Human-guided robot-environment interaction adaptive control method and related device

Abstract

The invention belongs to a robot control method, and provides a human-guided robot-environment interaction self-adaptive control method and a related device for solving the technical problems that an existing human-guided robot-environment interaction control method is insufficient in motion stability, low in human operator safety and low in control precision. And estimating an optimal impedance parameter through a state space model, and learning a human-guided reference trajectory through a neural network in combination with the optimal impedance parameter. Then, in combination with a robot Jacobian matrix and a human-guided reference trajectory, a reference joint speed and an estimated acceleration of a joint space are obtained through closed-loop inverse kinematics, then network parameters of a robot dynamic model are obtained through approximation of an uncertain dynamic model, and finally, the robot dynamic model is obtained through combination with the reference joint speed and the network parameters of the joint space. And the mechanical arm control torque is obtained through the offset width fuzzy neural network. Accurate and compliant control under uncertain motion and dynamics conditions is realized.

Description

Technical Field

[0001] This application belongs to a robot control method, and particularly relates to a human-guided robot-environment interaction adaptive control method and related devices. Background Art

[0002] Human-guided robot-environment interaction has been increasingly widely used in industrial fields and daily life. These tasks usually include trajectory and force guidance to achieve effective cooperation. Robots must simultaneously meet the requirements of operability in cooperation and stability in interaction to achieve optimal performance.

[0003] Existing control methods mainly focus on human-robot or robot-environment interaction. However, in many human-robot interaction tasks, robots may interact with the environment simultaneously. For example, in manually guided tool machining, this interaction may disrupt the stability of motion, damage task execution, and even endanger the safety of human operators. In addition, due to joint wear and differences in structural information, the uncertainty of robot kinematics and dynamics also reduces control accuracy. Summary of the Invention

[0004] Aiming at the technical problems of insufficient motion stability, low safety of human operators, and low control accuracy existing in the existing control methods for human-guided robot-environment interaction, this application provides a human-guided robot-environment interaction adaptive control method and related devices.

[0005] To achieve the above object, this application is implemented by adopting the following technical solutions: In a first aspect, this application proposes a human-guided robot-environment interaction adaptive control method, including: Estimating optimal impedance parameters through a state space model to minimize the tracking error of the robot-environment interaction; and combining the optimal impedance parameters, learning the human-guided reference trajectory through a neural network to optimize the interaction performance; Real-time correcting the Jacobian matrix of the robot through sensor data; Combining the robot Jacobian matrix and the human-guided reference trajectory, obtaining the reference joint velocity and estimated acceleration in the joint space through closed-loop inverse kinematics; at the same time, combining the estimated acceleration, synchronously adjusting the robot Jacobian matrix through Jacobian adaptation; According to the actual manipulator joint velocity, approximating through an uncertain dynamics model to obtain the network parameters of the robot dynamics model; combining the reference joint velocity in the joint space and the network parameters, obtaining the manipulator control torque through a bias width fuzzy neural network.

[0006] Further, the method for estimating optimal impedance parameters through a state space model includes: Set state variables; Rewrite the state variables into a linear state space system; Define the cost function of the impedance model; Use adaptive optimal impedance learning to solve the minimization of the cost function and obtain the optimal impedance parameters.

[0007] Furthermore, the expression of the state variables is:

[0008] where is the desired trajectory, , is the first adjustable matrix, , is the second adjustable matrix, , is the intermediate state variable, is the derivative of the intermediate state variable; The expression of the linear state space system is:

[0009] where , is the constructed diagonal matrix, is the first positive definite diagonal coefficient matrix, is the second definite diagonal coefficient matrix, is the introduced composite state variable, , is the derivative of the composite state variable, x is the trajectory; The expression of the cost function of the impedance model is:

[0010] where is the reference trajectory, is the actual interaction force, is the desired interaction force, , is the tracking error weight, is the cost function of the impedance model, , is the interaction force weight; .

[0011] Furthermore, the method of using adaptive optimal impedance learning to solve the minimization of the cost function includes: When , use as the input interaction force and calculate:

[0012] where is the rank of the matrix, is a custom matrix for subsequent calculations, is a custom matrix for subsequent calculations, is the dimension of, is the dimension of, is the assumed initial interaction force, is the assumed initial impedance gain, is the iteration termination threshold, is a custom matrix for subsequent calculations, is a custom matrix for subsequent calculations, is a custom matrix for subsequent calculations, is a custom matrix for subsequent calculations, is the m-dimensional identity matrix, is a custom matrix for subsequent calculations, is the matrix the (m - 1)-th element; When , solve to obtain the optimal impedance parameter ; where, is is to expand into the operator of, is the impedance parameter during iteration;

[0013] Among them, is a custom matrix for subsequent calculations, is the m-dimensional identity matrix, is the interaction force weight, is a custom matrix for subsequent calculations;

[0014] Among them, , is the trajectory error weight.

[0015] Furthermore, the method for learning the human-guided reference trajectory through the neural network includes: Taking the desired trajectory corresponding to the parameter value corresponding to the trajectory parameterization as the input of the neural network to obtain the reference trajectory; Among them, the neural network includes two hidden layers, and the node matrices of the two hidden layers include:

[0016] Among them, is the node matrix of the first hidden layer, is the node matrix of the second hidden layer, is the first conversion function, is the second conversion function, is the first connection weight, is the second connection weight, is the set of desired trajectories, is the first bias, is the second bias; The calculation method of the reference trajectory includes:

[0017] where, is the third connection weight, is the third bias.

[0018] Furthermore, the method for real-time correcting the Jacobian matrix of the robot through sensor data includes: Define the data-driven error through the following formula:

[0019] where, is the data-driven error function, is the end-effector velocity of the manipulator predicted based on the Jacobian estimation, is the actual end-effector velocity of the manipulator collected by the sensor, is the estimated Jacobian matrix, is the actual joint velocity of the manipulator; Correct the Jacobian matrix through the following formula:

[0020] where, is the corrected Jacobian matrix, is the learning rate, is the gradient; The method for obtaining the reference joint velocity and the estimated acceleration in the joint space through closed-loop inverse kinematics includes:

[0021] where, is the reference joint velocity, is a positive definite finite matrix, is the trajectory tracking error.

[0022] Furthermore, the method for obtaining the output of the bias width fuzzy neural network includes:

[0023] where, is the output of the offset-width fuzzy neural network, is the updated node matrix, is the unfolded connection weight matrix, is the extended fuzzy rule matrix, is the extended enhanced node matrix, is the fuzzy rule matrix before extension, is the th fuzzy rule, is the enhanced node matrix before extension, is the result defined by the transformation function corresponding to the th enhanced node using the triangular expansion scheme, is the introduced offset, is the number of enhanced nodes; The method for obtaining the control torque of the robotic arm includes: Introduce the output of the offset-width fuzzy neural network into the controller to obtain the control torque of the robotic arm:

[0024] where, is the control torque of the robotic arm, , and are the approximation results of the uncertain robot dynamics model using the offset-width fuzzy neural network, is the control gain, is the joint velocity, is the joint tracking error, is the interaction force of the environment, is the virtual control term, is the derivative of the virtual control term.

[0025] In a second aspect, the present application proposes a human-guided robot-environment interaction adaptive control system, including: An optimization module for estimating optimal impedance parameters through a state space model to minimize the force / position tracking error of the robot's interaction with the environment; and combining the optimal impedance parameters to learn the human-guided reference trajectory through a neural network to optimize the interaction performance; A correction module for real-time correcting the Jacobian matrix of the robot through sensor data; A conversion module for combining the robot's Jacobian matrix and the human-guided reference trajectory to obtain the reference joint velocity and estimated acceleration in the joint space through closed-loop inverse kinematics; at the same time, combining the estimated acceleration to synchronously adjust the robot's Jacobian matrix through Jacobian adaptation; An approximation control module, configured to approximate the network parameters of the robot dynamics model according to the actual robotic arm joint speed through an uncertain dynamics model; and combine the reference joint speed in the joint space and the network parameters to obtain the robotic arm control torque through a bias-width fuzzy neural network.

[0026] In a third aspect, the present application provides an electronic device, including: a memory and one or more processors; the memory is coupled to the processor; wherein, computer program code is stored in the memory, and the computer program code includes computer instructions, when the computer instructions are executed by the processor, the electronic device executes the steps of the above-mentioned human-guided robot-environment interaction adaptive control method.

[0027] In a fourth aspect, the present application provides a computer-readable storage medium, in which a computer program is stored, and when the computer program is executed by a processor, the steps of the above-mentioned human-guided robot-environment interaction adaptive control method are implemented.

[0028] Compared with the prior art, the present application has the following beneficial effects: The present application provides a human-guided robot-environment interaction adaptive control method. The optimal impedance parameters are estimated through a state space model, and the reference trajectory guided by a human is learned through a neural network in combination with the optimal impedance parameters. Then, in combination with the robot Jacobian matrix and the reference trajectory guided by a human, the reference joint speed and the estimated acceleration in the joint space are obtained through closed-loop inverse kinematics. Then, the network parameters of the robot dynamics model are obtained through approximation by an uncertain dynamics model. Finally, in combination with the reference joint speed in the joint space and the network parameters, the robotic arm control torque is obtained through a bias-width fuzzy neural network. The present application adopts impedance / reference hybrid adaptation. Among them, adaptive impedance learning autonomously infers human intentions and adjusts the impedance to adapt to unknown environments and human limbs, while reference adaptation adjusts the trajectory to effectively handle the interaction caused by oscillation. In addition, an adaptive controller that combines Jacobian matrix adaptive closed-loop inverse kinematics and a gain adjustment control strategy based on a width fuzzy neural network is also provided, realizing precise compliant control under uncertain motion and dynamics conditions.

[0029] The present application also provides a human-guided robot-environment interaction adaptive control system, an electronic device and a computer-readable storage medium, which have all the advantages of the above-mentioned human-guided robot-environment interaction adaptive control method. Description of the Drawings

[0030] To more clearly illustrate the technical solutions of the embodiments of the present application, the following will briefly introduce the accompanying drawings required for the embodiments. It should be understood that the following drawings only show some embodiments of the present application and should not be regarded as limiting the scope. For those of ordinary skill in the art, without creative efforts, other related drawings can also be obtained based on these drawings.

[0031] Figure 1 It is a schematic diagram of the human-guided robot-environment interaction of the present application; Figure 2 It is a schematic flow diagram of the human-guided robot-environment interaction adaptive control method of the present application; Figure 3 It is a schematic diagram of the overall principle of the human-guided robot-environment interaction adaptive control method of the present application; Figure 4 It is a schematic diagram of the impedance / reference hybrid adaptive principle in the embodiments of the present application; Figure 5 It is a structural diagram of the Jacobian matrix adaptive closed-loop inverse kinematics in the embodiments of the present application; Figure 6 It is a schematic diagram of the deviation width fuzzy neural network structure proposed by the present application; Figure 7 It is a schematic diagram of the human-guided robot-environment interaction adaptive control system of the present application. Detailed implementation manners

[0032] To make the objectives, technical solutions, and advantages of the embodiments of the present application clearer, the following will clearly and completely describe the technical solutions in the embodiments of the present application with reference to the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are some, but not all, of the embodiments of the present application. Usually, the components of the embodiments of the present application described and shown in the accompanying drawings here can be arranged and designed in various different configurations.

[0033] Therefore, the following detailed description of the embodiments of the present application provided in the drawings is not intended to limit the scope of the present application claimed, but merely represents the selected embodiments of the present application. Based on the embodiments of the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present application.

[0034] It should be noted that similar reference numerals and letters denote similar items in the following drawings. Therefore, once an item is defined in one drawing, it does not need to be further defined and explained in subsequent drawings.

[0035] In the description of the embodiments of the present application, it should be noted that if terms such as "upper", "lower", "horizontal", "inner", etc. indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, or the orientation or positional relationship in which the inventive product is usually placed during use, it is only for the convenience of describing the present application and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and thus should not be construed as a limitation to the present application. In addition, terms such as "first", "second", etc. are only used for distinguishing descriptions and cannot be construed as indicating or implying relative importance.

[0036] In addition, if the term "horizontal" appears, it does not mean that the component is required to be absolutely horizontal, but it can be slightly inclined. For example, "horizontal" only means that its direction is more horizontal relative to "vertical", and does not mean that the structure must be completely horizontal, but it can be slightly inclined.

[0037] In the description of the embodiments of the present application, it should also be noted that unless otherwise clearly specified and limited, if terms such as "set", "installed", "connected", "coupled" are understood in a broad sense. For example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be directly connected or indirectly connected through an intermediate medium, and it can be the communication inside two elements. For those of ordinary skill in the art, the specific meanings of the above terms in the present application can be understood according to specific circumstances.

[0038] In recent years, the application of human-robot collaborative robot systems has rapidly expanded in the fields of industrial manufacturing and daily services. Such systems usually need to perform complex trajectory planning and force-position hybrid control to achieve efficient collaborative operations. However, in the physical interaction scenario, the robot needs to satisfy dual constraints simultaneously: on the one hand, it is necessary to ensure the safety and operational compliance of human-robot collaboration, and on the other hand, it is necessary to maintain the dynamic stability of the interaction with the environment, which poses a major challenge to the design of the control system.

[0039] Current mainstream control frameworks mostly focus on a single interaction modality, that is, either dealing with force feedback and motion guidance in human-robot interaction, or focusing on the contact force control between the robot and the environment. However, in actual operation scenarios, such as precision assembly assisted by manual teaching or collaborative handling tasks, robots often need to handle multi-modal interactions simultaneously. This composite interaction mode may cause the destruction of the system stability boundary, resulting in the accumulation of trajectory tracking errors at least, and even equipment damage or casualties at worst. In addition, due to factors such as mechanical structure aging, non-linear joint clearances, and time-varying load distributions, the parameter uncertainties of the robot kinematic model will form a coupling effect with dynamic disturbances, further reducing the control accuracy and system robustness. Human-guided robot-environment interaction is a collaborative mode that integrates human intelligence and robot capabilities. Its core lies in that humans guide the robot through intuitive operations (such as teaching, force feedback, or voice commands) to accurately complete the tasks of interacting with the environment. This interaction mode breaks the limitation of the "preset program" of traditional robots, enabling the robot to dynamically adapt to complex environments and showing revolutionary potential in fields such as medical care, services, and industry. As Figure 1 shown, it is a schematic diagram of human-guided robot-environment interaction. These tasks usually include humans providing trajectory and force guidance to the robot to achieve effective collaboration and controlling the robot to interact with the environment. In this case, the robot must meet both the operability requirements in collaboration and the stability requirements in interaction to achieve optimal performance.

[0040] The above-mentioned problem of composite interaction control under multi-source uncertainties has become the key bottleneck restricting the development of the human-machine-material fusion system towards a higher level of intelligence, and there is an urgent need to develop a new control architecture to break through the current technical limitations.

[0041] Based on the above situation, this application proposes a human-guided robot-environment interaction adaptive control method and related devices. The following will make a detailed description of this application in combination with embodiments and drawings.

[0042] As Figure 2 shown, it is a schematic flow diagram of the human-guided robot-environment interaction adaptive control method of this application, which may include: S101, estimating the optimal impedance parameters through the state space model to minimize the tracking error of the robot-environment interaction. Combining the optimal impedance parameters, learning the reference trajectory guided by humans through the neural network to optimize the interaction performance.

[0043] The state - space model is a mathematical model used to describe the behavior of dynamic systems. It relates the state variables of the system (such as position, velocity, etc.) to inputs and outputs. In the scenario of robot - environment interaction, impedance control is an important control strategy. It adapts to different environments by adjusting the impedance characteristics of the robot (such as stiffness, damping, etc.). By establishing a state - space model, the interaction process between the robot and the environment can be accurately modeled. Based on this model, optimization algorithms (such as gradient descent method, genetic algorithm, etc.) are used to find the optimal impedance parameters, so that the tracking error of the robot during interaction with the environment, that is, the deviation between the actual motion trajectory of the robot and the desired motion trajectory, is minimized. This helps to improve the stability and accuracy of the robot when in contact with the environment, enabling it to perform tasks better.

[0044] After obtaining the optimal impedance parameters, they can be used as the basis for robot control. A neural network is a machine - learning model with powerful learning capabilities that can simulate the neuron structure and working mode of the human brain. Here, a neural network is used to learn the reference trajectory under human guidance. Humans can guide the robot's movement in various ways, such as manual operation, teaching, etc. The robot records these motion data, including position, velocity, acceleration, etc. as training samples. Through learning these samples, the neural network can understand the human motion intention and pattern and generate the corresponding reference trajectory. Combining the optimal impedance parameters, when the robot tracks these reference trajectories, it can better adapt to environmental changes, optimize the interaction performance with the environment, and improve the quality and efficiency of task execution.

[0045] S102, Real - time correct the Jacobian matrix of the robot through sensor data.

[0046] The Jacobian matrix is an important mathematical tool for describing the motion relationship between the robot's joint space and the Cartesian space (workspace), reflecting how small changes in joint variables cause motion changes of the end - effector in the Cartesian space. In practical applications, due to factors such as the structural deformation of the robot and environmental interference, the Jacobian matrix may change, thus affecting the motion control accuracy of the robot. Therefore, various sensors installed on the robot (such as position sensors, force sensors, etc.) are used to obtain the state information of the robot in real - time, such as joint positions, forces and torques of the end - effector. According to these sensor data, specific algorithms are used to correct the Jacobian matrix in real - time so that it can accurately reflect the current motion state of the robot and provide a more accurate basis for subsequent motion control.

[0047] S103, Combine the robot's Jacobian matrix and the reference trajectory guided by humans. Through closed - loop inverse kinematics, obtain the reference joint velocity and estimated acceleration in the joint space; at the same time, combine the estimated acceleration and synchronously adjust the robot's Jacobian matrix through Jacobian adaptation.

[0048] S104. Based on the actual robotic arm joint speed, approximate it through the uncertain dynamics model to obtain the network parameters of the robot dynamics model. Combine the reference joint speed in the joint space and the network parameters, and through the bias-width fuzzy neural network, obtain the control torque of the robotic arm.

[0049] Closed-loop inverse kinematics is a method for solving the joint motion of a robot. Given the desired motion trajectory of the robot's end effector, i.e., the reference trajectory guided by humans, and the current state of the robot, the current state of the robot can be described by the Jacobian matrix. Through iterative calculations, the reference joint speed and estimated acceleration in the joint space are solved. Specifically, by continuously adjusting the joint speed, the robot's end effector can track the reference trajectory, and at the same time, the corresponding acceleration is calculated according to the kinematic relationship. During this process, combined with the estimated acceleration, the Jacobian matrix is synchronously adjusted using the Jacobian adaptive algorithm. The Jacobian adaptive algorithm adjusts the parameters of the Jacobian matrix in real time according to information such as acceleration to adapt to the changes during the robot's movement, further improving the accuracy and stability of motion control.

[0050] The bias-width fuzzy neural network is an intelligent control method that combines the advantages of fuzzy logic and neural networks. After obtaining the reference joint speed in the joint space and the network parameters of the robot dynamics model, they can be used as the inputs of the bias-width fuzzy neural network. Fuzzy logic can handle uncertain and imprecise information, while the neural network has strong learning and adaptive capabilities. Through the calculation and processing of the bias-width fuzzy neural network, according to the input reference joint speed and network parameters, combined with the controller, an appropriate control torque of the robotic arm can be output. This control torque is used to drive the joint motion of the robotic arm, enabling the robotic arm to interact with the environment in the desired manner and achieve precise task execution.

[0051] This application proposes an adaptive control framework that comprehensively considers humans, robots, and the environment, solves the complexity of unknown and uncertain models in robot-environment interaction under human guidance, enhances the adaptability of the robot to environmental changes and human operations, and thus improves the safety and reliability of the interaction.

[0052] Such as Figure 3As shown in the figure, it is a schematic diagram of the overall principle of the human-guided robot-environment interaction adaptive control method of the present application. It mainly includes two core stages: the estimation of the unknown interaction object model, estimating the impedance parameters of the interaction between the robotic arm and the outside world, and outputting the reference trajectory; and the approximation of the uncertain robot kinematics / dynamics model. For the stage of estimating the unknown interaction object model, an impedance / reference hybrid adaptive method is proposed, which estimates the human intention based on the dynamic fine-tuning impedance model and adaptively learns the impedance gain to modify the reference trajectory. The adaptive impedance learning therein autonomously infers the human intention and adjusts the impedance to adapt to the unknown environment and the human body limbs, while the reference adaptation adjusts the trajectory to effectively handle the interaction caused by oscillation. For the approximation of the uncertain robot kinematics / dynamics model, the present application proposes an adaptive controller that combines the Jacobian matrix adaptive closed-loop inverse kinematics with the gain adjustment control strategy based on the width fuzzy neural network, realizing precise and flexible control under uncertain motion and dynamic conditions.

[0053] The following is a more detailed description of the present application through another embodiment of the present application.

[0054] As Figure 4 shown, it is a schematic diagram of the impedance / reference hybrid adaptive principle. It includes two parts: adaptive impedance learning and neural network-based reference adaptation. The adaptive impedance learning and the neural network-based reference adaptation are carried out simultaneously in the impedance / reference hybrid adaptation to fine-tune the impedance model. This closed-loop hybrid learning process effectively integrates the motion and force information, helps to optimize the human-guided robot-environment interaction, and shows great application potential.

[0055] When the human-guided robotic arm interacts with the environment, the dynamic model in the Cartesian space can be expressed as:

[0056] Among them, , , are the inertia matrix, the damping matrix, and the stiffness matrix respectively, is the force exerted by the human, is the interaction force with the environment, is the acceleration of the robot end, is the velocity of the robot end, is the position of the robot. In addition, the target impedance model of the interaction in the task space can be expressed as:

[0057] Among them, , , are all positive definite diagonal coefficient matrices, is the desired trajectory.

[0058] This application proposes an impedance modeling method for estimating impedance parameters. First, a composite state variable is introduced , where is used as an intermediate state variable and is defined as:

[0059] where, is the derivative of the intermediate state variable, is the desired trajectory, , is the first adjustable matrix, , is the second adjustable matrix, , is the intermediate state variable.

[0060] Next, the defining formula of can be rewritten as a linear state space system:

[0061] where, is the derivative of the composite state variable, , is a constructed diagonal matrix, is the first positive definite diagonal coefficient matrix, is the second definite diagonal coefficient matrix, is the introduced state variable, , x is.

[0062] Then, to achieve optimal interaction performance, the cost function of the impedance model is defined as:

[0063] where, is the reference trajectory, is the actual interaction force, is the desired interaction force, , is the tracking error weight, is the cost function of the impedance model, , is defined as:

[0064] Finally, the desired trajectory is parameterized as to optimize . To parameterize the end effector trajectory as , the dynamic model in the Cartesian space is combined with the target impedance model of the interaction in the task space to obtain:

[0065] Similarly, the force and can be parameterized by (2). Therefore, the cost function can be expressed as a function of that minimizes to . Parametric methods such as non-uniform rational B-spline curve fitting can fit any trajectory and adapt to specific task requirements.

[0066] Next, adaptive optimal impedance learning is used to solve for the optimal feedback gain, that is, to solve how to minimize : (1) When holds, first use as the input interaction force and calculate:

[0067] where is the rank of the matrix, is a custom matrix for subsequent calculations, is a custom matrix for subsequent calculations, is 's dimension, is 's dimension, is the assumed initial interaction force, is the assumed initial impedance gain, is the iteration termination threshold, is a custom matrix for subsequent calculations, is a custom matrix for subsequent calculations, is a custom matrix for subsequent calculations, is a custom matrix for subsequent calculations, is the m-dimensional identity matrix, is a custom matrix for subsequent calculations, is the matrix 's (m - 1)-th element.

[0068] It should be noted that if is not satisfied, then directly proceed to the subsequent step (2).

[0069] (2) When holds, solve .

[0070] where .

[0071] is a custom matrix for subsequent calculations, is the m-dimensional identity matrix, is the interaction force weight, is a custom matrix for subsequent calculations, is the impedance parameter during iteration, is an n-dimensional identity matrix, , is an operator that expands into , to obtain the optimal impedance parameter The G at the end of iteration is .

[0072] It should be noted that if is not satisfied, it means that the iteration can end and the optimal impedance parameter has been obtained .

[0073] Meanwhile, the impedance gain can be adjusted from to using (9) to improve the smoothness of the reference trajectory:

[0074] where, is the impedance gain corresponding to the adjustment time , is the adjustment time length, is the adjustment time.

[0075] Meanwhile, the present application proposes a neural network-based reference adaptation method by designing a simple backpropagation neural network to learn the reference trajectory. Using the corresponding desired trajectory as the input. The neural network includes two hidden layers, and each hidden layer has 16 neurons. The node matrices of these layers can be calculated as:

[0076] where, is the node matrix of the first hidden layer, is the node matrix of the second hidden layer, is the first transfer function, is the second transfer function, is the first connection weight, is the second connection weight, is the set of desired trajectories, is the first bias, is the second bias, randomly selected.

[0077] Therefore, the output of the neural network-based reference adaptation is calculated as:

[0078] Among them, is the third connection weight, the third bias.

[0079] In a neural network, the cost function of the impedance model serves as the cost function, and dynamic hierarchical backpropagation is used to iteratively adjust the weights of each layer, enabling the network to effectively learn from input-output pairs.

[0080] As Figure 5 shown, this is the Jacobian matrix adaptive closed-loop inverse kinematics structure diagram in this embodiment. The Jacobian adaptive method proposed in this application is data-driven, and its essence is to learn the kinematic model from the robot state information measured by sensors. First, define the data-driven error:

[0081] Among them, is the data-driven error function, is the end-effector velocity of the manipulator predicted based on the Jacobian estimate, is the actual end-effector velocity of the manipulator collected by the sensor, is the estimated Jacobian matrix, is the actual manipulator joint velocity.

[0082] Next, using the gradient descent method, the adaptive learning of the Jacobian coefficient can be designed as:

[0083] Among them, is the corrected Jacobian matrix, is the learning rate, is the gradient.

[0084] At the same time, this application proposes a Jacobian adaptive closed-loop inverse kinematics for converting the reference trajectory into the joint space under uncertain kinematic conditions. In the Jacobian adaptive closed-loop inverse kinematics, a convenient pure proportional control law is defined as:

[0085] Among them, is the reference joint velocity, is a positive definite finite matrix, and its eigenvalues affect the convergence rate of is the trajectory tracking error.

[0086] As Figure 6As shown, it is a schematic diagram of the bias width fuzzy neural network structure proposed in this application. The control accuracy and adaptability to complex tasks are effectively enhanced. The network is used to form a bias width fuzzy neural network controller for unknown dynamic model estimation and interactive control of the robot, and the interaction force is obtained by using a momentum-based force observer. The width fuzzy neural network is a node-self-increasing fuzzy neural network that combines the width learning system algorithm with the fuzzy neural network. The width learning system is a novel incremental learning method inspired by random vector function link neural networks. Within the framework of the width learning system, feature mapping is used to replace the input vector of the hidden layer and further expanded using enhanced nodes. The width fuzzy neural network uses the width learning system algorithm to dynamically increase the number of neural network nodes, thereby greatly enhancing the generalization ability of the network to adapt to new input samples. Most importantly, the need for subjective manual adjustment of network parameters and the number of nodes during network design is eliminated.

[0087] This application enhances the width fuzzy neural network to improve its performance in complex trajectory tracking control problems with optimal interactions. Unlike the width fuzzy neural network that may fall into local minima and lead to reduced control performance, the new method of this application introduces a local bias term in the transformation function matrix, which can improve the dynamic approximation accuracy of the width fuzzy neural network controller, thereby correspondingly improving the control accuracy. At the same time, the biased width fuzzy neural network can adaptively add neural nodes without manually adjusting the network parameters, making it more suitable for robot dynamic control problems.

[0088] In order to make the width learning system more suitable for robot control, a Gaussian function is selected as the transfer function:

[0089] in, is the center of the Gaussian function, is the standard deviation, is the number of member nodes, initially set to 1. According to the Euclidean distance between the reference trajectory and the center of the Gaussian function Adaptive increase, where yes The closest Before The average value of the centers, is the joint angle.

[0090] When the distance Exceeding the threshold When , a new member node is added. The new node center is defined as:

[0091] in, is the central update step size, is the new node center.

[0092] Next, apply the first-order fuzzy rule. The j th fuzzy rule is calculated as , where is the dimension of each sample, is the number of enhanced nodes. The transformation function corresponding to the enhanced nodes is defined using the triangular expansion scheme as: , thus reducing the computational burden.

[0093] Furthermore, the enhanced node matrix becomes:

[0094] where is the input dimension.

[0095] Meanwhile, a bias is introduced to improve the approximation accuracy of the controller for the dynamic model with significant errors. The output of the network is expanded as:

[0096] After adding new nodes, the updated node matrix becomes , then the connection weight matrix is expanded as , where is the old weight matrix.

[0097] Therefore, the network output is expressed as:

[0098] where and , , is the output of the bias-width fuzzy neural network, is the updated node matrix, is the expanded connection weight matrix, is the expanded fuzzy rule matrix, is the expanded enhanced node matrix, is the fuzzy rule matrix before expansion, is the th fuzzy rule, is the enhanced node matrix before expansion, is the result of the transformation function corresponding to the th enhanced node defined using the triangular expansion scheme, is the introduced bias, is the number of enhanced nodes.

[0099] Furthermore, an adaptive optimal interaction controller is constructed based on the bias-width fuzzy neural network. First, the virtual control term is defined , where is the reference trajectory in the joint space, is the control gain, is the tracking error.

[0100] Next, the interaction force is introduced into the controller, and the robot control torque is defined as:

[0101] where is the derivative of the virtual control term, is the virtual control term, is the manipulator control torque, is similar to control gain, , and are the approximation results of the formula for the uncertain robot dynamic model modified by the bias-width fuzzy neural network:

[0102] where is the error matrix, and are respectively transformation matrix and weight matrix:

[0103] where are all learning rates, are all discontinuous switching constants, defined as , is the weight update threshold.

[0104] As Figure 7 shown, it is a schematic diagram of a human-guided robot-environment interaction adaptive control system of the present application, which may include: An optimization module for estimating the optimal impedance parameters through a state space model to minimize the force / position tracking error of the robot's interaction with the environment; and combining the optimal impedance parameters, learning the human-guided reference trajectory through a neural network to optimize the interaction performance; A correction module for real-time correcting the Jacobian matrix of the robot through sensor data; A conversion module, configured to combine the robot Jacobian matrix and the human-guided reference trajectory, and obtain the reference joint velocity and estimated acceleration in the joint space through closed-loop inverse kinematics; meanwhile, combine the estimated acceleration and synchronously adjust the robot Jacobian matrix through Jacobian adaptation. An approximation control module, configured to obtain the network parameters of the robot dynamics model through approximation of the uncertain dynamics model according to the actual manipulator joint velocity; combine the reference joint velocity in the joint space and the network parameters, and obtain the manipulator control torque through a bias-width fuzzy neural network.

[0105] It should be noted that in several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the system embodiments described above are only illustrative. For example, the division of each module is only a logical function division, and there may be other division methods in actual implementation. For example, multiple modules can be combined or integrated into another device, or some features can be ignored or not executed. The modules described as separate components may or may not be physically separated, and the components shown as modules may be one physical unit or multiple physical units, that is, they may be located in one place or distributed to multiple different places. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0106] In addition, the modules in each embodiment of the present invention can be integrated in a processing unit, or each module can exist physically alone, or two or more modules can be integrated in one unit. The above integrated unit can be implemented in the form of hardware or in the form of a software functional unit.

[0107] The embodiment of this application also provides an electronic device, which may include one or more processors, a memory, and a communication interface.

[0108] Among them, the memory and the communication interface are coupled to the processor. For example, the memory and the communication interface can be coupled together through a bus.

[0109] Among them, the communication interface is used for data transmission with other devices. The memory stores computer program code. The computer program code includes computer instructions, and when the computer instructions are executed by the processor, the electronic device is caused to execute the steps of the above-mentioned human-guided robot-environment interaction adaptive control method.

[0110] Among them, the processor can be a processor or a controller. For example, it can be a Central Processing Unit (CPU), a general-purpose processor, a Digital Signal Processor (DSP), an Application-Specific Integrated Circuit (ASIC), a Field Programmable Gate Array (FPGA), or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. It can implement or execute various exemplary logical blocks, modules, and circuits described in conjunction with the present disclosure. The processor can also be a combination that implements computing functions, such as a combination of one or more microprocessors, a combination of a DSP and a microprocessor, and so on. The processor can be used to support the electronic device in executing the method steps provided in the above embodiments.

[0111] Among them, the bus can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. The above buses can be divided into an address bus, a data bus, a control bus, and so on.

[0112] A computer-readable storage medium provided by an embodiment of the present application stores a computer program, and when the computer program is executed by a processor, it implements the steps of the above human-guided robot-environment interaction adaptive control method.

[0113] The computer-readable storage medium involved in the present application includes a Random Access Memory (RAM), an internal memory, a Read-Only Memory (ROM), an Electrically Programmable ROM, an Electrically Erasable Programmable ROM, a register, a hard disk, a removable disk, a CD ROM, or any other form of storage medium well-known in the technical field.

[0114] The above are only the preferred embodiments of the present application and are not used to limit the present application. For those skilled in the art, the present application can have various changes and modifications. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included within the protection scope of the present application.

Claims

1. A human-guided robot-environment interaction adaptive control method, characterized in that, Including: Estimate the optimal impedance parameters through a state - space model to minimize the tracking error of the robot's interaction with the environment; Combined with the optimal impedance parameters, learn the reference trajectory guided by humans through a neural network to optimize the interaction performance; Real - time correct the Jacobian matrix of the robot through sensor data; Combined with the robot's Jacobian matrix and the reference trajectory guided by humans, through closed - loop inverse kinematics, obtain the reference joint velocity and estimated acceleration in the joint space; meanwhile, combined with the estimated acceleration, synchronously adjust the robot's Jacobian matrix through Jacobian adaptation; According to the actual manipulator joint velocity, approximate through an uncertain dynamics model to obtain the network parameters of the robot dynamics model; Combined with the reference joint velocity in the joint space and the network parameters, obtain the manipulator control torque through a bias - width fuzzy neural network.

2. The human-guided robot-environment interaction adaptive control method according to claim 1, characterized in that The method for estimating the optimal impedance parameters through a state - space model includes: Set state variables; Rewrite the state variables into a linear state - space system; Define the cost function of the impedance model; Use adaptive optimal impedance learning to solve the minimization of the cost function to obtain the optimal impedance parameters.

3. The human-guided robot-environment interaction adaptive control method according to claim 2, wherein The expression of the state variables is: wherein, is the desired trajectory, is the first adjustable matrix, is the second adjustable matrix, is the intermediate state variable, is the derivative of the intermediate state variable; The expression of the linear state - space system is: Among them, , is to construct a diagonal matrix, is the first positive definite diagonal coefficient matrix, is the second definite diagonal coefficient matrix, is the introduced composite state variable, , is the derivative of the composite state variable, x is the trajectory; The expression of the cost function of the impedance model is: Among them, is the reference trajectory, is the actual interaction force, is the desired interaction force, , is the tracking error weight, is the cost function of the impedance model, , is the interaction force weight; 。 4. The human-guided robot-environment interaction adaptive control method according to claim 2, characterized in that The method for using adaptive optimal impedance learning to solve the minimization of the cost function includes: When is adopted, is used as the input interaction force, and the following calculation is performed: Among them, is the rank of the matrix, is a custom matrix for subsequent calculations, is a custom matrix for subsequent calculations, is the dimension of, is the dimension of, is the assumed initial interaction force, is the assumed initial impedance gain, is the iteration termination threshold, is a custom matrix for subsequent calculations, is a custom matrix for subsequent calculations, is a custom matrix for subsequent calculations, is a custom matrix for subsequent calculations, is an m-dimensional identity matrix, is a custom matrix for subsequent calculations, is the matrix the (m - 1)-th element; When is true, solve to obtain the optimal impedance parameter ; where is is the operator that expands into , and is the impedance parameter during iteration; Among them, is a custom matrix for subsequent calculations, is an m-dimensional identity matrix, is the interaction force weight, is a custom matrix for subsequent calculations; Among them, , is the trajectory error weight.

5. The human-guided robot-environment interaction adaptive control method according to claim 1, characterized in that, The method for learning the reference trajectory guided by humans through a neural network includes: With The expected trajectory corresponding to the parameter value corresponding to the trajectory parameterization is used as the input of the neural network to obtain the reference trajectory; Among them, the neural network includes two hidden layers, and the node matrices of the two hidden layers include: Among them, is the node matrix of the first hidden layer, is the node matrix of the second hidden layer, is the first conversion function, is the second conversion function, is the first connection weight, is the second connection weight, is the set of desired trajectories, is the first bias, is the second bias; The calculation method of the reference trajectory includes: Among them, is the third connection weight, the third bias.

6. The human-guided robot-environment interaction adaptive control method according to claim 1, wherein The method for real - time correcting the Jacobian matrix of the robot through sensor data includes: Define the data - driven error through the following formula: wherein, is the data-driven error function, is the end-effector velocity of the robotic arm predicted based on the Jacobian estimation, is the actual end-effector velocity of the robotic arm collected by the sensor, is the estimated Jacobian matrix, is the actual robotic arm joint velocity; Correct the Jacobian matrix through the following formula: Among them, is the corrected Jacobian matrix, is the learning rate, is the gradient; The method for obtaining the reference joint velocity and estimated acceleration in the joint space through closed - loop inverse kinematics includes: where, is the reference joint velocity, is a positive definite finite matrix, is the trajectory tracking error.

7. The human-guided robot-environment interaction adaptive control method according to claim 1, wherein The method for obtaining the output of the bias - width fuzzy neural network includes: Among them, is the output of the bias-width fuzzy neural network, is the updated node matrix, is the expanded connection weight matrix, is the expanded fuzzy rule matrix, is the expanded enhanced node matrix, is the fuzzy rule matrix before expansion, is the th fuzzy rule, is the enhanced node matrix before expansion, is the result defined by the transformation function corresponding to the th enhanced node using the triangular expansion scheme, is the introduced bias, is the number of enhanced nodes; The method for obtaining the manipulator control torque includes: Introduce the output of the bias - width fuzzy neural network into the controller to obtain the manipulator control torque: wherein, is the control torque of the robotic arm, , and are the approximation results of the uncertain robot dynamic model using the bias-width fuzzy neural network, is the control gain, is the joint velocity, is the joint tracking error, is the interaction force of the environment, is the virtual control term, is the derivative of the virtual control term.

8. An adaptive control system for human-guided robot-environment interaction, characterized in that, Including: An optimization module for estimating the optimal impedance parameters through a state - space model to minimize the force / position tracking error of the robot's interaction with the environment; And combined with the optimal impedance parameters, learn the reference trajectory guided by humans through a neural network to optimize the interaction performance; A correction module for real - time correcting the Jacobian matrix of the robot through sensor data; A conversion module for combining the robot's Jacobian matrix and the reference trajectory guided by humans, through closed - loop inverse kinematics, to obtain the reference joint velocity and estimated acceleration in the joint space; meanwhile, combined with the estimated acceleration, synchronously adjust the robot's Jacobian matrix through Jacobian adaptation; An approximation control module for approximating according to the actual manipulator joint velocity through an uncertain dynamics model to obtain the network parameters of the robot dynamics model; Combined with the reference joint velocity in the joint space and the network parameters, obtain the manipulator control torque through a bias - width fuzzy neural network.

9. An electronic device, characterized in that, Including: A memory and one or more processors; the memory is coupled to the processor; wherein, computer program code is stored in the memory, and the computer program code includes computer instructions, and when the computer instructions are executed by the processor, the electronic device executes the steps of the human-guided robot-environment interaction adaptive control method according to any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, A computer program is stored in the computer-readable storage medium, and when the computer program is executed by a processor, the steps of the human-guided robot-environment interaction adaptive control method according to any one of claims 1-7 are implemented.

Images