):

[0057] -K h (xx h )=f (1)

[0058] Here x h Defined as a person's intention, x is the person's current position, and K h The force represents the stiffness of the human limbs, and the interaction force is represented by f. The current position (x) and velocity of the human hand are also considered. The position information from the robot encoder can be further calculated, and the interaction force f can be measured by a force sensor. Therefore, the stiffness of the human upper limb can be approximately expressed as:

[0059]

[0060] T represents a sampling period, and the location of a person's intention can be approximated as:

[0061]

[0062] Assuming that the stiffness of the human upper limbs follows a normal distribution, that is The likelihood function constructed from the observed data in equation (2) is:

[0063]

[0064] Take the logarithm of equation (4):

[0065]

[0066] Find equation (6) for μ, σ 2 Take the partial derivatives and set them equal to 0:

[0067]

[0068] Solving for:

[0069]

[0070] Therefore, the estimated upper limb stiffness of the human body was... lie in to The interval, i.e.:

[0071]

[0072] Then the person's intention Located in the interval For programming convenience, only the estimated average value will be used later.

[0073] 6. Further, the impedance control corresponds to formula (10):

[0074]

[0075] Where M d B d ,K d These are the robot's inertia, damping, and stiffness coefficients, which are manually set. r x is the robot's reference position. d The target position is determined by the interaction force f. When there is no interaction force f, the robot remains at x. r At point x, when an interaction force f exists, the robot moves from x... r Move from point x to the target position x generated by the interaction force. d Place.

[0076] The location of the person's intention estimated by maximum likelihood The reference trajectory x is set as the robot's impedance control. r That is, let x r =x h Integrating human intent into impedance control creates impedance control that takes human intent into account.

[0077] 7. Further, the force feedforward compensation system in step three is specifically implemented as follows:

[0078] To minimize fluctuations in interaction force and position tracking, a compensation force is defined.

[0079]

[0080] Assume the robot's actual trajectory can track the target position generated by the interaction force, i.e., x = x d Then we have:

[0081]

[0082]

[0083] The reference position x for robot impedance control r Set as the person's intended location x h It can reduce the interaction force, making x r =x h And define e = xx h Then equation (13) can be rewritten as:

[0084]

[0085] have

[0086]

[0087] Selecting Lyapunov candidate functions:

[0088]

[0089] Then, taking the derivative of V1 with respect to time, we have:

[0090]

[0091] Substituting equation (15) into equation (17), we have

[0092]

[0093] Let a = -K h If b > 0, then equation (18) is true.

[0094]

[0095] The variables required for feedforward force compensation can all be estimated from human intent. and the stiffness of the human upper limb Therefore, as time approaches infinity, according to the defined Lyapunov function, the robot's actual position and speed can track the human's intended position and speed, and the error converges to 0.

[0096] 8. Further, the position inner loop control system in step four is implemented as follows:

[0097] In order for the robotic arm to accurately track the target trajectory x generated by the interaction force d To improve position response speed, an RBF neural network is used for inner-loop position tracking control. The robotic arm dynamics in Cartesian space are as follows:

[0098]

[0099] in, These represent the position, velocity, and acceleration of the robotic arm in Cartesian space, respectively. It is a symmetric positive definite inertial matrix. The matrix of Coriolis force and centrifugal force. For the gravity matrix, This represents the total control input, and m represents the number of degrees of freedom of the robotic arm. Let h represent the interaction force between the robot and the object, and h denote the dimension of the interaction force.

[0100] Define the desired target position as x d Let x be the actual position of the robotic arm, then the tracking error is: z = x d -x, defines the error function:

[0101]

[0102] If Λ is a positive number, then we have

[0103]

[0104] Differentiating equation (22) and according to equation (20), we have

[0105]

[0106] For the unknown dynamic model in equation (23):

[0107]

[0108] An RBF neural network is used for approximation, where the input of the RBF neural network is:

[0109]

[0110] Then there is

[0111]

[0112] Where ε represents the approximation error of the neural network, let

[0113]

[0114] in This represents an estimate of f(Z). Let W represent an estimate of W.

[0115] The controller designed is as follows:

[0116]

[0117] Where K v It represents a positive constant.

[0118] Then equation (23) can be expressed as:

[0119]

[0120] Get Lyapunov function

[0121]

[0122] V2 Find the derivative with respect to time.

[0123]

[0124] Substituting equation (29) into the above equation, we have:

[0125]

[0126] make Then we have:

[0127]

[0128] because Therefore, the weight adjustment formula for the RBF neural network is:

[0129]

[0130] 9. Further, the control block diagram of the present invention is as follows: Figure 3 As shown, based on the position, speed and interaction force information of the robotic arm end effector, maximum likelihood estimation is used to obtain the human's motion intention and stiffness in human-robot collaboration and integrate them into impedance control. Force feedforward compensation is designed based on the obtained motion intention and stiffness to reduce the fluctuation of interaction force and position tracking. In order to improve the position tracking accuracy and response speed, RBFNN control is used for position inner loop control.

[0131] Next, we will demonstrate the Simulink simulation. We will perform simulations in both the x and y directions, setting the reference position for robot impedance control at x. r Set as the person's intended location x h This can reduce interaction forces, which we define as impedance control considering intent. The human upper limb model in the x-direction is set to f. x = -25(x-0.45), the human upper limb model in the y direction is set to f y = -30(x-0.5), meaning the person's intended location is (0.45m, 0.5m). The RBF neural network has 7 hidden layers, the center point of the radial basis function is set to [-1.5, 1.5], and the initial weights are 0. K v =15, in feedforward force compensation b=3N(m / s) -1 The PD controller used in the comparative experiment was set to... Control gain set to

[0132] A comparative experiment was conducted using four methods: impedance control, impedance control considering intent, impedance control considering intent plus force feedforward compensation (the first three methods use PD control in the inner loop of the position), and the method presented in this paper. The position curves in the x-axis and y-axis directions are shown below. Figure 4 , 5 As shown, since the selected robot impedance equation has a stiffness term, simple impedance control cannot move the robot to the intended position. Integrating the estimated human intention into the robot impedance equation constitutes an impedance control that takes intention into account. The robot can move to the target position, but there are large fluctuations. Adding force feedforward to the impedance control that takes intention into account can significantly improve the position fluctuations and improve human-robot cooperation. However, the inner loop of PD control is relatively slow. The method in this paper can move to the intended position more quickly and with smaller fluctuations.

[0133] Figure 6 , 7 The x-axis and y-axis interaction forces of the four methods are shown respectively. It can be seen that the impedance control has a large interaction force because it has not reached the target position. The interaction force of the method that integrates the intended position into the impedance control can converge to 0. However, the method in this paper has a faster convergence speed and smaller interaction force fluctuation.

Claims

1. A human-computer collaboration method based on motion intention estimation and force feedforward compensation, characterized in that, Includes the following steps: Based on the end-effector state information and interaction force information, the motion intention and limb stiffness of the human are estimated through the motion intention recognition system; Based on the estimated human movement intention and limb stiffness, the fluctuation of human-computer interaction force and position response is reduced by the force feedforward compensation system. Based on the end-effector state information and interaction force information, the human's motion intention, and the force feedforward compensation obtained from the force feedforward compensation system, the target position of the interaction force is obtained through the robot impedance model. Based on the target position generated by the interaction force and the position in the end-effector state information, the robot is controlled through the position inner loop control system to achieve human-robot collaboration. The step of estimating a person's movement intention and limb stiffness based on the end-effector state information of the robotic arm through a motion intention recognition system includes the following steps: Human limb dynamics model is composed of express, Human movement intentions x As a person's current position, Indicates the stiffness of human limbs. f Indicates interactive force; Based on the position, velocity, and interaction force of the robotic arm's end effector, maximum likelihood estimation is used to estimate the human's motion intention and limb stiffness, and the estimated limb stiffness is set as follows: And the estimated average human limb stiffness and variance Estimate from observation data The intention to move is located in the interval The estimated motion intention is then used in the robot's impedance model. Where T represents a sampling period, Let represent the i-th interaction force, and n represent the number of data points; , These represent the estimated maximum and minimum values ​​of human limb stiffness, respectively.

2. The human-computer collaboration method based on motion intention estimation and force feedforward compensation according to claim 1, characterized in that, The process of reducing human-computer interaction force and position response fluctuations through a force feedforward compensation system based on estimated human movement intentions and limb stiffness includes the following steps: The force feedforward compensation system consists of It means, and , The required variable for feedforward force compensation is the human motion intention. Human limb stiffness The intentions of the people being estimated respectively And estimated human limb stiffness express; in, Indicates feedforward force compensation. , All are coefficients.

3. The human-computer collaboration method based on motion intention estimation and force feedforward compensation according to claim 1, characterized in that, The step of controlling the robot through a position inner loop control system based on the target position generated by the interaction force and the position in the end-effector state information includes the following steps: The position inner loop control system is a robotic arm dynamics construction controller designed for Cartesian space; The dynamics of the robotic arm in Cartesian space are as follows: ； in, These represent the current position of the person, i.e., the position, velocity, and acceleration of the robotic arm in Cartesian space. It is a symmetric positive definite inertial matrix. The matrix of Coriolis force and centrifugal force. For the gravity matrix, This represents the control quantity output by the controller, and m represents the number of degrees of freedom of the robotic arm. Let h represent the interaction force between the robot and the object, and h denote the dimension of the interaction force. The controller is: ; in, Represents a positive constant. The input to the neural network is represented by the radial basis functions. Represents the error function; This represents the estimated value of the weight parameter W for the RBF neural network; ； It is a positive definite matrix.

4. The human-computer collaboration method based on motion intention estimation and force feedforward compensation according to claim 1, characterized in that, The robot is guided to work collaboratively by dragging a six-dimensional force sensor at the end of the robotic arm.

5. A human-computer collaboration system based on motion intention estimation and force feedforward compensation, the system being used to implement the human-computer collaboration method based on motion intention estimation and force feedforward compensation as described in any one of claims 1-4, characterized in that, include: The motion intention recognition system is used to estimate a person's motion intention and limb stiffness based on the end-effector state information and interaction force information of the robotic arm. Force feedforward compensation system is used to reduce fluctuations in human-computer interaction force and position response based on the estimated human movement intention and limb stiffness. The position inner loop control system is used to control the robot based on the target position generated by the interaction force and the position in the end-effector state information, so as to realize human-robot collaboration. The target position generated by the interaction force is obtained by using the robot impedance model based on the end-effector state information, interaction force information, human movement intention and force feedforward compensation obtained from the force feedforward compensation system.

6. A human-machine collaborative system based on motion intention estimation and force feedforward compensation according to claim 5, characterized in that, It includes a collaborative robot and a six-dimensional force sensor. The six-dimensional force sensor is installed at the end of the collaborative robot, and a person guides the robot to work collaboratively by dragging the six-dimensional force sensor at the end of the robotic arm.

7. A human-machine collaborative device based on motion intention estimation and force feedforward compensation, characterized in that, It includes a memory and a processor; the memory is used to store a computer program; the processor is used to implement, when executing the computer program, a human-computer collaboration method based on motion intention estimation and force feedforward compensation as described in any one of claims 1-4.

8. A computer-readable storage medium, characterized in that, The storage medium stores a computer program, which, when executed by a processor, implements a human-computer collaboration method based on motion intention estimation and force feedforward compensation as described in any one of claims 1-4.

Images