Adaptive force control hierarchical shared control method and system based on variable impedance and prior neural network

Abstract

The application discloses a self-adaptive force control hierarchical shared control method and system based on variable impedance and prior neural network, relates to the technical field of robot control, and discloses the self-adaptive force control hierarchical shared control method and system based on variable impedance and prior neural network, which can effectively reduce the cognitive load of an operator and improve the stability of remote operation contact force control and operation precision through hierarchical shared control, dynamic adjustment of damping parameters according to human-computer interaction force of a master end to obtain a smooth trajectory, adaptive processing of nonlinear characteristics of an environment of a slave end through a prior neural network, and relieving of the pressure of multi-dimensional information processing of the operator, and solving the problem that fixed parameter control cannot adapt to unknown nonlinear environment and leads to unstable force control.

Description

Technical Field

[0001] This application relates to the field of robot control technology, and in particular to an adaptive force control hierarchical shared control method and system based on variable impedance and prior neural networks. Background Technology

[0002] Teleoperated robotic systems play a crucial role in high-risk or specialized operating environments such as machinery manufacturing and aerospace, enabling operators to remotely control slave robotic arms to perform precision tasks from a safe distance. However, traditional teleoperated systems have significant drawbacks. Operators must simultaneously process image information from a visual display, tactile signals from a force feedback device, and control inputs from manually operating the master device. This multi-dimensional information processing leads to a sharp increase in cognitive load, resulting in decreased operational efficiency and a higher error rate. Especially in complex operating scenarios, operator distraction is more pronounced, easily causing trajectory deviations or accidental collisions. Furthermore, slave robotic arms struggle to maintain stable contact forces when interacting with the environment. When environmental dynamics are unknown or exhibit nonlinear changes, such as in flexible material processing or irregular surface operations, fixed-parameter impedance control methods cannot be dynamically adjusted, leading to frequent force control overshoot, trajectory jitter, and contact force fluctuations. These problems not only affect machining accuracy and workpiece quality but can also damage precision components due to unstable contact forces. In existing technologies, force control strategies based on linear models cannot accurately describe the nonlinear characteristics of the environment, such as hysteresis effects and nonlinear stiffness changes, while compliant control mechanisms with fixed damping parameters lack environmental adaptability and are difficult to cope with dynamically changing operating conditions.

[0003] The above content is only used to help understand the technical solution of this application and does not represent an admission that the above content is prior art. Summary of the Invention

[0004] The main objective of this application is to provide an adaptive force control hierarchical shared control method and system based on variable impedance and prior neural network, which aims to improve the stability and operational accuracy of contact force control during remote operation.

[0005] To achieve the above objectives, this application proposes an adaptive force-controlled hierarchical shared control method based on variable impedance and prior neural networks. The method includes: The real-time position data of the master robotic arm is acquired, and the real-time position data is processed by Kalman filtering to obtain filtered position data. Based on the filtered position data and the human-machine interaction external torque data obtained from the master robotic arm, smooth trajectory data is obtained through variable impedance compliant control; wherein, the variable impedance compliant control dynamically adjusts the damping parameters according to the human-machine interaction external torque data. The smooth trajectory data is scaled and mapped according to a preset ratio coefficient to obtain the desired trajectory data from the slave end. Real-time contact force data between the slave robotic arm and the environment is acquired. Based on the real-time contact force data and the slave-end desired trajectory data, the data is processed by a priori neural network adaptive force controller to obtain updated trajectory data. The priori neural network adaptive force controller is adaptively adjusted according to the nonlinear dynamics model of the environment. Based on the updated trajectory data, the control torque data is calculated by the trajectory tracking controller; The control torque data is sent to the motor of the slave robotic arm, driving the slave robotic arm to perform the corresponding action.

[0006] In one embodiment, the step of obtaining smooth trajectory data through variable impedance compliant control processing based on the filtered position data and the human-machine interaction external torque data obtained from the master robotic arm includes: Obtain the reference trajectory position data, reference trajectory velocity data, and reference trajectory acceleration data corresponding to the filtered position data, and obtain the human-computer interaction external torque data; Based on the reference trajectory position data, reference trajectory velocity data, reference trajectory acceleration data, and the human-machine interaction external torque data, smooth trajectory data is calculated using a variable impedance rigid-flexible control structure; wherein, the variable impedance rigid-flexible control structure includes a variable damping function, and the variable damping function dynamically adjusts the damping value according to the rate of change of the human-machine interaction external torque data.

[0007] In one embodiment, the calculation of the variable damping function includes: Calculate the rate of change of the external torque data in the human-computer interaction; The first damping component is calculated using a nonlinear function based on the absolute value of the rate of change. The first damping component is added to the preset basic damping value to obtain the final damping value.

[0008] In one embodiment, the step of scaling and mapping the smoothed trajectory data according to a preset scaling factor to obtain the desired trajectory data from the slave end includes: Get the slave position data from the previous moment; Calculate the difference between the filtered position data at the current time and the filtered position data at the previous time. Multiply the difference data by the preset scaling factor to obtain the location increment data; The incremental position data is added to the slave position data of the previous moment to obtain the slave desired trajectory data.

[0009] In one embodiment, the step of acquiring real-time contact force data between the slave robotic arm and the environment, and processing the real-time contact force data and the slave-end desired trajectory data using a priori neural network adaptive force controller to obtain updated trajectory data includes: Based on the expected trajectory data from the slave end, predict the environmental contact location data; Based on the real-time contact force data and the environmental contact location data, an environmental dynamics model is constructed, which includes linear spring terms and nonlinear terms. The nonlinear term is modeled by a pre-trained neural network initialized based on prior knowledge, and the neural network output data is obtained, wherein the prior knowledge includes pre-measured environmental parameter data. Based on the linear spring term, the neural network output data, and the real-time contact force data, the force error data is calculated. Based on the force error data, the neural network weight data is updated using an adaptive law; The updated trajectory data is calculated based on the updated neural network weight data and the environmental dynamics model.

[0010] In one embodiment, the step of updating the neural network weight data using an adaptive law based on the force error data includes: Based on the force error data and the preset learning rate, the weight update data is calculated; The updated weight data is added to the current neural network weight data to obtain the updated neural network weight data.

[0011] In one embodiment, the step of calculating control torque data by the trajectory tracking controller based on the updated trajectory data includes: Based on the updated trajectory data, the expected joint angle data and the expected joint angular velocity data are calculated; Acquire the actual joint angle data and actual joint angular velocity data of the slave robotic arm; Calculate the angle error data between the expected joint angle data and the actual joint angle data, and the angular velocity error data between the expected joint angular velocity data and the actual joint angular velocity data; Based on the angle error data and the angular velocity error data, the control torque data is calculated by a proportional-derivative controller.

[0012] In one embodiment, the method further includes: Acquire real-time contact force data from the slave robotic arm; The real-time contact force data is scaled according to a preset ratio to obtain the main end feedback force data; The master-end feedback force data is applied to the master-end robotic arm, allowing the operator to perceive the contact force between the slave end and the environment.

[0013] In one embodiment, the method further includes: The working image data of the robotic arm is collected in real time by a camera; The work screen data is displayed on the operator's screen so that the operator can monitor the status of the slave device.

[0014] Furthermore, to achieve the above objectives, this application also proposes an adaptive force-controlled hierarchical sharing control system based on variable impedance and a priori neural network. The adaptive force-controlled hierarchical sharing control system based on variable impedance and a priori neural network includes: a memory, a processor, and an adaptive force-controlled hierarchical sharing control program based on variable impedance and a priori neural network stored in the memory and executable on the processor. The adaptive force-controlled hierarchical sharing control program based on variable impedance and a priori neural network is configured to implement the steps of the adaptive force-controlled hierarchical sharing control method based on variable impedance and a priori neural network.

[0015] The adaptive force control hierarchical shared control method and system proposed in this application, based on variable impedance and prior neural network, achieves a smooth trajectory by dynamically adjusting damping parameters at the master end according to the human-machine interaction force through hierarchical shared control. At the slave end, the nonlinear characteristics of the environment are adaptively processed through prior neural network, which alleviates the pressure of multi-dimensional information processing for the operator. At the same time, it solves the problem that fixed parameter control cannot adapt to unknown nonlinear environments, resulting in unstable force control. It can effectively reduce the cognitive load of the operator and improve the stability and operation accuracy of remote operation contact force control. Attached Figure Description

[0016] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0017] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart illustrating an embodiment of the adaptive force-controlled hierarchical shared control method based on variable impedance and prior neural network of this application. Figure 2 This is a schematic diagram of a structure provided for an embodiment of the adaptive force-controlled hierarchical shared control system based on variable impedance and prior neural network of this application; Figure 3This is a schematic diagram of the three-dimensional trajectory of the master-slave mapping under shared control in this application; Figure 4 This is a schematic diagram of the two-dimensional trajectory in each of the XYZ directions under the shared control of this application; Figure 5 This is a schematic diagram of the experimental results of the force tracking effect of the end-effector robot in this application.

[0019] Explanation of icon numbers: 10. Memory; 20. Processor.

[0020] The purpose, features, and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0021] The technical solutions of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. The components of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of this application, but merely represents selected embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0022] It should be understood that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. Furthermore, in the description of this application, the terms "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0023] In existing technologies, teleoperated robot systems suffer from problems such as excessive cognitive load on operators, low operational efficiency, and insufficient force control accuracy and stability of the slave robotic arm in complex and unknown environments when performing precision tasks. In existing technologies, impedance control with fixed parameters is difficult to adapt to dynamically changing working environments, while force control methods based on linear models cannot effectively handle the nonlinear characteristics of the environment, resulting in limited work quality and potential workpiece damage risks.

[0024] Based on this, embodiments of this application provide an adaptive force-controlled hierarchical shared control method based on variable impedance and prior neural networks, referring to... Figure 1 The adaptive force-controlled hierarchical shared control method based on variable impedance and prior neural network includes steps S100 to S600, wherein: Step S100: Obtain the real-time position data of the master robotic arm, and perform Kalman filtering on the real-time position data to obtain filtered position data; Step S200: Based on the filtered position data and the human-machine interaction external torque data obtained from the main robotic arm, smooth trajectory data is obtained through variable impedance compliant control processing; wherein, the variable impedance compliant control dynamically adjusts the damping parameters according to the human-machine interaction external torque data. Step S300: The smooth trajectory data is scaled and mapped according to a preset scaling factor to obtain the desired trajectory data from the slave end; Step S400: Obtain real-time contact force data between the slave robotic arm and the environment; based on the real-time contact force data and the slave desired trajectory data, process the data using a priori neural network adaptive force controller to obtain updated trajectory data; wherein, the priori neural network adaptive force controller adaptively adjusts according to the nonlinear dynamics model of the environment. Step S500: Based on the updated trajectory data, calculate the control torque data through the trajectory tracking controller; In step S600, the control torque data is sent to the motor of the slave robotic arm to drive the slave robotic arm to perform the corresponding action.

[0025] To more clearly describe the technical solution of this application, the Cartesian space dynamics of the master and slave robotic arms are first modeled. For the master robotic arm, its Cartesian space dynamic equation can be expressed as: ; in For the control torque of the robotic arm itself, Human-computer interaction power It is a scaled-down environmental contact force from the slave end. and as well as These represent the inertia, centripetal force, and gravity terms of the main arm, respectively. and This indicates the speed and acceleration of the master robotic arm.

[0026] Similarly, the dynamic model of the slave robotic arm in Cartesian space is as follows: ; in, and as well as This represents the inertial term, centripetal force term, and gravity term from the starting point. Indicates the control torque from the slave end. It is interactive force. and This indicates the speed and acceleration of the robotic arm at the end.

[0027] In this architecture, the operator controls the movement of the master robotic arm by applying interactive forces, and these motion commands are mapped to the slave robotic arm. The slave robotic arm, based on a designed controller, simultaneously tracks the desired trajectory generated by the operator's intent and the desired contact force required to satisfy the manufacturing process. This collaborative human-machine working mode, jointly completing the control task, is defined as shared control.

[0028] In this embodiment, Kalman filtering is an algorithm that uses the linear system state equations and system input / output observation data to optimally estimate the system state. Its function is to extract more accurate state estimates from noisy measurement data, thereby improving data accuracy and smoothness. Variable impedance compliant control is a robot control strategy that adjusts the interaction force between the robot and its environment by simulating the impedance characteristics of a mechanical system. Its core lies in dynamically adjusting these impedance parameters according to the actual interaction, enabling the robot to compliantly adapt to environmental changes and achieve force control or trajectory tracking. Human-robot interaction external torque data refers to the external torque information applied to the system by the operator through the master robotic arm. This data reflects the operator's intention and expectations for the slave robotic arm's movement, and is an important component of the master input in shared control.

[0029] In this embodiment, the damping parameter is a key parameter in variable impedance compliant control, used to describe the system's resistance to motion. Dynamically adjusting the damping parameter can change the robot arm's response characteristics to external disturbances, thereby affecting its compliance and stability. The prior neural network adaptive force controller combines prior knowledge with the learning capabilities of neural networks. This controller uses pre-set knowledge or models as initial conditions and, through the adaptive learning capabilities of neural networks, adjusts control parameters online to adapt to unknown or changing system dynamics, making it particularly suitable for handling nonlinear environments. The environmental nonlinear dynamics model refers to a mathematical model describing the complex nonlinear interaction between the robot arm and its environment. This model can capture the nonlinear characteristics of the environment, such as contact stiffness and friction, providing the force controller with more accurate environmental information. The trajectory tracking controller is used to ensure that the robot arm's actual motion trajectory accurately follows the desired trajectory. This controller calculates the required control torque or force to eliminate the deviation between the actual trajectory and the desired trajectory, ensuring that the robot arm moves along a preset path.

[0030] In this embodiment, the adaptive force control hierarchical shared control method based on variable impedance and prior neural networks first acquires the real-time position data of the master robotic arm. This real-time position data can be directly collected through position sensors or encoders on the master robotic arm. To improve the accuracy of the data, the real-time position data can be processed, for example, by using a basic moving average filter to average the real-time position data of the most recent sampling points to obtain filtered position data.

[0031] Specifically, the main trajectory is acquired in real time through human-computer interaction, and after Kalman filtering, the position of the main terminal at time t can be represented as: ; and This indicates that Kalman filtering is performed on the original data. It is the first of the main end At any given moment, This is the size of the filtering window.

[0032] In this embodiment, based on the filtered position data and the human-machine interaction external torque data obtained from the master robotic arm, smooth trajectory data is obtained through variable impedance compliant control. This variable impedance compliant control can employ a basic impedance model, where the damping parameter is linearly adjusted according to the magnitude of the human-machine interaction external torque data. For example, when the human-machine interaction external torque data increases, the damping parameter can increase by a preset linear gain to increase the system's compliance; when the human-machine interaction external torque data decreases, the damping parameter can decrease by a preset linear gain.

[0033] In this embodiment, the smooth trajectory data is scaled and mapped according to a preset scaling factor to obtain the desired trajectory data for the slave end. This scaling mapping can be achieved by directly multiplying the smooth trajectory data by a fixed scaling factor. For example, if the range of motion of the master robotic arm is X and the working range of the slave robotic arm is Y, the scaling factor can be set to Y / X, thereby mapping the master trajectory proportionally to the slave workspace.

[0034] After Kalman filtering and scaling, the master-end trajectory is mapped onto the slave-end robotic arm control system for further calculations. The slave-end position at time t can be represented as: ; in, The scaling factor is the mapping coefficient. From end Location at any given moment It represents the position of the slave at the previous moment. This incremental mapping method ensures the continuity and smoothness of the slave trajectory.

[0035] In this embodiment, real-time contact force data between the slave robotic arm and the environment is acquired. This real-time contact force data can be obtained through a force sensor installed at the end of the slave robotic arm. Based on this real-time contact force data and the desired trajectory data from the slave end, it is processed by a priori neural network adaptive force controller to obtain updated trajectory data. This priori neural network adaptive force controller can employ a neural network structure with a fixed number of layers and neurons, and its initial weights can be set based on experience or a basic mathematical model. The neural network adjusts its internal weights according to the deviation between the real-time contact force data and the desired force using a basic backpropagation algorithm, thereby achieving adaptive adjustment to the nonlinear dynamics model of the environment and outputting updated trajectory data.

[0036] In this embodiment, control torque data is calculated based on the updated trajectory data using a trajectory tracking controller. This trajectory tracking controller can be a basic proportional controller that directly calculates the required control torque data based on the positional deviation between the updated trajectory data and the current position of the slave robotic arm. Finally, the calculated control torque data is sent to the motors of the slave robotic arm. This control torque data is transmitted to the motor drivers of each joint via the robotic arm's control bus, driving the motors of the slave robotic arm to perform corresponding actions, thereby achieving precise control of the slave robotic arm.

[0037] In this embodiment, by introducing variable impedance compliance control, the damping parameters are dynamically adjusted to adapt to the operator's intentions, effectively reducing the operator's cognitive load. Simultaneously, combined with a priori neural network adaptive force controller, the interaction between the slave robotic arm and the unknown nonlinear environment can be precisely handled, ensuring stable constant force contact in complex operations, avoiding force control overshoot and trajectory jitter, thereby improving work efficiency and processing quality, and reducing the risk of damage to precision workpieces.

[0038] In some embodiments described above in this application, damping parameters are dynamically adjusted based on the external torque data from human-computer interaction using variable impedance compliance control to obtain smooth trajectory data. However, in actual human-computer interaction, the external torque applied by the operator may change rapidly. If the damping parameters are adjusted only based on the instantaneous value of the external torque, the damping response may be delayed or excessive, thereby affecting the smoothness of the trajectory and the stability of the system, making it difficult to effectively cope with complex and ever-changing human-computer interaction scenarios.

[0039] To address this, this application further proposes a step for obtaining smooth trajectory data based on filtered position data and human-machine interaction external torque data obtained from the master robotic arm, through variable impedance compliant control processing. This step includes: acquiring reference trajectory position data, reference trajectory velocity data, and reference trajectory acceleration data corresponding to the filtered position data, and acquiring human-machine interaction external torque data; calculating smooth trajectory data based on the reference trajectory position data, the reference trajectory velocity data, the reference trajectory acceleration data, and the human-machine interaction external torque data using a variable impedance rigid-flexible control structure; wherein the variable impedance rigid-flexible control structure includes a variable damping function, and the variable damping function dynamically adjusts the damping value according to the rate of change of the human-machine interaction external torque data.

[0040] In this embodiment, the filtered position data is the position information of the master robotic arm after Kalman filtering, which eliminates measurement noise and provides a more accurate current position. Based on this filtered position data, the corresponding reference trajectory position data, reference trajectory velocity data, and reference trajectory acceleration data can be derived or calculated. These data collectively describe the motion state of the master robotic arm on the desired trajectory, providing a foundation for compliant control. The human-machine interaction external torque data directly reflects the operator's intention and the force applied to the master robotic arm. Acquiring these data is a prerequisite for precise compliant control, ensuring that the system can understand the operator's motion intention and respond accordingly. Specifically, the reference trajectory position data can be directly obtained from the filtered position data, while the reference trajectory velocity data and reference trajectory acceleration data can be obtained by performing first-order and second-order differences on the filtered position data or by estimating them through a state observer. The human-machine interaction external torque data is usually obtained by directly measuring the force / torque sensor installed on the master robotic arm.

[0041] In this embodiment, the variable impedance rigid-flexible control structure is an advanced compliant control strategy that combines the precision of rigid control with the adaptability of compliant control. This structure uses the position, velocity, and acceleration information of a reference trajectory as desired inputs and combines them with external torque data from human-machine interaction to adjust the dynamic response of the robotic arm. Its core lies in its ability to dynamically change the "stiffness" and "damping" characteristics of the robotic arm according to the magnitude and changes of the external interaction torque, thereby maintaining trajectory tracking capability while achieving compliant interaction with the operator's intentions. Through this control structure, the operator's intentions can be effectively translated into a smooth robotic arm motion trajectory, avoiding problems such as discomfort due to excessive stiffness or failure to follow the operator's intentions due to insufficient compliance. For example, this variable impedance rigid-flexible control structure can be constructed based on a mass-spring-damping model, where the mass term is typically fixed, while the spring and damping terms are dynamically adjusted according to external interaction to calculate the desired smooth trajectory data.

[0042] Specifically, the formula for the designed variable impedance rigid-flexible control structure is as follows: ; in It is a nonlinear damping function that is related to external forces. These are the position, velocity, and acceleration of the reference trajectory, respectively. These are the desired position, velocity, and acceleration, respectively. and It is the mass and stiffness matrix. The external torque for human-computer interaction obtained through decoupling.

[0043] In this embodiment, the variable damping function is a key component of the variable impedance rigid-flexible control structure. It is responsible for adjusting the system's damping value based on the dynamic characteristics of the external torque data from human-machine interaction. Traditional impedance control may adjust damping only based on the instantaneous magnitude of the external torque, which may lead to hysteresis or overshoot when the torque changes rapidly. By introducing the rate of change of the external torque data from human-machine interaction, the variable damping function can predict the changing trend of the operator's intention, thereby adjusting the damping value more proactively. When the rate of change of the external torque is large, the system can quickly increase the damping to absorb the impact and prevent trajectory oscillation; when the rate of change of the external torque is small, the damping can be appropriately reduced to improve the system's response sensitivity. This dynamic adjustment mechanism based on the rate of change enables the system to respond more intelligently and smoothly to the operator's complex operations, improving the naturalness and comfort of human-machine interaction. For example, the variable damping function can be designed as a nonlinear function, with its input being the absolute value of the rate of change of the external torque data from human-machine interaction, to achieve dynamic adjustment of the damping value.

[0044] In this embodiment, based on the filtered position data and human-machine interaction external torque data obtained through the above technical solution, this application further introduces reference trajectory position data, reference trajectory velocity data, and reference trajectory acceleration data, and employs a variable impedance rigid-flexible control structure to calculate smooth trajectory data. The unique feature of this variable impedance rigid-flexible control structure is that it includes a variable damping function, which dynamically adjusts the damping value according to the rate of change of the human-machine interaction external torque data. This design enables the system not only to respond to the operator's instantaneous torque but also to predict and adapt to rapid changes in torque, effectively avoiding the problem of damping response lag or over-damping caused by rapid changes in external torque. Through intelligent adjustment of the damping value, the robotic arm exhibits better compliance and stability when interacting with the operator, generating a smoother trajectory, improving the naturalness and comfort of human-machine interaction, and making the system more adaptable and robust in complex and ever-changing human-machine interaction scenarios.

[0045] In some embodiments described above in this application, a variable impedance compliant control is proposed that dynamically adjusts damping parameters based on human-machine interface (HMI) external torque data. The variable impedance compliant control structure includes a variable damping function, which dynamically adjusts the damping value based on the rate of change of the HMI external torque data. However, in its implementation, simply adjusting the damping value based on the rate of change of the HMI external torque data may lead to an overly sensitive damping response, especially when the HMI external torque data contains noise or rapid fluctuations, potentially causing system instability or operator discomfort.

[0046] In response, this application further proposes that the calculation of the variable damping function includes: calculating the rate of change of the human-machine interaction external torque data; calculating the first damping component based on the absolute value of the rate of change using a nonlinear function; and adding the first damping component to a preset basic damping value to obtain the final damping value.

[0047] In this embodiment, calculating the rate of change of the human-computer interaction external torque data refers to obtaining the rate of change over time by performing time differentiation or difference operations on the continuously collected human-computer interaction external torque data. For example, the finite difference method can be used, that is, subtracting the human-computer interaction external torque data from the previous moment from the current moment's human-computer interaction external torque data, and then dividing by the time interval to obtain its approximate rate of change. To improve the accuracy and noise resistance of the rate of change, signal processing methods such as Kalman filtering can also be used to smooth the original data before calculating the rate of change.

[0048] In this embodiment, the first damping component is calculated using a nonlinear function based on the absolute value of the rate of change. This means first obtaining the absolute value of the rate of change of the human-machine interaction external torque data to ensure that the damping component is always positive and that only the amplitude of change is considered, unaffected by direction. Subsequently, this absolute value is input into a pre-designed nonlinear function to calculate the first damping component. This nonlinear function can be designed according to actual needs; for example, it can be a piecewise function, an exponential function, or an sigmoid function. Through the nonlinear function, differentiated responses to different rates of change can be achieved. For example, when the rate of change is small, the first damping component may increase slowly to maintain system stability; when the rate of change is large, the first damping component may increase rapidly to quickly increase damping and suppress oscillations or overshoot. This nonlinear mapping helps the system exhibit more intelligent and stable compliance characteristics when facing different interaction intensities.

[0049] Specifically, the calculation expression for variable damping can be written as: ; in, and All of these are parameters of a function of variable damping. This is the initial damping value. The rate at which torque is expressed. This is achieved by adjusting... and The parameters can be flexibly controlled to adjust the damping response characteristics as the rate of change of torque, thereby achieving different compliance control effects.

[0050] In this embodiment, the first damping component is added to a preset basic damping value to obtain the final damping value. The preset basic damping value is a constant determined during the system design or debugging phase; it represents the minimum damping level the system should possess when there are no significant changes in the external torque generated by human-machine interaction. This basic damping value ensures that the system has a certain degree of stability under all circumstances, avoiding system oscillations that may result from excessively low damping values. By superimposing the dynamically calculated first damping component with the preset basic damping value, the final damping value includes both the real-time response to the rate of change of the external torque data generated by human-machine interaction and guarantees the inherent stability and compliance of the system.

[0051] In this embodiment, the above technical solution firstly, by calculating the rate of change of the human-computer interaction external torque data and taking its absolute value, the dynamic intensity of the human-computer interaction can be accurately captured, while avoiding the complexity of damping adjustment logic caused by different directions of change. Secondly, a nonlinear function is introduced to calculate the first damping component, so that the damping adjustment is no longer a simple linear relationship, but can make a more refined and intelligent response based on the magnitude of the rate of change of the human-computer interaction external torque data. For example, for small changes that may be caused by noise, the nonlinear function can reduce its impact on damping, thereby enhancing the system's anti-interference ability; while for significant changes caused by the operator's intention, the nonlinear function can quickly increase damping to provide stronger compliance and stability. Finally, the dynamically calculated first damping component is added to a preset base damping value to ensure that the system always has a stable reference damping. Even when the change of the human-computer interaction external torque is not significant, the inherent compliance and stability of the system can be maintained, effectively avoiding system oscillation or discomfort that may be caused by excessively low damping. Overall, this damping calculation method makes variable damping compliant control more robust and adaptive, improving the smoothness, comfort, and safety of human-machine interaction. This results in smoother trajectory generation of the master robotic arm, allowing operators to obtain a more natural and intuitive force feedback experience.

[0052] In some embodiments described above in this application, a method is proposed to scale and map smooth trajectory data according to a preset scaling factor to obtain the desired trajectory data from the slave end. However, in its implementation, if the smooth trajectory data is simply scaled absolutely, the motion trajectory of the slave end robot arm may become discontinuous or unsmooth, especially when the scaling factor is large or the environmental interaction is complex, making it difficult to ensure the stability and accuracy of the slave end motion.

[0053] To address this, this application further proposes a step of scaling and mapping the smoothed trajectory data according to a preset scaling factor to obtain the desired trajectory data from the slave end, including: acquiring the slave end position data at the previous moment; calculating the difference between the filtered position data at the current moment and the filtered position data at the previous moment; multiplying the difference data by the preset scaling factor to obtain position increment data; and adding the position increment data to the slave end position data at the previous moment to obtain the desired trajectory data from the slave end.

[0054] In this embodiment, the control system periodically acquires the current actual position information of the slave robotic arm from its joint encoders, end effector sensors, or other position measurement devices. This real-time position data is stored at the end of each control cycle and read and used as the "slave position data from the previous moment" at the beginning of the next control cycle. In this way, the system can always plan its future movements based on the actual historical state of the slave robotic arm, ensuring a smooth transition of the trajectory.

[0055] In this embodiment, the filtered position data obtained by Kalman filtering at the current moment is vector subtracted from the filtered position data stored in the previous control cycle. This difference calculation can effectively filter out the influence of the absolute position drift or initial position difference of the master robot arm on the slave motion planning, so that the subsequent scaling mapping focuses on the changes in the master operator's motion intention, thereby improving the accuracy and responsiveness of the slave motion.

[0056] In this embodiment, the preset scaling factor can be a scalar used to uniformly scale the motion of all degrees of freedom; it can also be a vector or matrix, allowing different scaling ratios to be applied to different degrees of freedom to meet the specific task's requirements for operational precision or range of motion. This scaling factor is typically calibrated offline or adjusted online during operation based on the actual application scenario (e.g., a smaller scaling factor when precise operation is required, and a larger scaling factor when a large range of movement is required) to optimize the human-computer interaction experience and task execution efficiency.

[0057] In this embodiment, the desired absolute position of the slave robot at the current moment is generated by superimposing the relative motion increment of the slave robot onto its actual position at the previous moment. This is a simple vector addition operation, which superimposes the calculated position increment data with the position data of the slave robot at the previous moment. This incremental superposition method ensures that the desired trajectory of the slave robot is always continuous and smooth, avoiding trajectory jumps or discontinuities that may be caused by directly scaling the absolute position, thus providing a stable and reliable input for the subsequent trajectory tracking control of the slave robot.

[0058] In this embodiment, the previous position data of the slave robot arm is used as a reference, and the difference between the filtered position data of the master robot arm at the current moment and the previous moment is calculated to obtain the relative motion intention of the master robot arm. Then, this difference data is multiplied by a preset scaling factor to obtain the incremental position data of the slave robot arm. Finally, this incremental position data is added to the previous position data of the slave robot arm to generate the desired trajectory data of the slave robot arm. This incremental trajectory generation method effectively avoids the problem of discontinuous or unsmooth slave trajectory that may be caused by directly scaling the absolute position of the master robot arm. It ensures that the movement of the slave robot arm is always a smooth transition based on its current state. Even when the scaling factor is large or the master operator makes rapid and large-amplitude movements, the stability and predictability of the slave robot arm's movement are maintained. This improves the operational accuracy and smoothness of the shared control system in complex environments, allowing the operator to control the slave robot arm to perform fine tasks more intuitively and stably.

[0059] In some embodiments described above in this application, an adaptive force control hierarchical shared control method based on variable impedance and a priori neural network is proposed. This method achieves compliant interaction between master and slave robotic arms through the collaborative operation of variable impedance compliant control and a priori neural network adaptive force controller. However, in practical applications, the interaction between the slave robotic arm and the environment often involves complex nonlinear dynamics. These nonlinear characteristics are difficult to accurately capture with simple models, which may lead to decreased force control accuracy, impaired system stability, or even failure to achieve effective compliant operation in unknown or changing environments.

[0060] To address this, this application further proposes a step for obtaining real-time contact force data between the slave robotic arm and the environment, and processing the real-time contact force data and the slave-end desired trajectory data using a priori neural network adaptive force controller to obtain updated trajectory data. This step includes: predicting environmental contact position data based on the slave-end desired trajectory data; constructing an environmental dynamics model based on the real-time contact force data and the environmental contact position data, the environmental dynamics model including linear spring terms and nonlinear terms; modeling the nonlinear terms using a pre-trained neural network initialized based on prior knowledge to obtain neural network output data, wherein the prior knowledge includes pre-measured environmental parameter data; calculating force error data based on the linear spring terms, the neural network output data, and the real-time contact force data; updating the neural network weight data using an adaptive law based on the force error data; and calculating the updated trajectory data based on the updated neural network weight data and the environmental dynamics model.

[0061] In this embodiment, when predicting environmental contact location data, this step aims to estimate the possible location where the slave robot arm will come into contact with the environment based on the expected motion trend of the slave arm. This is typically achieved by analyzing the expected trajectory data of the slave arm, combining the geometric model of the slave robot arm with coarse information about the potential environment (such as workspace boundaries), and performing forward kinematics calculations or geometric collision detection. The predicted environmental contact location data provides key reference points for the subsequent construction of the environmental dynamics model, enabling the model to focus on the actual interaction area.

[0062] In this embodiment, the step of constructing the environmental dynamics model aims to establish a mathematical model that accurately describes the mechanical relationship between the slave manipulator and the environment. The environmental dynamics model is designed to include two parts: linear spring terms and nonlinear terms. Linear spring terms capture the basic elastic properties of the environment; for example, when the slave manipulator contacts a rigid surface, its mechanical response can be approximated as a linear spring. Nonlinear terms characterize more complex environmental properties that are difficult to describe with simple linear relationships, such as friction, viscous damping, plastic deformation of materials, or irregular surface geometry. This decompositional modeling approach allows the system to handle both linear and nonlinear environmental behaviors simultaneously, improving the model's universality and accuracy.

[0063] Specifically, environmental dynamics are nonlinear, and this application empirically models environmental dynamics as follows: ; middle This represents a nonlinear dynamics model term related to the location of the unknown environment. Indicates environmental stiffness. and These represent the position of the robotic arm's end effector and the position of the environment, respectively. The model decomposes the environmental forces into linear spring terms. and nonlinear terms The two parts allow for a more accurate description of the mechanical properties of complex environments.

[0064] In this embodiment, to effectively handle the nonlinear terms of the environment, this application employs a pre-trained neural network initialized with prior knowledge. This neural network is specifically designed to learn and approximate the nonlinear dynamic characteristics of the environment. Prior knowledge, such as pre-measured environmental parameter data (e.g., known ranges of elastic moduli of materials, estimated values ​​of friction coefficients, or historical interaction data collected from similar environments), is used to initialize the weights of the neural network or for initial training. This pre-training and introduction of prior knowledge accelerates the convergence speed of the neural network, improves its modeling accuracy and robustness in the early stages of practical operation, avoids the inefficiency of learning from scratch, and enables the system to adapt more quickly to the nonlinear behavior of unknown environments and output neural network output data representing the nonlinear response of the environment.

[0065] The nonlinear term of the environment estimated using a neural network can be expressed as: ; in, These are the weights of the neural network. It is a radial basis function. This represents the approximation error of the network. The universal approximation property of neural networks allows for the effective modeling of complex nonlinear characteristics of the environment.

[0066] In this embodiment, after constructing the environmental dynamics model and obtaining the neural network output data, this step compares the model's predicted force with the actual measured real-time contact force data to calculate the force error data. Specifically, the environmental dynamics model (including linear spring terms and nonlinear terms represented by the neural network output data) predicts a desired contact force, while the real-time contact force data is the force actually felt by the end-effector. The force error data is the difference between these two; it quantifies the degree of mismatch between the current environmental model and the actual environment and is a key signal driving subsequent adaptive adjustments by the neural network.

[0067] Ignoring reconstruction error, the force errors of model prediction and sensor measurement are: ; in, It is a force measurement and predictive power The error between them. and These are the predicted environmental stiffness and location. It is the prediction result of environmental nonlinear terms.

[0068] Define intermediate variables: ; ; ; in, and These are the defined variables and the unknown parameters that need to be identified. This parameterization method allows the force error to be represented as the inner product of the parameter error and the regression vector, facilitating subsequent adaptive law design.

[0069] In this embodiment, to ensure that the environmental dynamics model can continuously and accurately reflect environmental characteristics, this application utilizes the calculated force error data to update the weight data of the neural network through an adaptive law. This adaptive law adjusts the connection weights within the neural network based on the magnitude and direction of the force error, enabling the neural network's output to gradually approximate the true nonlinear response of the environment. This online adaptive mechanism ensures that even if environmental characteristics change, the neural network can adjust its model in a timely manner, thereby maintaining the accuracy and stability of force control. The specific adaptive law can employ gradient descent, least squares, or other adaptive control algorithms based on Lyapunov stability.

[0070] To prove the convergence of the force error, a Lyapunov function is designed: ; in This represents the error between the predicted and actual values ​​of the parameters. It is the defined learning rate. It is a defined Lyapunov function.

[0071] The derivative of the Lyapunov function is: ; in, It is the derivative of the predicted value of the unknown parameter; Design Adaptive Law: ; Then we have: ; The above process demonstrates the convergence of the force error. In practical applications, the environmental parameters and the weights of the neural network are initialized a priori based on pre-measured environmental parameters to ensure rapid convergence of the force error and prevent overshoot.

[0072] In this embodiment, after the neural network weight data is updated using an adaptive law, the environmental dynamics model provides a more accurate description of the environment. Based on this updated environmental dynamics model, the system can calculate updated trajectory data that better meets the actual interaction requirements. This updated trajectory data is the final output of the prior neural network adaptive force controller. It takes into account the actual mechanical response of the environment, thereby guiding the trajectory tracking controller to generate precise control torque, ensuring that the slave robotic arm can achieve the desired mechanical behavior when interacting with the environment, such as maintaining a constant contact force or achieving compliant following.

[0073] In this embodiment, the above technical solution effectively addresses the challenges of force control accuracy and stability when a slave robotic arm interacts with a complex nonlinear environment. Specifically, by predicting environmental contact position data and constructing an environmental dynamics model including linear spring terms and nonlinear terms, the system can comprehensively characterize the mechanical properties of the environment. In particular, by using a pre-trained neural network initialized with prior knowledge to model the nonlinear terms, the system can quickly and robustly learn and adapt to the complex behavior of unknown or changing nonlinear environments. Force error data serves as a feedback signal, driving the adaptive law to continuously update the neural network weight data, ensuring that the environmental model always maintains high accuracy. Finally, the updated trajectory data calculated based on the updated environmental dynamics model can more accurately guide the movement of the slave robotic arm, thereby achieving more precise and stable force control when interacting with the environment. This improves the compliance and operational safety of the shared control system, enabling operators to perform remote precision operations more intuitively and reliably.

[0074] In some of the above embodiments, a priori neural network adaptive force controller adaptively adjusts the nonlinear dynamic model of the environment and updates the neural network weight data based on force error data using an adaptive law to achieve precise force control of the slave robotic arm. However, in practical applications, how to efficiently and stably update the neural network weights to ensure that the controller can quickly adapt to environmental changes and maintain good control performance is a problem that needs further refinement and resolution. If the weight update mechanism is unclear or inefficient, it may lead to a decrease in force control accuracy or a slow system response.

[0075] In response, this application further proposes a step of updating neural network weight data based on force error data using an adaptive law, including: calculating weight update amount data based on the force error data and a preset learning rate; adding the weight update amount data to the current neural network weight data to obtain the updated neural network weight data.

[0076] In this embodiment, force error data is a key indicator measuring the difference between the actual contact force and the expected contact force of the slave robotic arm, reflecting the deviation between the current control effect and the target. The preset learning rate is a hyperparameter used to control the adjustment magnitude of the neural network weights in each update; its magnitude directly affects the learning speed and convergence stability. The weight update amount data is calculated based on the force error data and the preset learning rate, indicating the direction and magnitude of adjustment required for each weight in the neural network. For example, gradient descent or its variants (such as stochastic gradient descent, Adam, etc.) can be used to calculate the weight update amount data. The weight update amount data can be represented as the preset learning rate multiplied by the gradient of the force error data with respect to the neural network weights; the gradient calculation can be implemented using the backpropagation algorithm.

[0077] In this embodiment, the weight update data is added to the current neural network weight data to obtain the updated neural network weight data. The current neural network weight data is the state parameter of the neural network at the previous moment, which determines the current mapping relationship of the neural network. The calculated weight update data is added to the current neural network weight data to obtain the updated neural network weight data. This process is the core of neural network learning. By continuously adjusting the weights, the neural network can gradually reduce the force error data, thereby more accurately modeling the nonlinear dynamics of the environment and improving the performance of the adaptive force controller. This iterative update mechanism allows the neural network to learn from the data and dynamically adjust its internal parameters as the environment changes to adapt to different contact conditions and task requirements.

[0078] In this embodiment, the specific update mechanism for neural network weight data is clearly defined through the above technical solution. The weight update amount is calculated based on force error data and a preset learning rate, and then added to the current neural network weight data. This ensures that the neural network weights are effectively and controllably adjusted. This explicit update strategy allows the prior neural network adaptive force controller to learn and adapt in a controlled manner based on the force error data generated by the real-time contact force data between the slave robotic arm and the environment and the slave's desired trajectory data. The introduction of a preset learning rate allows the system to balance learning speed and stability while ensuring convergence, avoiding problems such as system oscillation due to excessively fast weight updates or insufficient adaptability due to excessively slow updates. Ultimately, this helps improve the modeling accuracy of the environmental nonlinear dynamics model and the response speed of the adaptive force controller, thereby achieving more accurate and stable force control performance. Especially when facing complex, unknown, or dynamically changing environments, it can effectively improve the compliance and operational safety of the slave robotic arm.

[0079] In some embodiments described above, updated trajectory data considering environmental interaction can be obtained after processing by a priori neural network adaptive force controller. However, in practical applications, how to accurately and stably convert this updated trajectory data into the actual motion of the slave robotic arm and enable it to accurately track the desired trajectory is a key challenge in achieving efficient human-machine collaboration and force control. If the trajectory tracking is not accurate enough or the response is sluggish, it will directly affect the accuracy and compliance of the slave robotic arm's interaction with the environment, thereby reducing the performance of the entire shared control system and the operator's immersion.

[0080] To address this, this application further proposes a step of calculating control torque data based on updated trajectory data using a trajectory tracking controller. Specifically, this includes: calculating desired joint angle data and desired joint angular velocity data based on the updated trajectory data; acquiring actual joint angle data and actual joint angular velocity data of the slave-end robotic arm; calculating the angle error data between the desired joint angle data and the actual joint angle data, and the angular velocity error data between the desired joint angular velocity data and the actual joint angular velocity data; and calculating the control torque data using a proportional-derivative controller based on the angle error data and the angular velocity error data.

[0081] In this embodiment, the updated trajectory data is obtained after processing by a priori neural network adaptive force controller. It comprehensively considers the intention of the master operator, the smoothness of the variable impedance compliant control, and the real-time contact force between the slave robot and the environment, representing the ideal motion path of the slave robot in the current environment. This trajectory data is typically represented by position and orientation information in Cartesian space. To enable the robot to be controlled in joint space, the updated trajectory data (position and orientation) in Cartesian space needs to be converted into the desired angle data of each joint of the robot using an inverse kinematics algorithm. Simultaneously, by taking the time derivative of the desired joint angle data, the desired joint angular velocity data can be obtained. These desired joint space data provide a target reference for subsequent trajectory tracking. The slave robot is typically equipped with sensors such as encoders and rotary transformers to measure the current angular position of each joint in real time. Actual joint angular velocity data can be obtained by differentiating these real-time angle data or using specialized sensors (such as tachometer motors). These actual measurement data reflect the current true motion state of the robot. Angle error data refers to the difference between the desired joint angle data and the actual joint angle data, reflecting the deviation of the robot in position. Angular velocity error data refers to the difference between the desired joint angular velocity data and the actual joint angular velocity data, reflecting the deviation of the robotic arm in terms of movement speed. This error data forms the basis for feedback control by the trajectory tracking controller. The proportional-derivative (PDC) controller is a classic feedback controller that calculates the control output based on the current error (proportional term) and the rate of change of the error (derivative term). In this application, the PDC controller receives angular error data and angular velocity error data as input, and weights these errors using proportional gain and derivative gain to calculate the control torque data required to drive the slave-end robotic arm motor. The proportional term helps reduce static errors, enabling the robotic arm to quickly approach the desired position; the derivative term helps suppress oscillations, improving system stability and response speed, thereby achieving accurate tracking of the desired trajectory.

[0082] In this embodiment, the above-described technical solution effectively addresses the challenge of precise trajectory tracking by the slave-end robotic arm in complex interactive environments. Specifically, by converting the updated trajectory data into desired angles and angular velocities in joint space and comparing them in real time with the actual joint states of the slave-end robotic arm, the deviations in position and velocity of the robotic arm can be accurately quantified. Based on this, a proportional-derivative controller can be used to dynamically calculate the required control torque in real time based on these error data. The proportional term ensures that the robotic arm can quickly and accurately approach the desired trajectory, while the derivative term effectively suppresses overshoot and oscillations during the tracking process, improving the system's stability and response speed. This trajectory tracking mechanism based on error feedback enables the slave-end robotic arm to accurately follow the updated trajectory generated by the prior neural network adaptive force controller, thereby ensuring that in human-machine shared control mode, the slave-end robotic arm can accurately execute the operator's intentions and smoothly interact with the environment, improving the overall control performance and user experience of the system.

[0083] In some of the embodiments described above in this application, although the slave robotic arm can achieve adaptive force control based on variable impedance and prior neural networks to effectively cope with the nonlinear dynamics of the environment, the operator lacks direct perception of the contact force between the slave robotic arm and the environment from the master robotic arm. This lack of information may make it difficult for the operator to accurately judge the working state of the slave robotic arm and the characteristics of the environment, thereby affecting the intuitiveness and accuracy of the operation, especially in tasks that require fine force control or cope with complex environments.

[0084] In response, this application further proposes an adaptive force control hierarchical shared control method based on variable impedance and prior neural network. The method further includes: acquiring real-time contact force data of the slave robotic arm; scaling the real-time contact force data according to a preset ratio to obtain master feedback force data; and applying the master feedback force data to the master robotic arm so that the operator can perceive the contact force between the slave and the environment.

[0085] In this embodiment, acquiring real-time contact force data of the slave robotic arm refers to measuring the mechanical information generated when the slave robotic arm comes into contact with the external environment in real time using force sensors installed at the end effector or joint of the slave robotic arm. These force sensors are typically six-dimensional force sensors, capable of simultaneously measuring force and torque in three axes, thereby comprehensively and accurately reflecting the interaction between the slave robotic arm and the environment. The acquired raw force data undergoes necessary filtering and calibration to eliminate noise and ensure data accuracy.

[0086] In this embodiment, the real-time contact force data is scaled according to a preset ratio to obtain the master-end feedback force data. This step aims to convert the actual contact force sensed by the slave robotic arm into a force feedback signal suitable for application to the master robotic arm. Setting the preset scaling factor is crucial; it requires comprehensive consideration of the mechanical characteristics of the slave robotic arm, the operator's perception threshold, and the force feedback capability of the master robotic arm. For example, when the contact force between the slave robotic arm and the environment is large, a scaling factor less than 1 can be set to avoid excessive feedback force from the master robotic arm, thereby protecting the operator and the equipment; while when the contact force is small, a scaling factor greater than 1 can be set to enhance the operator's perception sensitivity. This scaling process is typically achieved through simple multiplication operations to ensure that the intensity of the feedback force matches the operator's perception needs.

[0087] In this embodiment, the master-end feedback force data is applied to the master-end robotic arm, allowing the operator to perceive the contact force between the slave end and the environment. This is typically achieved through force feedback actuators (such as high-precision motors or electromagnetic brakes) integrated within the master-end robotic arm. These actuators generate corresponding forces or torques based on the received master-end feedback force data, acting on the part of the master-end robotic arm that the operator holds or touches. For example, when the slave-end robotic arm encounters a hard obstacle in a remote environment, the master-end robotic arm applies a reaction force to the operator's hand, simulating this sense of obstruction, allowing the operator to intuitively "touch" the hardness, shape, or contact condition of the remote environment. This tactile feedback mechanism greatly enhances the operator's immersion and the precision of control over remote operations.

[0088] In this embodiment, the real-time contact force data of the slave robotic arm is scaled up according to a preset ratio and applied as master-end feedback force data to the master robotic arm. The operator can intuitively perceive the interaction force between the slave robotic arm and the environment through touch. This force feedback mechanism enhances the operator's sense of presence and understanding of the remote environment, enabling them to more accurately judge the working state of the slave robotic arm and environmental characteristics. Especially when performing delicate operations or dealing with complex and unknown environments, this solution effectively compensates for the shortcomings of relying solely on visual feedback, improving the intuitiveness, accuracy, and safety of operations, thereby achieving more efficient and stable shared control.

[0089] In some of the embodiments described above in this application, although the force control of the master and slave robotic arms is achieved through variable impedance compliant control and a priori neural network adaptive force controller, and adaptive adjustment can be made according to the external torque data of human-machine interaction and the environmental contact force data, in actual operation, the operator may not be able to intuitively understand the real-time working status of the slave robotic arm and the surrounding environment information. This may lead to the operator lacking a comprehensive perception of the movements of the slave robotic arm, thereby affecting the accuracy and safety of the operation.

[0090] In this regard, this application further proposes that the method also includes acquiring the working screen data of the slave robotic arm through a real-time camera, and displaying the working screen data on the operator's screen for the operator to monitor the status of the slave arm.

[0091] In this embodiment, the real-time camera is an imaging device capable of continuously capturing images or video streams. It can be configured as an industrial camera, USB camera, or endoscope camera, and is typically integrated near the end effector of the slave robotic arm or within its working area, connected to the control system via wired or wireless means. This real-time camera provides sufficient resolution, frame rate, and viewing angle to clearly capture the working image of the slave robotic arm. The working image data of the slave robotic arm is visual information collected by the real-time camera, reflecting the state of the slave robotic arm and its working environment. This data can be a continuous video stream or a series of high-frame-rate image frames, typically transmitted after encoding and transmission protocols. Its function is to provide the operator with visual feedback from the slave robotic arm, enabling them to observe the posture of the slave robotic arm, the movement of the end effector, its contact with the environment, and other objects within the working area. The operator screen is a human-machine interface for displaying visual information. It can be a high-resolution display, touchscreen, or VR / AR headset, etc. Its function is to receive and decode the working image data from the real-time camera and present it in a way that is easy for the operator to understand, ensuring that the operator can clearly and smoothly monitor the slave status. The monitoring of the slave status refers to the operator observing the work screen data displayed on the operator's screen to understand the real-time position, posture, interaction with the environment, and task execution progress of the slave robotic arm. This allows the operator to determine whether the slave robotic arm moves along the expected trajectory, whether it has an accidental collision with the environment, and whether it has successfully grasped the target object, helping the operator to promptly identify and correct potential problems.

[0092] In this embodiment, through the above technical solution, the operator can obtain real-time and intuitive data on the working screen of the slave robotic arm, which is then displayed on the operator's screen. This greatly enhances the operator's perception of the working status of the slave robotic arm and the environment, compensating for the lack of visual information that may exist when relying solely on force feedback or trajectory control. Based on real-time visual information, the operator can more accurately determine whether the movements of the slave robotic arm are as expected, and whether it has come into contact or collided with the environment, thereby adjusting the operating strategy in a timely manner and improving the accuracy, safety, and efficiency of remote operation. Especially in complex or delicate operational tasks, this real-time visual monitoring capability can improve the quality of human-machine collaboration and the success rate of task completion.

[0093] To further verify the actual control effect of the method proposed in this application, a master-slave teleoperation shared control experimental platform was built to conduct grinding operation verification experiments. The experimental system consists of two parts: a desktop six-axis robotic arm at the master end and a six-axis industrial robotic arm at the slave end. A six-dimensional force sensor is installed at the end of the slave robotic arm to detect the contact force with the environment in real time; an electric spindle is mounted at the end of the force sensor to perform the grinding task, and its start / stop and speed are adjusted by an independent frequency converter. The workpiece to be processed is firmly fixed on the multi-hole platform by a fixture to ensure the stability of the workpiece's posture during the experiment.

[0094] Based on the above experimental platform, shared control trajectory tracking and constant force contact experiments were conducted. The experimental results are as follows: Figure 3 The shared control 3D trajectory diagram shown illustrates the complete 3D motion trajectory mapped from the master end to the slave end. Experimental results demonstrate that the slave-end robotic arm follows the trajectory very smoothly without significant jitter, verifying the smoothing effect of the combined variable impedance compliant control and Kalman filtering. The trajectory includes the start point, end point, and constant force stable contact area, clearly reflecting the complete operation process of the slave-end robotic arm from free movement to environmental contact.

[0095] like Figure 4 The shared two-dimensional trajectory diagrams in the X, Y, and Z directions show the motion trajectory change curves in the X, Y, and Z directions, respectively. The Z-direction curve shows that when the robotic arm's end effector contacts the environment, adaptive force control ensures that the robotic arm maintains constant force contact with the environment, eliminating the need for manual adjustment of the contact depth by the operator and effectively reducing the operator's cognitive load. The X and Y-direction trajectories maintain good following accuracy, meeting the trajectory requirements for surface grinding operations.

[0096] like Figure 5 The force tracking experiment results shown in the figure illustrate the comparison curve between the expected force and the actual measured force. Based on these results, the force control error is within ±5% of the expected value, demonstrating extremely high accuracy. This verifies the stability and accuracy of the prior neural network adaptive force controller in unknown nonlinear environments, effectively ensuring the processing quality of constant force contact operations.

[0097] It should be noted that the above examples are only for understanding this application and do not constitute a limitation on the adaptive force control hierarchical shared control method based on variable impedance and prior neural network of this application. Any simple transformations based on this technical concept are within the protection scope of this application.

[0098] This application also provides an adaptive force-controlled hierarchical shared control system based on variable impedance and prior neural networks, referenced... Figure 2The adaptive force control hierarchical sharing control system based on variable impedance and prior neural network includes: a memory 10, a processor 20, and an adaptive force control hierarchical sharing control program based on variable impedance and prior neural network stored in the memory 10 and executable on the processor 20. The adaptive force control hierarchical sharing control program based on variable impedance and prior neural network is configured to implement the steps of the adaptive force control hierarchical sharing control method based on variable impedance and prior neural network.

[0099] The adaptive force control hierarchical sharing control system based on variable impedance and a priori neural network provided in this application, employing the adaptive force control hierarchical sharing control method based on variable impedance and a priori neural network in the above embodiments, can improve the stability and operational accuracy of contact force control during remote operation. Compared with the prior art, the beneficial effects of the adaptive force control hierarchical sharing control system based on variable impedance and a priori neural network provided in this application are the same as those of the adaptive force control hierarchical sharing control method based on variable impedance and a priori neural network provided in the above embodiments, and other technical features in the adaptive force control hierarchical sharing control system based on variable impedance and a priori neural network are the same as those disclosed in the methods of the above embodiments, and will not be repeated here.

[0100] It should be understood that the various parts disclosed in this application can be implemented using hardware, software, firmware, or a combination thereof. In the description of the above embodiments, specific features, structures, materials, or characteristics can be combined in any suitable manner in one or more embodiments or examples.

[0101] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. All equivalent structural transformations made under the technical concept of this application using the contents of the specification and drawings of this application, or direct / indirect applications in other related technical fields, are included within the scope of patent protection of this application.

Claims

1. An adaptive force-controlled hierarchical shared control method based on variable impedance and prior neural networks, characterized in that, The method includes: The real-time position data of the master robotic arm is acquired, and the real-time position data is processed by Kalman filtering to obtain filtered position data. Based on the filtered position data and the human-machine interaction external torque data obtained from the master robotic arm, smooth trajectory data is obtained through variable impedance compliant control; wherein, the variable impedance compliant control dynamically adjusts the damping parameters according to the human-machine interaction external torque data. The smooth trajectory data is scaled and mapped according to a preset ratio coefficient to obtain the desired trajectory data from the slave end. Real-time contact force data between the slave robotic arm and the environment is acquired. Based on the real-time contact force data and the slave-end desired trajectory data, the data is processed by a priori neural network adaptive force controller to obtain updated trajectory data. The priori neural network adaptive force controller is adaptively adjusted according to the nonlinear dynamics model of the environment. Based on the updated trajectory data, the control torque data is calculated by the trajectory tracking controller; The control torque data is sent to the motor of the slave robotic arm, driving the slave robotic arm to perform the corresponding action; Based on the filtered position data and the human-machine interaction external torque data obtained from the master robotic arm, the steps to obtain smooth trajectory data through variable impedance compliant control processing include: Obtain the reference trajectory position data, reference trajectory velocity data, and reference trajectory acceleration data corresponding to the filtered position data, and obtain the human-computer interaction external torque data; Based on the reference trajectory position data, reference trajectory velocity data, reference trajectory acceleration data, and the human-machine interaction external torque data, smooth trajectory data is calculated through a variable impedance rigid-flexible control structure; wherein, the variable impedance rigid-flexible control structure includes a variable damping function, and the variable damping function dynamically adjusts the damping value according to the rate of change of the human-machine interaction external torque data. The calculation of the variable damping function includes: Calculate the rate of change of the external torque data in the human-computer interaction; The first damping component is calculated using a nonlinear function based on the absolute value of the rate of change. The first damping component is added to the preset basic damping value to obtain the final damping value.

2. The adaptive force-controlled hierarchical shared control method based on variable impedance and prior neural network as described in claim 1, characterized in that, The step of scaling and mapping the smoothed trajectory data according to a preset scaling factor to obtain the desired trajectory data from the slave end includes: Get the slave position data from the previous moment; Calculate the difference between the filtered position data at the current time and the filtered position data at the previous time. Multiply the difference data by the preset scaling factor to obtain the location increment data; The incremental position data is added to the slave position data of the previous moment to obtain the slave desired trajectory data.

3. The adaptive force-controlled hierarchical shared control method based on variable impedance and prior neural network as described in claim 1, characterized in that, The steps of acquiring real-time contact force data between the slave robotic arm and the environment, and processing the updated trajectory data based on the real-time contact force data and the slave-end desired trajectory data through a priori neural network adaptive force controller include: Based on the expected trajectory data from the slave end, predict the environmental contact location data; Based on the real-time contact force data and the environmental contact location data, an environmental dynamics model is constructed, which includes linear spring terms and nonlinear terms. The nonlinear term is modeled by a pre-trained neural network initialized based on prior knowledge, and the neural network output data is obtained, wherein the prior knowledge includes pre-measured environmental parameter data. Based on the linear spring term, the neural network output data, and the real-time contact force data, the force error data is calculated. Based on the force error data, the neural network weight data is updated using an adaptive law; The updated trajectory data is calculated based on the updated neural network weight data and the environmental dynamics model.

4. The adaptive force-controlled hierarchical shared control method based on variable impedance and prior neural network as described in claim 3, characterized in that, The steps of updating the neural network weight data using an adaptive law based on the force error data include: Based on the force error data and the preset learning rate, the weight update data is calculated; The updated weight data is added to the current neural network weight data to obtain the updated neural network weight data.

5. The adaptive force-controlled hierarchical shared control method based on variable impedance and prior neural network as described in claim 1, characterized in that, Based on the updated trajectory data, the step of calculating the control torque data using the trajectory tracking controller includes: Based on the updated trajectory data, the expected joint angle data and the expected joint angular velocity data are calculated; Acquire the actual joint angle data and actual joint angular velocity data of the slave robotic arm; Calculate the angle error data between the expected joint angle data and the actual joint angle data, and the angular velocity error data between the expected joint angular velocity data and the actual joint angular velocity data; Based on the angle error data and the angular velocity error data, the control torque data is calculated by a proportional-derivative controller.

6. The adaptive force-controlled hierarchical shared control method based on variable impedance and prior neural network as described in claim 1, characterized in that, The method further includes: Acquire real-time contact force data from the slave robotic arm; The real-time contact force data is scaled according to a preset ratio to obtain the main end feedback force data; The master-end feedback force data is applied to the master-end robotic arm, allowing the operator to perceive the contact force between the slave end and the environment.

7. The adaptive force-controlled hierarchical shared control method based on variable impedance and prior neural network as described in claim 1, characterized in that, The method further includes: The working image data of the robotic arm is collected in real time by a camera; The work screen data is displayed on the operator's screen so that the operator can monitor the status of the slave device.

8. An adaptive force-controlled hierarchical shared control system based on variable impedance and prior neural network, characterized in that, The adaptive force control hierarchical sharing control system based on variable impedance and prior neural network includes: a memory, a processor, and an adaptive force control hierarchical sharing control program based on variable impedance and prior neural network stored in the memory and executable on the processor. The adaptive force control hierarchical sharing control program based on variable impedance and prior neural network is configured to implement the steps of the adaptive force control hierarchical sharing control method based on variable impedance and prior neural network as described in any one of claims 1 to 7.

Images