Robot compliance force control interaction method, system and equipment facing collaborative sewing scene and storage medium

Abstract

The invention discloses a robot compliance force control interaction method, system and device for a collaborative sewing scene and a storage medium. The method comprises the steps of setting a preset period of a contact force track, a speed track and a pose track; before the task is started, collecting data of a plurality of groups of force sensors, completing dynamic gravity compensation, and calculating real contact force in real time; in the task process, the real contact force is continuously obtained, the speed deviation and the pose deviation are obtained, the real contact force, the speed deviation and the pose deviation are input into a compliant controller based on an admittance model, and the target pose of the mechanical arm in the next step is dynamically adjusted; the compliant cooperation is repeatedly executed until the whole cooperation sewing task is completed; after the task is finished, a parameter optimization algorithm based on smooth iterative learning is constructed, the real contact force trajectory, the speed trajectory and the pose trajectory are used as input, the smooth iterative learning algorithm autonomously optimizes parameters of a compliant controller, and the optimal force-position trajectory tracking performance is achieved. According to the invention, a safe and stable robot cooperation sewing task is realized.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of robot intelligent control, and particularly relates to a robot compliance force control interaction method, system and device for a flexible fabric autonomous sewing scene and a storage medium. BACKGROUND

[0002] The intelligent transformation of the textile manufacturing industry is accelerating the application of robot technology in the sewing process. However, the operation mode of the traditional industrial robot based on rigid position control faces significant challenges in dealing with the special scene of fabric sewing: the inherent flexibility and nonlinear deformation characteristics of the fabric make the force-deformation coupling effect a key factor affecting the sewing quality, often leading to uneven stitching, fabric wrinkling and other process defects. Especially in the collaborative sewing scene, the dynamic interaction between the robot and the environment puts higher requirements on the real-time force control accuracy and compliance.

[0003] The current technology faces two major bottlenecks: one is the insufficient trajectory self-adaptability, and the rigid trajectory is difficult to cope with the nonlinear deformation of the fabric, which is easy to produce tracking deviation or fabric damage; the second is the low safety of collaborative interaction, and there is a lack of efficient force / position hybrid control strategy, which cannot respond to sudden contact force changes in real time. The existing impedance control or PID force control method has response lag or overshoot problems in complex scenes, and fixed control parameters are difficult to adapt to the diversified sewing task requirements. It is urgent to develop a high-precision, self-adaptive compliant force control method to break through the key technical bottlenecks of collaborative sewing. SUMMARY

[0004] The purpose of the present application is to provide a robot compliant force control interaction method for collaborative sewing scene, to realize good trajectory tracking performance and collaborative sewing effect, and to have strong environmental adaptability.

[0005] Technical solution: The method disclosed by the present application comprises the following steps:

[0006] presetting a desired contact force trajectory, a robot speed trajectory and a pose trajectory;

[0007] Before the task starts, a plurality of groups of force sensor data are collected and dynamic gravity compensation is completed, and the real contact force between the end effector and the sewing platform is calculated in real time;

[0008] During the task, the real contact force is continuously obtained, the desired robot speed trajectory and pose trajectory are subtracted from the real robot speed trajectory and pose trajectory to obtain speed deviation and pose deviation, and then the real contact force, speed deviation and pose deviation are input into a compliance controller based on a mobility model to dynamically adjust the target pose of the mechanical arm in the next step, thereby realizing the compliant collaboration of the robot; the compliant collaboration is repeatedly performed until the entire collaborative sewing task is completed;

[0009] After the task, a parameter optimization algorithm based on smooth iterative learning is constructed, which takes the recorded real contact force trajectory, robot speed trajectory and pose trajectory as input, and the smooth iterative learning algorithm autonomously optimizes the compliant controller parameters to achieve the best force and pose trajectory tracking performance.

[0010] Further, a plurality of force sensor data are collected and dynamic gravity compensation is completed, and the real contact force f(t) between the end effector and the sewing platform is calculated in real time, specifically including:

[0011] The robot arm is placed in a free state at a static position, i.e., no external force contact, and then the force sensor collects and records the original data f s (t) of the force sensor under N different static poses, N≥3; the gravity value G of the end effector is measured, and the gravity component f g (t) in the force sensor coordinate system is calculated according to the pose information of the end effector; the least square method is used to calculate the zero point data f0 of the force sensor; the real contact force f(t) is calculated as f s (t)-f0-f g (t).

[0012] Further, the real contact force, speed deviation and pose deviation are input into the compliant controller based on the admittance model, and the next target pose of the robot arm is dynamically adjusted; including the following steps:

[0013] (1) Construct a parameterized robot compliant controller:

[0014]

[0015] wherein, represents the acceleration increment of the robot output by the compliant controller, and x e (t) represent the speed deviation and pose deviation of the robot respectively, f d (t) and f(t) represent the expected contact force and the real contact force respectively, m(t), b(t) and k(t) are all time-varying parameter quantities, representing the virtual inertia, damping and stiffness parameters of the expected dynamic interaction process respectively;

[0016] (2) The compliant controller updates the next robot target pose x(t+1) in real time:

[0017] According to the acceleration increment output by the compliant controller, the adjustment amount of speed and pose is obtained by integration, so as to calculate the next robot target pose x(t+1), which is expressed as:

[0018]

[0019] x t(t+1) = x(t+1) - x e (t+1)

[0020] where, and x e (t+1) respectively represent the robot velocity increment and pose increment at t+1 time, x(t+1) and x t (t+1) respectively represent the real robot pose and target pose at t+1 time, and△t is the sampling time.

[0021] Further, a parameter optimization algorithm based on smooth iterative learning is constructed, which takes the recorded real contact force trajectory, robot velocity trajectory and pose trajectory as input, and the smooth iterative learning algorithm optimizes the compliant controller parameters autonomously to achieve the best force-position trajectory tracking performance, including the following steps:

[0022] (1) Construct the state space equation of the parameterized compliant controller, and convert it into a discrete expression form to assist in deriving the iterative learning controller;

[0023] (2) Calculate the controller output at the next time according to the discrete expression of the robot compliant controller, and derive the controller error correction expression;

[0024] (3) Design an adaptive iterative learning update law, and construct an environment stiffness estimation algorithm based on data driving;

[0025] (4) Design a smoothing update method based on the bilinear interpolation principle to enhance the continuity of the iterative process, and combine it with the adaptive iterative learning update law to form a parameter optimization algorithm based on smooth iterative learning.

[0026] Further, step (1) specifically includes the following steps:

[0027] (11) Let the environment model of the interaction between the robot end effector and the sewing platform be f(t) = k c (t)x(t), and establish the state space equation of the control system:

[0028]

[0029] where, f(t), f d (t) are the real contact force and expected contact force trajectory respectively, k c (t) represents the environment stiffness, x(t), x are the real robot pose trajectory and velocity trajectory respectively, and x e (t+1) respectively represent the robot velocity increment and pose increment at t+1 time, m(t), b(t), k(t) are all time-varying parameter quantities, representing the virtual inertia, damping and stiffness parameters of the expected dynamic interaction process; xd (t), are the desired robot pose trajectory and velocity trajectory, respectively, x e (t), are the pose error and velocity error, respectively, and△t is the sampling time;

[0030] (12) Convert the state-space equation of the robot compliant controller into a typical discrete expression:

[0031] z(t+1) = (βu T (t)A+B)(r(t)-y(t))

[0032] y(t) = C(t)z(t) + D(t)r(t)

[0033] where z(t+1) is the state vector composed of the velocity error and the pose error at t+1, β, A and B are intermediate variables, u T (t) is the transpose of the vector of controller parameters at t, C(t) is the matrix of environmental stiffness parameters at t, D(t) is the matrix of intermediate variables at t, z(t) is the state vector composed of the velocity error and the pose error at t, y(t) is the output vector composed of the target velocity, pose and contact force at t, and r(t) is the input vector composed of the desired contact force and robot motion trajectory at t.

[0034] Further, step (2) specifically comprises the following steps:

[0035] (21) Let the compliant controller parameter matrix U(t) = βu T (t)A+E, β, A and E are intermediate variables, and according to the discrete expression of the robot compliant controller, the output vector expression at t+1 is:

[0036] y(t+1) = U(t)(r(t)-y(t))+D(t+1)r(t+1)

[0037] where y(t+1) is the output vector composed of the target velocity, pose and contact force at t+1, D(t+1) is the matrix of intermediate variables at t+1, r(t) is the input vector composed of the desired contact force and robot motion trajectory at t, y(t) is the output vector composed of the target velocity, pose and contact force at t, and r(t+1) is the vector composed of the desired contact force and robot motion trajectory at t+1;

[0038] (22) Construct an error correction term △U(t), after the error correction term is added, the output of the system at t+1 is compensated to the reference value; construct the output error vector e(t+1) of the controller at t+1, r(t+1)-y(t+1), then according to the system output vector expression and the error correction term:

[0039] y(t+1)+e(t+1)=C(t+1)(U(t)+△U(t))e(t)+D(t+1)r(t+1)

[0040] Wherein, C(t+1) is the environmental stiffness parameter matrix at t+1, e(t) is the output vector error of the controller at t;

[0041] Subtract the above two formulas and use the concept of matrix generalized inverse, obtain the error correction expression:

[0042] △U(t)=C + (t+1)e(t+1)e + (t)

[0043] Wherein, C + (t+1) is the generalized inverse of the environmental stiffness parameter matrix at t+1, e + (t) is the generalized inverse of the output error vector of the controller at t.

[0044] Further, step (3) specifically includes the following steps:

[0045] (31) According to the characteristics of continuously iterating to approach the convergence region according to the iterative learning algorithm, multiply △U(t) expression by learning rate α to construct the parameter update law of the robot compliant controller based on iterative learning as:

[0046]

[0047] Wherein, U k+1 (t) is the compliant controller parameter matrix of the k+1th iteration, U k (t) is the compliant controller parameter matrix of the kth iteration, △U k (t) is the error correction term of the kth iteration, is the generalized inverse of the environmental stiffness parameter matrix of the kth iteration, e k (t+1) is the output error vector of the controller of the kth iteration, is the generalized inverse of the output error vector of the controller of the kth iteration;

[0048] (32) Construct an adaptive learning rate α, and its adaptive expression is:

[0049]

[0050] where |e(t)| is the norm of the output error vector of the controller at the current iteration round t;

[0051] (33) In the parameter update law of the iterative learning based robot compliant controller, C(t) contains the unknown environmental stiffness parameter matrix, and the corresponding estimation matrix is constructed as

[0052]

[0053] where y k-1 (t) is the output vector composed of the target velocity, pose and contact force of the k-1th iteration, z k-1 (t) is the state vector composed of the velocity deviation and pose deviation of the k-1th iteration, is the intermediate variable matrix, and μ is the weight factor, is the estimated environmental stiffness parameter matrix of the k-1th iteration, and r(t) is the input vector composed of the desired contact force and the robot motion trajectory at t.

[0054] Let The estimated environmental stiffness parameter is calculated as

[0055]

[0056] where T D is the intermediate variable matrix, 0<η≤2 is the step factor, and μ is the weight factor.

[0057] Further, step (4) specifically comprises the following steps:

[0058] (41) Assuming that the output of the control system at time t is y(t), the leftmost control system output is y(t-1), and the rightmost control system output is y(t+1), it is obvious that there are error correction terms △U k (t-1) and △U k (t+1) in the kth iteration process, then a smoothing update method of the error correction term is designed based on the bilinear interpolation principle, and the expression is:

[0059]

[0060] where U k+1 (t) is the compliant controller parameter matrix of the k+1th iteration, and U k (t) is the compliant controller parameter matrix of the kth iteration.

[0061] (42) Similar to the smoothing update process of the error correction term, the smoothing update expression of the environmental stiffness parameter is:

[0062]

[0063] where C k+1 (t) is the environment stiffness parameter matrix of the k+1th iteration, C k (t) is the environment stiffness parameter matrix of the kth iteration, △C k (t+1) is the environment stiffness parameter matrix deviation of the kth and k-1th iterations at time t+1, △C k (t-1) is the environment stiffness parameter matrix deviation of the kth and k-1th iterations at time t-1;

[0064] (43) The complete parameter optimization algorithm based on smooth iterative learning is:

[0065] z k (t+1) = U k (t)(r(t) - y k (t))

[0066] y k (t) = C(t)z k (t) + D(t)r(t)

[0067]

[0068] where z k (t+1) is the state vector composed of the velocity deviation and the pose deviation of the kth iteration at time t+1, z k (t) is the state vector composed of the velocity deviation and the pose deviation of the kth iteration at time t, y k (t) is the output vector composed of the target velocity, the pose and the contact force of the kth iteration at time t, U k-1 (t) is the compliant controller parameter matrix of the k-1th iteration at time t, and α is an adaptive learning rate, is the generalized inverse of the environment stiffness parameter matrix of the k-1th iteration at time t+1, e k-1 (t+1) is the output error vector of the controller of the k-1th iteration, is the generalized inverse of the output error vector of the controller of the k-1th iteration, is the estimated environment stiffness parameter matrix of the kth iteration, y k-1 (t) is the output vector composed of the target velocity, the pose and the contact force of the k-1th iteration, z k-1 (t) is the state vector composed of the velocity deviation and the pose deviation of the k-1th iteration, is the estimated environment stiffness parameter matrix of the k-1th iteration, T D is an intermediate variable matrix, 0<η≤2 is a step factor, and μ is a weight factor.

[0069] The system comprises:

[0070] The force-position operation trajectory preset unit is configured to preset the expected contact force trajectory, the robot speed trajectory and the pose trajectory;

[0071] The real contact force calculation unit is configured to collect a plurality of groups of force sensor data and complete dynamic gravity compensation before the task starts, and calculate the real contact force between the end effector and the sewing platform in real time;

[0072] The cooperative sewing unit is configured to continuously acquire the real contact force during the task, subtract the real robot speed trajectory and the pose trajectory from the expected robot speed trajectory and the pose trajectory to obtain the speed deviation and the pose deviation, then input the real contact force, the speed deviation and the pose deviation into the compliance controller based on the admittance model, dynamically adjust the target pose of the mechanical arm in the next step, so as to realize the compliant cooperation of the robot; and repeat the compliant cooperation until the entire cooperative sewing task is completed.

[0073] The parameter optimization unit is configured to, after the task is completed, construct a parameter optimization algorithm based on smooth iterative learning, take the recorded real contact force trajectory, the robot speed trajectory and the pose trajectory as input, and autonomously optimize the parameters of the compliant controller by using the smooth iterative learning algorithm, so as to realize the best force-position trajectory tracking performance.

[0074] The electronic device comprises a memory, a processor and a computer program / instruction stored on the memory and executable on the processor, and the computer program / instruction is executed by the processor to realize the steps of the robot compliant force control interaction method for the cooperative sewing scene.

[0075] Advantages: Compared with the prior art, the significant technical effects of the present application are as follows: a complete set of effective and universal robot compliant force control interaction algorithm can be provided, the performance problem of force-position trajectory tracking and the safety interaction problem of the robot and the environment in the cooperative sewing process can be effectively solved, the autonomous optimization of the robot compliant control parameters in the sewing process can be realized, the tracking performance in the cooperative process and the actual sewing effect can be improved, and the present application can be widely applied to various production and manufacturing processes, and has good practicability and application prospect. BRIEF DESCRIPTION OF DRAWINGS

[0076] Figure 1 FIG. 1 is a schematic diagram of the robot compliant force control interaction process of the present application;

[0077] Figure 2 FIG. 5 is a variable parameter compliant control structure diagram based on iterative learning of the present application. DETAILED DESCRIPTION

[0078] The application will be described further below with reference to the drawings and specific embodiments. Additional aspects and advantages of the application will be described in the following description, will become apparent from the following description, or will be learned through practice of the application.

[0079] The application proposes a high-precision and adaptive robot soft force control interaction method for collaborative sewing scenarios to solve the key technical problem of force-position collaborative control in collaborative sewing and provide reliable technical support for intelligent upgrading of the textile industry. The method abstracts the robot collaborative sewing process as a tracking problem of contact force trajectory and robot motion trajectory, reconstructs a soft controller through a parameterization method, and proposes a parameter optimization method based on iterative learning to obtain optimal control parameters in force / position operation, thereby achieving good trajectory tracking performance and collaborative sewing effect and having strong environmental adaptability.

[0080] As shown in Figure 1 , the method of the application has the following specific steps:

[0081] S1, presetting an expected contact force trajectory and a robot motion trajectory, including giving an expected contact force f d (t) between the end effector and the sewing platform; giving an expected speed and pose trajectory x d (t) of the robot in the sewing task.

[0082] S2, before the task starts, collecting multiple sets of force sensor data and completing dynamic gravity compensation, and calculating the real contact force f(t) between the end effector and the sewing platform in real time, specifically including:

[0083] The force sensor is installed between the flange of the robot arm and the end effector, the robot arm is placed in a free state (i.e. no external force contact) at a static position, and then the original data f s (t) of the force sensor at N (N≥3) different static poses are collected and recorded; the gravity value G of the end effector is measured, the gravity component f g (t) in the force sensor coordinate system is calculated according to the pose information of the end effector; the least square method is used to calculate the zero point data f0 of the force sensor; and the real contact force f(t) = f s (t)-f0-f g (t) is calculated.

[0084] S3, in the task process, the real contact force f(t) is continuously obtained, the expected robot speed trajectory and the pose trajectory x d (t) are subtracted from the real robot speed trajectory and the pose trajectory x(t) to obtain the speed deviation and the pose deviation x e(t) = x d (t) - x(t), then the real contact force f(t), velocity deviation and pose deviation x e (t) input into the compliance controller based on the mobility model, dynamically adjust the next step of the robot target pose x(t+1), so as to realize the compliant cooperation of the robot.

[0085] Further, the step S3 comprises the following steps:

[0086] S31, constructing a parameterized robot compliance controller:

[0087]

[0088] wherein, represents the robot acceleration increment output by the compliance controller, and x e (t) respectively represent the velocity deviation and pose deviation of the robot, f d (t) and f(t) respectively represent the expected contact force and the real contact force, m(t), b(t), k(t) are all time-varying parameter quantities, respectively representing the virtual inertia, damping and stiffness parameters of the expected dynamic interaction process.

[0089] S32, the compliance controller updates the next step of the robot target pose x(t+1) in real time:

[0090] According to the acceleration increment output by the compliance controller, the adjustment amount of velocity and pose is obtained by integration, so as to calculate the next step of the robot target pose x(t+1), which is expressed by the formula:

[0091]

[0092] x t (t+1) = x(t+1) - x e (t+1)

[0093] wherein, and x e (t+1) respectively represent the robot velocity increment and pose increment at t+1, x(t+1) and x t (t+1) respectively represent the real pose and target pose of the robot at t+1, and △t is the sampling time.

[0094] S4, repeating the compliant cooperation operation of step S3 until the whole cooperation sewing task is completed.

[0095] S5, after the task is completed, a parameter optimization algorithm based on smooth iterative learning is constructed, the recorded real contact force trajectory, robot speed trajectory and pose trajectory are taken as inputs, the smooth iterative learning algorithm autonomously optimizes the compliant controller parameters, and the best force and position trajectory tracking performance is realized. Specifically, the following steps are included:

[0096] S51, the state space equation of the parameterized compliant controller is constructed, and is converted into a discrete form to assist in deriving the iterative learning controller;

[0097] Further, the step S51 includes the following steps:

[0098] S511, the environment model of the robot end effector interacting with the sewing platform is f(t)=k c (t)x(t), the state space equation of the control system can be established:

[0099]

[0100] Wherein, k c (t) represents the stiffness of the environment;

[0101] S512, the state space equation of the robot compliant controller is converted into a typical discrete expression:

[0102] z(t+1)=(βu T (t)A+B)(r(t)-y(t))

[0103] y(t)=C(t)z(t)+D(t)r(t)

[0104] Wherein, z(t+1) is the state vector composed of velocity deviation and pose deviation at t+1, β, A and B are intermediate variables, D(t) is the intermediate variable matrix at t, u T (t) is the transpose of the controller parameter vector at t, C(t) is the environment stiffness parameter matrix at t.

[0105] S52, according to the discrete expression of the robot compliant controller, the controller output at the next moment is calculated, and the controller error correction expression is derived;

[0106] Further, the step S52 includes the following steps:

[0107] S521, let the compliant controller parameter matrix U(t)=βu T (t)A+E, according to the discrete expression of the robot compliant controller, the output vector expression at t+1 can be obtained as:

[0108] y(t+1) = U(t)(r(t) - y(t)) + D(t+1)r(t+1)

[0109] where y(t+1) is the output vector of target velocity, pose and contact force at t+1, D(t+1) is the intermediate variable matrix at t+1, and r(t+1) is the vector of desired contact force and robot trajectory at t+1.

[0110] S522, construct an error correction term △U(t), which compensates the output of the system at t+1 to the reference value after the error correction term is added; construct the output error vector e(t+1) of the controller at t+1, which is r(t+1)-y(t+1), then according to the system output vector expression and the error correction term, the following equation can be obtained:

[0111] y(t+1)+e(t+1)=C(t+1)(U(t)+△U(t))e(t)+D(t+1)r(t+1)

[0112] where C(t+1) is the environmental stiffness parameter matrix at t+1, and e(t) is the output vector error of the controller at t. Subtract the above two equations and use the concept of matrix generalized inverse to obtain the error correction expression:

[0113] △U(t)=C + (t+1)e(t+1)e + (t)

[0114] where C + (t+1) is the generalized inverse of the environmental stiffness parameter matrix at t+1, and e + (t) is the generalized inverse of the output error vector of the controller at t.

[0115] S53, design an adaptive iterative learning update law, and construct an environment stiffness estimation algorithm based on data driving;

[0116] Figure 2 is the variable parameter compliant control structure diagram based on iterative learning of the application.

[0117] Further, the step S53 comprises the following steps:

[0118] S531, according to the characteristics of continuously iterating to approach the convergence region according to the iterative learning algorithm, multiply the △U(t) expression by the learning rate α to construct the parameter update law of the robot compliant controller based on iterative learning as follows:

[0119]

[0120] where U k+1 (t) is the compliant controller parameter matrix of the k+1th iteration, and Uk (t) is the compliance controller parameter matrix of the kth iteration, △U k (t) is the error correction term of the kth iteration, is the generalized inverse of the environmental stiffness parameter matrix of the kth iteration, e k (t+1) is the output error vector of the controller of the kth iteration, is the generalized inverse of the output error vector of the controller of the kth iteration.

[0121] S532, the adaptive learning rate a is constructed to improve the convergence speed of the iterative learning process, the adaptive learning rate a is highly related to the output error vector of the control system, and the adaptive expression is:

[0122]

[0123] Where |e(t)| is the modulus of the output error vector of the controller at the current iteration round t.

[0124] S533, in the robot compliance controller parameter update law based on iterative learning, C(t) contains unknown environmental stiffness parameter matrix, the corresponding estimation matrix and the cost function J(C(t)) of the estimation:

[0125]

[0126] Where y k-1 (t) is the output vector composed of target speed, pose and contact force of the k-1th iteration, z k-1 (t) is the state vector composed of speed deviation and pose deviation of the k-1th iteration, is an intermediate variable matrix, μ is a weight factor, is the estimated environmental stiffness parameter matrix of the k-1th iteration.

[0127] Let The estimated environmental stiffness parameter can be calculated, and the expression is:

[0128]

[0129] Where T D is an intermediate variable matrix, 0<η≤2 is a step factor; μ is a weight factor.

[0130] S54, a smoothing update method based on the principle of bilinear interpolation is designed to enhance the continuity of the iterative process, and combined with the adaptive iterative learning update law to form a parameter optimization algorithm based on smoothing iterative learning;

[0131] S541、Assuming the control system output at time t is y(t), the leftmost control system output is y(t-1), and the rightmost control system output is y(t+1), it is obvious that there are error correction terms △U k (t-1) and △U k (t+1) in the kth iteration process. Based on the bilinear interpolation principle, a smooth updating method for the error correction term can be designed, and the expression is:

[0132]

[0133] S542、Similar to the smooth updating process of the error correction term, the smooth updating expression of the environmental stiffness parameter is:

[0134]

[0135] where C k+1 (t) is the environmental stiffness parameter matrix of the k+1th iteration, C k (t) is the environmental stiffness parameter matrix of the kth iteration, △C k (t+1) is the environmental stiffness parameter matrix deviation at time t+1 between the kth and k-1th iterations, △C k (t-1) is the environmental stiffness parameter matrix deviation at time t-1 between the kth and k-1th iterations.

[0136] S543、The complete parameter optimization algorithm based on smooth iterative learning is:

[0137] z k (t+1) = U k (t)(r(t) - y k (t))

[0138] y k (t) = C(t)z k (t) + D(t)r(t)

[0139]

[0140] where z k (t+1) is the state vector composed of the velocity deviation and pose deviation at time t+1 in the kth iteration, z k (t) is the state vector composed of the velocity deviation and pose deviation at time t in the kth iteration, y k (t) is the output vector composed of the target velocity, pose, and contact force at time t in the kth iteration, U k-1 (t) is the compliant controller parameter matrix at time t in the k-1th iteration, and α is the adaptive learning rate. is the generalized inverse of the environment stiffness parameter matrix at time t+1 of the k-1th iteration, e k-1 (t+1) is the output error vector of the controller of the k-1th iteration, is the generalized inverse of the output error vector of the controller of the k-1th iteration.

[0141] The method of the application realizes safe and stable robot collaborative sewing tasks through accurate force-position control, provides a key basic solution for robot autonomous sewing technology, and has important economic value and application prospect for promoting intelligent upgrading of the textile industry.

[0142] The system provided by the application comprises:

[0143] The force-position operation trajectory preset unit is configured to preset the expected contact force trajectory, the robot speed trajectory and the pose trajectory.

[0144] The real contact force calculation unit is configured to collect multiple groups of force sensor data and complete dynamic gravity compensation before the task starts, and calculate the real contact force between the end effector and the sewing platform in real time.

[0145] The collaborative sewing unit is configured to continuously acquire the real contact force during the task, subtract the real robot speed trajectory and pose trajectory from the expected robot speed trajectory and pose trajectory to obtain the speed deviation and the pose deviation, then input the real contact force, the speed deviation and the pose deviation into the compliance controller based on the admittance model, dynamically adjust the target pose of the mechanical arm in the next step, so as to realize the compliant cooperation of the robot; and the compliant cooperation is repeatedly performed until the entire collaborative sewing task is completed.

[0146] The parameter optimization unit is configured to, after the task is completed, construct a parameter optimization algorithm based on a smoothing iterative learning, input the recorded real contact force trajectory, the robot speed trajectory and the pose trajectory, and automatically optimize the parameters of the compliance controller by the smoothing iterative learning algorithm, so as to realize the best force-position trajectory tracking performance.

[0147] The electronic device provided by the application comprises a memory, a processor and a computer program / instruction stored on the memory and executable on the processor, and the computer program / instruction is executed by the processor to realize the steps of the robot compliant force control interaction method for the collaborative sewing scene.

[0148] The computer readable storage medium provided by the application stores computer instructions, and the computer instructions are used to execute the steps of the robot compliant force control interaction method for the collaborative sewing scene when called.

[0149] The above has described the relevant content of the present application. The person skilled in the art can implement the present application based on the description. Based on the above content of the present application, all other embodiments obtained by the person skilled in the art without creative labor shall belong to the protection scope of the present application.

Claims

1. A robot compliant force control interaction method for a collaborative sewing scenario, characterized in that, The method comprises the following steps: a preset desired contact force trajectory, a robot speed trajectory and a pose trajectory; Before the task starts, a plurality of groups of force sensor data are collected and dynamic gravity compensation is completed, and real contact force between an end effector and a sewing platform is calculated in real time; During the task, the real contact force is continuously obtained, the desired robot speed trajectory and the pose trajectory are subtracted from the real robot speed trajectory and the pose trajectory to obtain speed deviation and pose deviation, then the real contact force, the speed deviation and the pose deviation are input into a compliance controller based on a mobility model, and the target pose of the robot in the next step is dynamically adjusted, so that the compliant cooperation of the robot is realized; the compliant cooperation is repeatedly performed until the whole cooperative sewing task is completed; After the task is completed, a parameter optimization algorithm based on smooth iterative learning is constructed, the recorded real contact force trajectory, the robot speed trajectory and the pose trajectory are taken as input, the smooth iterative learning algorithm autonomously optimizes the parameters of the compliance controller, and the best force and position trajectory tracking performance is realized.

2. The robot soft force control interaction method for a collaborative sewing scene according to claim 1, wherein, The plurality of groups of force sensor data are collected and the dynamic gravity compensation is completed, and the real contact force f(t) between the end effector and the sewing platform is calculated in real time, and specifically comprises: The mechanical arm is placed in a free state of static position, i.e. no external force contact, and then the force sensor collects and records the original data f of the force sensor in N different static positions s (t), N≥3; measure the gravity value G of the end effector, and calculate the gravity component f in the force sensor coordinate system according to the pose information of the end effector g (t); calculate the force sensor zero point data f0 by using the least square method; calculate the real contact force f(t)=f s (t)-f0-f g (t).

3. The robot soft force control interaction method for a collaborative sewing scene according to claim 1, wherein, The real contact force, the speed deviation and the pose deviation are input into the compliance controller based on the mobility model, and the target pose of the robot in the next step is dynamically adjusted; comprising the following steps: (1) constructing a parameterized robot compliance controller: wherein, represents the robot acceleration increment output by the compliance controller, and x e (t) represent the velocity and pose deviations of the robot, respectively, f d (t) and f(t) represent the desired and real contact forces, respectively, m(t), b(t), k(t) are all time-varying parameter quantities, representing the virtual inertia, damping and stiffness parameters of the desired dynamic interaction process, respectively; (2) the compliance controller updates the robot target pose x(t+1) in the next step in real time: According to the acceleration increment output by the compliance controller, the adjustment amount of the speed and the pose is obtained through integration, so that the robot target pose x(t+1) in the next step is calculated, and the formula is represented as: x t (t+1) = x(t+1) - x e (t+1) wherein, and x e (t+1) respectively represent the robot velocity increment and the pose increment at time t+1, x(t+1) and x t (t+1) respectively represent the robot real pose and the target pose at time t+1, and At is the sampling time.

4. The robot soft force control interaction method for a collaborative sewing scene according to claim 1, wherein, The parameter optimization algorithm based on smooth iterative learning is constructed, the recorded real contact force trajectory, the robot speed trajectory and the pose trajectory are taken as input, the smooth iterative learning algorithm autonomously optimizes the parameters of the compliance controller, and the best force and position trajectory tracking performance is realized, comprising the following steps: (1) constructing a state space equation of the parameterized compliance controller, and converting it into a discrete expression form to assist in deducing the iterative learning controller; (2) calculating the controller output at the next time according to the discrete expression of the robot compliance controller, and deducing the controller error correction expression; (3) designing an adaptive iterative learning update law, and constructing an environment stiffness estimation algorithm based on data driving; (4) designing a smooth updating method based on the bilinear interpolation principle to enhance the continuity of the iterative process, and combining with the adaptive iterative learning update law to form the parameter optimization algorithm based on smooth iterative learning.

5. The robot soft force control interaction method for collaborative sewing scene according to claim 4, characterized in that, Step (1) specifically comprises the following steps: (11) Let the environment model for the robot end effector interacting with the sewing platform be f(t) = k c (t)x(t), establish the state space equation of the control system: where f(t), f d (t) are the real contact force, the desired contact force trajectory respectively, k c (t) represents the environmental stiffness, x(t) are the real robot pose trajectory and velocity trajectory respectively, and x e (t+1) represent the robot velocity increment and pose increment at t+1 respectively, m(t), b(t), k(t) are all time-varying parameter quantities, representing the virtual inertia, damping and stiffness parameters of the desired dynamic interaction process respectively; x d (t), are the desired robot pose trajectory and velocity trajectory respectively, x e (t), are the pose deviation and velocity deviation respectively, and △t is the sampling time; (12) converting the state space equation of the robot compliance controller into a typical discrete expression: z(t + 1) = (βu T (t)A + B)(r(t) - y(t)) y(t)=C(t)z(t)+D(t)r(t) where z(t + 1) is the state vector composed of velocity and pose errors at time t + 1, β, A and B are intermediate variables, u T (t) is the transpose of the controller parameter vector at time t, C(t) is the environment stiffness parameter matrix at time t, D(t) is the intermediate variable matrix at time t, z(t) is the state vector composed of velocity and pose errors at time t, y(t) is the output vector composed of target velocity, pose and contact force at time t, and r(t) is the input vector composed of desired contact force and robot motion trajectory at time t.

6. The robot soft force control interaction method for collaborative sewing scene according to claim 4, characterized in that, Step (2) specifically comprises the following steps: (21) Let the compliance controller parameter matrix U(t) = βu T (t)A + E, β, A and E are intermediate variables, according to the discrete expression of the robot compliance controller, the output vector expression at t+1 time is: y(t+1)=U(t)(r(t)-y(t))+D(t+1)r(t+1) wherein, y(t+1) is an output vector composed of target velocity, pose and contact force at t+1, D(t+1) is an intermediate variable matrix at t+1, r(t) is an input vector composed of desired contact force and robot motion trajectory at t, y(t) is an output vector composed of target velocity, pose and contact force at t, r(t+1) is a vector composed of desired contact force and robot motion trajectory at t+1; (22) constructing an error correction term △U(t), which makes the output of the system at t+1 compensate to the reference value after the error correction term is added; constructing an output error vector e(t+1) of the controller at t+1, r(t+1)-y(t+1), then according to the system output vector expression and the error correction term, we have: y(t+1)+e(t+1)=C(t+1)(U(t)+△U(t))e(t)+D(t+1)r(t+1) wherein, C(t+1) is an environmental stiffness parameter matrix at t+1, e(t) is an output vector error of the controller at t; subtracting the above two formulas and using the concept of generalized inverse of matrix, the error correction expression is obtained: ΔU(t) = C + (t+1)e(t+1)e + (t) where C + (t + 1) is the generalized inverse of the environmental stiffness parameter matrix at t + 1, e + (t) is the generalized inverse of the output error vector of the controller at t.

7. The robot soft force control interaction method for collaborative sewing scene according to claim 4, characterized in that, Step (3) specifically includes the following steps: (31) according to the characteristics of continuously iterating to approach the convergence region of the iterative learning algorithm, multiplying the expression of △U(t) by the learning rate α to construct the parameter update law of the robot compliant controller based on iterative learning as: where U k+1 (t) is the compliance controller parameter matrix for the k+1 iteration, U k (t) is the compliance controller parameter matrix for the k iteration, ΔU k (t) is the error correction term for the k iteration, is the generalized inverse of the environment stiffness parameter matrix for the k iteration, e k (t+1) is the output error vector for the k iteration of the controller, is the generalized inverse of the output error vector for the k iteration of the controller; (32) constructing an adaptive learning rate α, and its adaptive expression is: wherein, |e(t)| is the module of the output error vector of the controller at the current iteration round t; (33) In the robot compliant controller parameter updating law based on iterative learning, C(t) contains unknown environment stiffness parameter matrix, and the corresponding estimation matrix is constructed with the estimated cost function J(C(t)): where y k-1 (t) is the output vector consisting of the target velocity, pose and contact force of the k-1th iteration, z k-1 (t) is the state vector consisting of the velocity error and the pose error of the k-1th iteration, is an intermediate variable matrix, μ is a weight factor, is the estimated environment stiffness parameter matrix of the k-1th iteration, r(t) is the input vector consisting of the desired contact force and the robot motion trajectory at time t; Let The estimated environmental stiffness parameter is calculated as: where T D is an intermediate variable matrix, 0 < η ≤ 2 is a step size factor; μ is a weight factor.

8. The robot soft force control interaction method for collaborative sewing scene according to claim 4, characterized in that, Step (4) specifically includes the following steps: (41)Assuming the control system output at time t is y(t), the leftmost control system output is y(t-1), and the rightmost control system output is y(t+1), it is obvious that there are error correction terms △U k (t-1) and △U k (t+1) in the kth iteration process. Based on the bilinear interpolation principle, a smoothing update method for the error correction term is designed, and the expression is as follows: wherein U k+1 (t) is the compliance controller parameter matrix of the k+1 iteration, U k (t) is the compliance controller parameter matrix of the k iteration; (42) similar to the smoothing update process of the error correction term, the smoothing update expression of the environmental stiffness parameter is: wherein C k+1 (t) is the environmental stiffness parameter matrix of the k+1th iteration, C k (t) is the environmental stiffness parameter matrix of the kth iteration, C k (t+1) is the environmental stiffness parameter matrix deviation at time t+1 of the kth and k-1th iterations, C k (t-1) is the environmental stiffness parameter matrix deviation at time t-1 of the kth and k-1th iterations; (43) the complete parameter optimization algorithm based on smoothing iterative learning is: z k (t+1) = U k (t)(r(t) - y k (t)) y k (t) = C(t)z k (t) + D(t)r(t) where z k (t+1) is the state vector consisting of velocity error and pose error at time t+1 of the kth iteration, z k (t) is the state vector consisting of velocity error and pose error at time t of the kth iteration, y k (t) is the output vector consisting of target velocity, pose and contact force at time t of the kth iteration, U k-1 (t) is the compliance controller parameter matrix at time t of the k-1th iteration, and a is the adaptive learning rate, is the generalized inverse of the environment stiffness parameter matrix at time t+1 of the k-1th iteration, e k-1 (t+1) is the output error vector of the controller at time t+1 of the k-1th iteration, is the generalized inverse of the output error vector of the controller at time t+1 of the k-1th iteration, is the estimated environment stiffness parameter matrix at time t+1 of the kth iteration, y k-1 (t) is the output vector consisting of target velocity, pose and contact force at time t of the k-1th iteration, z k-1 (t) is the state vector consisting of velocity error and pose error at time t of the k-1th iteration, is the estimated environment stiffness parameter matrix at time t of the k-1th iteration, T D is the intermediate variable matrix, 0 < η < 2 is the step factor, and μ is the weight factor.

9. A robot soft force control interaction system for a collaborative sewing scenario, characterized in that, including: a force-position operation trajectory preset unit, configured to preset a desired contact force trajectory, a robot velocity trajectory and a pose trajectory; a real contact force calculation unit, configured to collect multiple groups of force sensor data and complete dynamic gravity compensation before the task starts, and calculate the real contact force between the end effector and the sewing platform in real time; a collaborative sewing unit, configured to continuously acquire the real contact force during the task, subtract the real robot velocity trajectory and pose trajectory from the desired robot velocity trajectory and pose trajectory to obtain a velocity deviation and a pose deviation, then input the real contact force, the velocity deviation and the pose deviation into the compliant controller based on the admittance model, dynamically adjust the target pose of the mechanical arm in the next step, so as to realize the compliant collaboration of the robot; repeatedly performing the compliant collaboration until the entire collaborative sewing task is completed; a parameter optimization unit, configured to, after the task is completed, construct a parameter optimization algorithm based on smoothing iterative learning, take the recorded real contact force trajectory, robot velocity trajectory and pose trajectory as input, and optimize the compliant controller parameters by the smoothing iterative learning algorithm to realize the best force-position trajectory tracking performance.

10. An electronic device, comprising: A computer program product comprising a memory, a processor and computer programs / instructions stored on the memory and executable on the processor, said computer programs / instructions implementing the steps of the method for robot compliant force control interaction for a collaborative sewing scenario according to any one of claims 1-8 when executed by the processor.

Images