Dynamic encoding event-triggered adaptive control method and system for flexible joint robot arm
By using a dynamic coding event triggering mechanism and adaptive control with a preset performance function for switching monotonic tubes, the problems of control stability and communication efficiency of flexible joint robotic arms under complex working conditions are solved, and smooth control and efficient communication of flexible joint robotic arms under the switching of multiple discontinuous reference trajectories are realized.
Patent Information
- Application Number
- CN202611376180.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-09-07
- Publication Date
- 2026-11-13
AI Technical Summary
Existing flexible joint robotic arm control systems, when faced with complex working conditions, cannot cover instantaneous error jumps due to the fixed boundaries of preset performance control, and the fixed thresholds for event triggering cannot balance communication efficiency and transient performance. The lack of collaborative design leads to problems such as oscillation and excessive communication burden in the system under sudden working conditions.
The design incorporates a dynamic encoding event triggering mechanism and a preset performance function for switching monotonic tubes. By using dynamic threshold parameters and online approximation of unknown nonlinear dynamics, an adaptive controller is constructed. The encoded control signal is transmitted as a discrete string, and the performance boundary is adaptively adjusted at trajectory switching moments. Combined with a state feedback controller, efficient closed-loop control of the flexible joint robotic arm is achieved.
It significantly reduces the number of triggers and communication burden, ensuring smooth control of the system under the switching of multiple discontinuous reference trajectories. The trigger frequency is reduced by 70%, communication efficiency is improved, steady-state accuracy is maintained, and the trigger ratio in the simulation is only 29.15%, which expands the applicability of the method in practical systems.
Smart Images

Figure CN123080698A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of robotic arm control, and in particular to a dynamic coding event-triggered adaptive control method and system for a flexible joint robotic arm. Background Technology
[0002] With the rapid development of modern industrial automation, aerospace exploration, and medical rehabilitation assistance, robotic manipulators are increasingly being used in complex working environments. While traditional rigid manipulators have a relatively mature control theory system, their inherent drawbacks, such as large mass, poor human-machine interaction safety, and limited environmental adaptability, restrict their further application in unstructured environments and human-machine collaborative scenarios. To address these limitations, flexible joint manipulators (FJMs), which incorporate elastic elements at the joints, not only effectively reduce contact forces during collisions and significantly improve operational safety, but also offer significant advantages such as lighter weight, larger workspace, and lower energy consumption, making them a cutting-edge research direction in robotics.
[0003] However, while the introduction of joint flexibility brings the aforementioned advantages, it also introduces complex dynamic characteristics into the system: on the one hand, the system exhibits highly nonlinear and underactuated coupling, resulting in more frequent vibrations and uncertainties such as unmodeled dynamics and external disturbances; on the other hand, the presence of elastic elements creates strong coupling between the dynamics of the link end and the dynamics of the rotor end, posing fundamental challenges to controller design. How to effectively suppress flexible vibrations, compensate for system uncertainties, and simultaneously ensure the transient and steady-state performance of the system under high-performance trajectory tracking requirements has become a core challenge in the research of flexible joint manipulator control.
[0004] Preset performance control methods and their limitations: To ensure the transient and steady-state performance of nonlinear systems under complex uncertainties and external disturbances, the Prescribed Performance Control (PPC) method has received widespread attention since its inception. The core idea of this method is to force the system's tracking error to converge strictly within a pre-designed error envelope, thereby transforming performance indicators such as overshoot, convergence speed, and steady-state accuracy into explicit constraints. In the fields of mechanical systems and manipulator control, PPC has been widely used to limit actuator overshoot and ensure steady-state accuracy. For example, researchers use PPC to combine system performance objectives with physical constraints, ensuring that the manipulator system achieves preset dynamic and steady-state performance while satisfying multiple constraints; other scholars use fuzzy logic systems to identify unknown nonlinear dynamics and design PPCs to guarantee the transient performance of the control system.
[0005] Regarding the construction of performance functions, existing research mainly falls into two categories: one is the pre-defined performance function based on the error funnel, whose boundary depends on the initial error and must satisfy specific initial conditions; the other is the monotonic tube pre-defined performance function (Monotonic Tube PPF). Compared to the funnel-type PPF, the monotonic tube PPF can directly map intuitive control indicators such as overshoot, convergence time, and steady-state accuracy as design parameters into the controller, thereby constraining the transient and steady-state performance of the system more strictly and conveniently, effectively solving the problem of indirect pre-setting of performance indicators and reliance on trial-and-error tuning.
[0006] However, existing performance-constrained control research generally suffers from a key limitation: most schemes rely on performance boundaries with fixed attenuation forms or monotonically decaying performance boundaries. In real-world complex conditions such as task switching and sudden obstacle avoidance, the reference tracking trajectory often becomes discontinuous or experiences instantaneous jumps. In such cases, the initial error can easily exceed the constraint range of the fixed performance function, leading to error transformation singularities or system instability. Although some studies have attempted to address transient errors by extending the funnel boundary, the extension methods used are usually open-loop fixed quantities, lacking a mechanism for adaptive adjustment based on the actual error amplitude. This results in insufficient or excessive conservatism when dealing with sudden jumps. Furthermore, due to the inherent strong nonlinearity and elastic coupling characteristics of FJMs, traditional unconstrained or fixed-constraint controllers are prone to generating severe transient oscillations when facing abrupt signals, making it difficult to achieve high-quality control objectives.
[0007] Event-triggered communication mechanisms and their bottlenecks: On the other hand, when transmitting signals within a control system, to reduce communication bandwidth consumption and improve resource utilization, event-triggered mechanisms (ETMs) have been widely used to replace traditional periodic time-triggered control. Existing event-triggered strategies are mainly divided into three categories: static event-triggered, memory-based event-triggered, and dynamic event-triggered. For example, for uncertain switching nonlinear systems, researchers have designed an adaptive dynamic event-triggered control strategy that combines asynchronous switching with state-dependent switching laws, effectively reducing network communication burden and avoiding frequent system switching while ensuring tracking accuracy. Furthermore, for nonlinear multi-agent systems affected by unknown disturbances, some scholars have proposed dynamic event-triggered mechanisms that rely solely on output information. By fusing state / disturbance observers with dynamic filtering techniques, this effectively reduces the communication load between adjacent agents.
[0008] In recent years, Coded Event-Triggered Control (CEC) has emerged to further reduce the data load of information transmission. CEC maps trigger signals to short, discrete digital codes—such as symbols and amplitude codes that only transmit threshold values—significantly reducing the time and network resources required for information transmission and improving transmission security. However, existing CEC research still has prominent problems: when the input falls within a certain range, it corresponds to a fixed coding mapping and a fixed event trigger threshold. When the FJM faces trajectory jumps or changes in operating conditions, a fixed threshold either leads to excessively frequent triggering during transient phases, thus losing the advantage of event triggering, or causes a significant decrease in transient tracking performance due to an excessively large threshold. In addition, most existing ETC research has failed to properly resolve the inherent contradiction between communication efficiency and transient performance under sudden operating conditions—that is, the conflict between the expectation of saving resources by reducing communication frequency and the expectation of suppressing transient errors by densely updating control signals. Summary of the Invention
[0009] The purpose of this application is to address the problems that existing preset performance control fixed boundaries cannot encompass instantaneous error jumps and existing event trigger fixed thresholds cannot balance communication efficiency and transient performance when a flexible joint robotic arm is executing multiple discontinuous reference trajectories, and that the two lack coordinated design. This application provides a dynamic coding event triggering adaptive control method and system for flexible joint robotic arms.
[0010] The above-mentioned objective of this application is achieved through the following technical solution: Step S1: Establish a dynamic model of the flexible joint manipulator described by the coupling of link dynamics and rotor dynamics; Step S2: Design a dynamic encoding event triggering mechanism to reduce the communication burden between the controller and the actuator; encode the control signal to be transmitted into a discrete string of finite bit length, the discrete string containing both the sign and amplitude of the trigger error; introduce a dynamic threshold parameter into the triggering condition; Step S3: Design a preset performance function for switching monotonic tubes to constrain link position tracking error; for the case where the robotic arm executes multiple discontinuous reference trajectories, construct a monotonically decaying performance boundary for each continuous sub-task, and adaptively expand the performance boundary at the switching moment when the reference trajectory changes. Step S4: Based on the preset performance function of the switching monotonic tube and the dynamic coding event triggering mechanism, the state feedback controller is designed using the backstepping method. The radial basis function neural network is used to approximate the lumped unknown nonlinear dynamics of the system online to obtain the actual control torque of each joint. Step S5: Based on the actual control torque of each joint, the control signal is encoded and transmitted through a dynamic coding event triggering mechanism. When the event triggering condition is met, the actual control signal of the actuator is updated, and the dynamic threshold parameter is updated at the same time. In each control cycle, it is determined whether the triggering condition is met. If it is met, steps S4 to S5 are repeated; if it is not met, the current control input is kept unchanged to realize the dynamic closed-loop control of the flexible joint robot arm.
[0011] Optionally, step S1 includes: Dynamic model of flexible joint robotic arm:
[0012] in, These represent the angular position, angular velocity, and angular acceleration vectors of the link, respectively. This represents the angular position, angular velocity, and angular acceleration vectors on the rotor side. The inertia matrix; The vectors of the Coriolis force and the centrifugal force; It is the gravity vector; This is the friction term; These are the joint stiffness matrix, actuator inertia matrix, and damping matrix, respectively. The input torque vector for each actuator; denoted as the number of degrees of freedom of the robotic arm.
[0013] Optionally, step S2 includes: The encoding and triggering logic of the dynamic encoding event triggering mechanism includes: Continuous control signal With the actual output of the actuator The difference between them is defined as the trigger error. ; Event trigger time sequence Determined by the following triggering conditions:
[0014] in, Indicates the sequence number of the trigger time; Represents the continuous control signal to be designed; even integer. The base of the encoding number system is an even number. Indicates the first The encoding base of the power; For the first Weighting coefficients for joke-related tasks; This is a dynamic threshold parameter; Gain is adjusted using a dynamic threshold. The variables in equation (2) are governed by the following update law:
[0015]
[0016]
[0017]
[0018] in, This represents the amplitude level index, used to determine the quantization range to which the current control input amplitude belongs; This represents a range index, with values ranging from... arrive An integer used to identify the interval number where the control input amplitude is located; Indicates the total number of intervals; Indicates the first The lower boundary of each amplitude quantization interval and Together constitute the first A range of intervals; Indicates the bit length of the transmitted string; Indicates the first Each trigger time base The first in the encoded string The value of the bit takes the value of arrive Integers between; Indicates the encoding radix of The power is used to calculate the weighted value of each bit in the encoded string; The base of the encoded number system is used, and even numbers are preferred. Representation of range index The corresponding trigger threshold weight coefficient is a piecewise constant positive scalar; This represents the dynamic threshold attenuation coefficient, which is a design parameter. Represents dynamic threshold parameter The first derivative with respect to time is used to describe the continuous evolution of the dynamic threshold; When the trigger condition (2) is satisfied, the sign and magnitude of the trigger error are encoded into a string of length [length missing]. discrete strings Transmission is performed, with the most significant bit... The symbol representing the trigger error, the rest. The bit indicates the magnitude level of the trigger error.
[0019] Optionally, step S2 may further include: The decoding rule of the actuator is as follows:
[0020] in, This is the actual control input after decoding. The value is the control signal value from the previous trigger moment; the actuator then determines the value based on the received discrete string. and local storage , Parameters, combined with dynamic threshold parameters Reconstruct the control signal.
[0021] Optionally, step S3 includes: The method for constructing the preset performance function of the switching monotonic tube includes: Definition of the first Performance function of joke task for:
[0022] in, For the first The initial boundary magnitude at the start of the joke task. To the maximum allowable steady-state tracking error, The rate of convergence is exponential. and The first The start and end times of the joke-writing task; At the switching time The performance boundaries are reset according to the following rules:
[0023] in, The left limit value of the boundary before the switching time. For the extent of boundary expansion, This is the proportionality coefficient. To adjust the adjustable coefficients for boundary asymmetry, The instantaneous value of the tracking error at the switching moment; Linkage position tracking error Constrained as:
[0024] in, Let be the upper boundary function. It is the lower boundary function; and The basic performance funnel curve, and This is an auxiliary transient function used to achieve finite-time convergence after the switching time. The basic performance funnel curve and The expression is:
[0025] in, For symbolic functions, This is an adjustable coefficient used to adjust the degree of asymmetry of the performance boundary relative to the initial sign of the tracking error; The auxiliary transient function and The expression is:
[0026]
[0027] in, For scaling parameters, The initial value for tracking error, and These are the initial values for the upper and lower boundaries, respectively. A finite convergence time is set for the user; when hour, and The error tends to zero uniformly, ensuring that the error is minimized within a finite time. Return to the basic performance funnel.
[0028] Optionally, step S4 includes: Define state variables and Equation (1) can be transformed into the following form:
[0029] in, This represents the angular velocity vector of the connecting rod; This represents the angular acceleration vector of the link; Representing the inertia matrix The inverse matrix; Represents the joint stiffness matrix; The friction term relates to the angular velocity of the connecting rod. The function; The gravity vector is given by the angular position of the link. The function; Represents the matrix of Coriolis force and centrifugal force; Represents the rotor angular velocity vector; This represents the rotor angular acceleration vector; Represents the actuator inertia matrix; This represents the actuator input torque vector; Represents the actuator damping matrix; Represents the system output vector; Based on the state equation of the flexible robotic arm system described in equation (9), a state feedback controller is designed as follows: Define tracking error
[0030]
[0031] in, and For system status, To track the desired trajectory, This indicates the virtual control law that is about to be defined; The derivative of equation (10) is
[0032] Based on the state function (9) and the tracking error (11), equation (12) can be transformed into
[0033] Combine the preset performance function of the preset switching monotonic tube with the control tracking error:
[0034]
[0035] in, This represents the upper normalized error vector, containing the tracking errors of all joints relative to the upper boundary. The normalized ratio; Indicates the first The tracking error of each joint, i.e., the difference between the link angular position and the desired trajectory; Indicates the first The upper performance boundary of each joint is determined by the preset performance function of the switching monotonic tube; This represents the lower normalized error vector, containing the tracking errors of all joints relative to the lower boundary. The normalized ratio; Indicates the first Tracking error of each joint; This represents the joint index, with a value range of [value range missing]. to ; This indicates the number of degrees of freedom of the robotic arm, i.e., the number of joints. Indicates the first The performance boundary center offset of each joint is used to convert the asymmetric upper and lower boundaries into a symmetric form. Indicates the first The performance boundary half-width of each joint is used to normalize and scale the tracking error. For robotic arms, the following asymmetric time-varying barrier Lyapunov function is proposed:
[0036] but, The time derivative is
[0037] Substituting equation (13) into equation (14) yields
[0038] Design virtual control law To ensure the stability of the first-order subsystem:
[0039] in, Indicates the first The first virtual control law for each joint is used to stabilize the first-order error subsystem and ensure that the tracking error converges within the preset performance boundary. Indicates the first Desired trajectory of each joint The first derivative with respect to time, i.e., the desired angular velocity; Indicates the first Joint performance boundary center offset The first derivative with respect to time; Indicates the first Half-width of joint performance boundary The first derivative with respect to time; Indicates the first The positive definite matrices of each joint in the first virtual control law are used to adjust the error convergence rate. Substituting equation (16) into equation (15) yields
[0040] From the above, we can see that The time derivative is
[0041] Introducing another virtual control law Construct error variables as follows:
[0042] Consider Lyapunov functions :
[0043] Differentiating equation (18) and utilizing the oblique symmetry property of the robotic arm, i.e. Coriolis item Can be rewritten as ; item Completely offset, resulting in
[0044]
[0045]
[0046] According to equation (19), the virtual control law is set. To ensure the stability of the second-order subsystem; Define the lumped unknown continuous function as
[0047] in, It represents a lumped unknown continuous function, including the Coriolis force and centrifugal force terms, gravity terms, friction terms, and the coupling terms of the inertial matrix and the virtual control law derivative, and is used to uniformly describe all modelable but parameter-unknown nonlinear dynamics in a robotic arm system; The definition is that the left-hand symbol is defined by the right-hand expression, and it is an assignment operator; The inertia matrix is given by the angular position of the link. A symmetric positive definite matrix, describing the inertial coupling characteristics of each joint of the robotic arm; Represents the first virtual control law The first derivative with respect to time is used to compensate for dynamic changes in the virtual control law in subsequent controller design; Using radial basis function neural networks in compact sets Approximating with arbitrary desired accuracy :
[0048] in, This represents a Gaussian function vector, where each component is a radial basis function used to transform the input vector. Mapping to a high-dimensional feature space to approximate unknown nonlinear functions; It represents the bounded approximation error, reflecting the residual between the ideal approximation of the neural network and the true function; The estimated value of the weight matrix is adjusted online by an adaptive update law; The ideal value of the weight matrix is the optimal weight matrix that minimizes the approximation error. The weight estimation error matrix is defined as follows: That is, the difference between the estimated value and the ideal value, which is used to characterize the convergence of neural network adaptive learning in stability analysis; The transpose operation represents a matrix or vector; The input vector of the neural network is represented by the link angle position. Angular velocity of the connecting rod Expected trajectory Expected speed and expected acceleration Together they form; The approximation function is
[0049] Virtual control law Designed for
[0050] in ; Substituting (20) into (19), we get The time derivative is:
[0051]
[0052] From the above, we can see that The time derivative is
[0053] Introducing another virtual control law Construct error variables for
[0054] Thus obtain
[0055] Consider Lyapunov functions for
[0056] Differentiating it and using (23), we get
[0057] To eliminate cross terms and ensure the stability of the third-order subsystem, Designed for
[0058] in, This represents the positive definite gain matrix of the third virtual control law. For scalar design parameters, for 3D identity matrix; Substituting (21) and (25) into (24), we get
[0059]
[0060] in This represents the Lyapunov function designed in the third step, used to analyze the stability of the first three error subsystems; The time derivative is:
[0061] in The fourth error variable is defined as the difference between the rotor angular velocity and the third virtual control law. ; Represents the fourth error variable The first derivative with respect to time; Represents the rotor angular velocity vector The first derivative with respect to time, i.e., the rotor angular acceleration; Represents the third virtual control law The first derivative with respect to time; Represents the actuator inertia matrix The inverse matrix; This represents the actual control input torque vector, which is the final control signal applied to each joint actuator; This represents the actuator damping matrix, which is a positive definite diagonal matrix. This represents the joint stiffness matrix, which is a positive definite diagonal matrix. Represents the rotor angular position vector; Represents the angular position vector of the link; Consider Lyapunov functions :
[0062] Then, The time derivative is
[0063] Ideal continuous control law Defined as
[0064] in This represents the positive definite gain matrix of the fourth virtual control law. , Scalar design parameters; By introducing a dynamic coding event triggering mechanism, the actual control signal applied to the system is ; Define the trigger error as , can be obtained
[0065] Substituting (27) and (28) into (26), we get The time derivative is (29) The weight update law of the radial basis function neural network is designed as follows:
[0066] in Represents the estimated value of the neural network weight matrix The first derivative with respect to time is the weight adaptive update law; This represents the positive definite adaptive gain matrix, used to adjust the learning rate of the neural network weights; express - Correct parameters, Scalar design parameters are used to enhance the robustness of weight updates; Consider the following global composite Lyapunov function:
[0067] Differentiating (31) yields
[0068] Combining (30) and (32), and noting , can be obtained
[0069] Substituting (29) into (33), we get
[0070] in The trace operation represents the sum of the elements on the main diagonal of a matrix. Represents the positive definite adaptive gain matrix The inverse matrix; Represents the overall composite Lyapunov function The first derivative with respect to time; Representation matrix traces; This indicates that the scalar expression obtained by expanding the cross terms is equal to the result of its trace operation; This represents the trace of the product of the weight estimation error matrix and the weight estimation matrix; Represents the third error variable The square of the Euclidean norm; Represents the fourth error variable The square of the Euclidean norm; Indicates the second error variable The square of the Euclidean norm; ; Using Young's inequality, the cross term can be defined as
[0071] Substituting inequality (35) into (34), we obtain the final continuous dynamics:
[0072] in This represents the neural network approximation error vector; Indicates triggering error; Represents the weight estimation error matrix The square of the Frobenius norm is used to characterize the overall estimation bias of all elements of the weight matrix; The square of the Euclidean norm of the neural network approximation error vector is represented. The square of the Euclidean norm of the trigger error vector is represented. Represents the ideal weight matrix The square of the Frobenius norm.
[0073] Optionally, the method further includes an output feedback control mode: when the speed state of the robotic arm link end cannot be directly measured, the unmeasurable state is estimated by designing a state observer, and the estimated value is used to replace the actual state value. Output feedback control is realized based on an adaptive state feedback controller framework; the state observer adopts a high-gain observer or a dimension-reduced observer structure.
[0074] A dynamic coded event-triggered adaptive control system for a flexible joint robotic arm, the system comprising: The signal acquisition module is used to acquire the angular position signals of each joint link and the rotor in real time; The storage module is used to store the desired tracking trajectory, the parameter configuration of the preset performance function of the switching monotonic tube, the encoding parameters of the dynamic encoding event triggering mechanism, and the initial weights of the radial basis function neural network. The module for calculating the preset performance function of the switching monotonic tube is used to calculate the time-varying performance boundary of each joint based on the current running stage and real-time tracking error, and to adaptively reset the boundary amplitude based on the error jump amplitude at the trajectory switching moment. The dynamic encoding event trigger judgment module is used to calculate the trigger error between the ideal continuous control signal and the actual control signal in real time, compare the absolute value of the trigger error with the dynamic trigger threshold, and generate a trigger signal and determine the encoding string at the current moment when the trigger error reaches or exceeds the dynamic trigger threshold. The encoding and decoding module is used to encode the symbol information and amplitude level of the trigger error into a discrete string of finite bit length for transmission at the trigger time, and decode it into the actual control torque command at the actuator end according to the received encoded string and the local storage parameters. The adaptive state feedback controller module includes a cascaded error control law based on backstepping and a radial basis function neural network adaptive approximator, which is used to generate an ideal continuous control torque signal based on the connecting rod angular position signal, the rotor angular position signal and the desired trajectory. The actual control signal output module is used to keep the control torque unchanged at the previous triggering time when it is not triggered, update it to the currently decoded control torque at the triggering time, and output the control torque to each joint actuator. The output of the switching monotonic tube preset performance function calculation module is connected to the input of the adaptive state feedback controller module, providing the controller with real-time error constraint boundaries; the output of the adaptive state feedback controller module is connected to the input of the dynamic encoding event trigger judgment module and the encoding and decoding module; the output of the encoding and decoding module is connected to the input of the actual control signal output module; the output of the actual control signal output module is connected to each joint actuator, forming a closed-loop control.
[0075] The beneficial effects of the technical solution provided in this application are: 1. To address the limitation of existing preset performance control methods where fixed boundaries cannot handle instantaneous changes in the reference trajectory, this application proposes a switching monotonic tube preset performance function. This function directly maps intuitive indicators such as overshoot, convergence time, and steady-state accuracy to design parameters, and adaptively resets the performance boundaries based on the actual error amplitude at the trajectory switching time. This ensures that the transient error is always enveloped. Simultaneously, an auxiliary transient function is introduced to guarantee that the error is contained within a finite time. The process returns to the basic funnel. Beneficial effects: It fundamentally avoids the singularity problem of error transformation and achieves smooth switching control under multi-segment discontinuous tasks.
[0076] Dynamic encoding event triggering mechanism: To address the issue of a sudden increase in communication burden during transient phases when the fixed trigger threshold is used, this application dynamically evolves the parameters. Introducing trigger conditions, its update law is: This allows the threshold to adaptively adjust according to transient errors. Simultaneously, the control signal is encoded as... This method transmits discrete strings while simultaneously encoding both symbol and amplitude information. Beneficial effects include a reduction in trigger counts of approximately 70%, significantly reducing communication overhead while maintaining steady-state accuracy; the trigger ratio in simulations was only 29.15%.
[0077] State / output feedback dual-architecture collaborative design: This application further designs an observer-based output feedback scheme based on state feedback to solve the engineering problem of directly measuring the velocity of flexible joints. Beneficial effects: It expands the applicability of the method to practical physical systems and forms a complete theoretical system. Attached Figure Description
[0078] The present application will be further described below with reference to the accompanying drawings and embodiments. In the accompanying drawings: Figure 1 This is a flowchart of the dynamic coding event triggering scheme in the embodiments of this application; Figure 2 This is a schematic diagram of the preset performance function of the switching monotonic tube in the embodiments of this application; Figure 3 This is a simulation result diagram of the joint 1 state feedback control in the embodiment of this application; Figure 4 This is a simulation result diagram of the joint 2 state feedback control in the embodiment of this application. Detailed Implementation
[0079] To provide a clearer understanding of the technical features, objectives, and effects of this application, the specific embodiments of this application will now be described in detail with reference to the accompanying drawings.
[0080] The embodiments of this application provide a dynamic coding event-triggered adaptive control method for a flexible joint robotic arm.
[0081] Please refer to Figure 1 , Figure 1 This is a flowchart of the dynamic coding event triggering scheme of a dynamic coding event triggering adaptive control method for a flexible joint robotic arm according to an embodiment of this application, including: Step S1: Establish a dynamic model of the flexible joint manipulator described by the coupling of link dynamics and rotor dynamics; Step S2: Design a dynamic encoding event triggering mechanism to reduce the communication burden between the controller and the actuator; encode the control signal to be transmitted into a discrete string of finite bit length, the discrete string containing both the sign and amplitude of the trigger error; introduce a dynamic threshold parameter into the triggering condition; Step S3: Design a preset performance function for switching monotonic tubes to constrain link position tracking error; for the case where the robotic arm executes multiple discontinuous reference trajectories, construct a monotonically decaying performance boundary for each continuous sub-task, and adaptively expand the performance boundary at the switching moment when the reference trajectory changes. As one embodiment, the performance boundary is adaptively expanded at the switching moment to enclose the instantaneous error jump, thus avoiding the singularity problem of error transformation.
[0082] Step S4: Based on the preset performance function of the switching monotonic tube and the dynamic coding event triggering mechanism, the state feedback controller is designed using the backstepping method. The radial basis function neural network is used to approximate the lumped unknown nonlinear dynamics of the system online to obtain the actual control torque of each joint. Step S5: Based on the actual control torque of each joint, the control signal is encoded and transmitted through a dynamic coding event triggering mechanism. When the event triggering condition is met, the actual control signal of the actuator is updated, and the dynamic threshold parameter is updated at the same time. In each control cycle, it is determined whether the triggering condition is met. If it is met, steps S4 to S5 are repeated; if it is not met, the current control input is kept unchanged to realize the dynamic closed-loop control of the flexible joint robot arm.
[0083] Step S1 includes: Dynamic model of flexible joint robotic arm:
[0084] in, These represent the angular position, angular velocity, and angular acceleration vectors of the link, respectively. This represents the angular position, angular velocity, and angular acceleration vectors on the rotor side. The inertia matrix; The vectors of the Coriolis force and the centrifugal force; It is the gravity vector; This is the friction term; These are the joint stiffness matrix, actuator inertia matrix, and damping matrix, respectively. The input torque vector for each actuator; denoted as the number of degrees of freedom of the robotic arm.
[0085] As one embodiment, the dynamic model of the flexible joint robotic arm should satisfy the following assumptions: Assumption 1. Assume that the dynamics of the link end of the flexible joint manipulator (as shown in equation (1)) have the basic physical characteristics of a rigid manipulator. Specifically, the system parameters satisfy the following properties: Assumption 1-1): Inertia Matrix It is symmetric positive definite. It has positive constants. and , making .
[0086] Assumptions 1-2): Matrix Oblique symmetry, satisfying: for any vector , .
[0087] Assumptions 1-3): Gravity vector Friction terms and external unmodeled friction terms Both are bounded, that is, for some positive constants and , and .
[0088] Assumptions 1-4): Assume the desired tracking trajectory It is piecewise smooth. Specifically, within each interval between adjacent switching moments, and its end The time derivatives of the order are all continuous and bounded.
[0089] Note 1. Although hypotheses 1-1) to 1-3) are fairly standard in existing research, the boundedness in hypotheses 1-3) can be reasonably explained by the finite energy of the system and is a mathematically rigorous requirement for presupposed performance and event-triggered designs. Furthermore, the piecewise smooth trajectory in hypotheses 1-4) is specifically designed for switching frameworks, allowing for instantaneous transitions in reference signals between tasks while ensuring smoothness within tasks. Compared to standard global smooth tracking, this piecewise characteristic makes the design of relevant controllers and the proof of stability more complex.
[0090] Question 1. For a flexible joint robotic arm, in a continuous interval Execute the defined segmented tracing task on ( Design a state / output feedback controller based on a Dynamic Coded Event Triggering Mechanism (DCEM) to enable... 1) Although the task transition occurs in any initial state The change causes a momentary error jump, but it is still strictly and continuously limited within the switching monotonic transistor preset performance function (SMT-PPF), i.e. ,in and Indicates the first The specified monotonic lower and upper bounds of the interval performance envelope.
[0091] 2) Ensure all closed-loop signals are semi-globally consistent and eventually bounded (SGUUB). Furthermore, ensure the event interval has a strictly positive lower bound, i.e. It can strictly exclude Zeno behavior in the communication channel, where Indicates the first The first interval One trigger moment, and It is a positive number.
[0092] Note 2. Unlike general problems where initial constraint violations only occur at system startup, the switching framework assumes (1-4) that due to instantaneous trajectory jumps, at each switching moment... This will introduce repeated boundary escapes, which always occur in flexible jointed robotic arms and trigger severe transient errors. Therefore, solving problem 1 requires ensuring that the error after each task transition is within the convergence time... Rapid convergence within the region is crucial.
[0093] Step S2 includes: The encoding and triggering logic of the dynamic encoding event triggering mechanism includes: Continuous control signal With the actual output of the actuator The difference between them is defined as the trigger error. ; Event trigger time sequence Determined by the following triggering conditions:
[0094] in, Indicates the sequence number of the trigger time; Represents the continuous control signal to be designed; even integer. The base of the encoding number system is an even number. Indicates the first The encoding base of the power; For the first Weighting coefficients for joke-related tasks; This is a dynamic threshold parameter; Gain is adjusted using a dynamic threshold. The variables in equation (2) are governed by the following update law:
[0095]
[0096]
[0097]
[0098] in, This represents the amplitude level index, used to determine the quantization range to which the current control input amplitude belongs; This represents a range index, with values ranging from... arrive An integer used to identify the interval number where the control input amplitude is located; Indicates the total number of intervals; Indicates the first The lower boundary of each amplitude quantization interval and Together constitute the first A range of intervals; Indicates the bit length of the transmitted string; Indicates the first Each trigger time base The first in the encoded string The value of the bit takes the value of arrive Integers between; Indicates the encoding radix of The power is used to calculate the weighted value of each bit in the encoded string; The base of the encoded number system is used, and even numbers are preferred. Representation of range index The corresponding trigger threshold weight coefficient is a piecewise constant positive scalar; This represents the dynamic threshold attenuation coefficient, which is a design parameter. Represents dynamic threshold parameter The first derivative with respect to time is used to describe the continuous evolution of the dynamic threshold; As one embodiment, the dynamic threshold parameter is adaptively adjusted according to the amplitude of the trigger error to buffer the abnormal increase in trigger frequency caused by trajectory jumps. A set of boundaries is defined to delineate the control input amplitude. Weighting coefficients A user-defined positive scalar. (Symbol) Corresponding to the The base of the update time The first in the encoding Position. Furthermore, It is a design parameter used to ensure that initial conditions are met. .
[0099] When the trigger condition (2) is satisfied, the sign and magnitude of the trigger error are encoded into a string of length [length missing]. discrete strings Transmission is performed, with the most significant bit... The symbol representing the trigger error, the rest. The bit indicates the magnitude level of the trigger error.
[0100] As one embodiment, the triggering rule in equation (2) is determined. The symbols are a prerequisite for accurate encoding and decoding. The logical relationship in equations (3) and (4) of the encoding mechanism shows that if... Then set Otherwise set Therefore, in any interval Inside, the control station uses encoded strings. Communicate with the actuator, and whenever the execution condition (2) is met, send a message of length [length missing]. string Functionally speaking, It also includes the sign and magnitude of the trigger error tolerance. Specifically, the most significant bit... Acting as a symbolic indicator bit (i.e. mean ), the remaining positions Then the amplitude of the precise encoding threshold.
[0101] Step S2 also includes: The decoding rule of the actuator is as follows:
[0102] in, This is the actual control input after decoding. The value is the control signal value from the previous trigger moment; the actuator then determines the value based on the received discrete string. and local storage , Parameters, combined with dynamic threshold parameters Reconstruct control signals.
[0103] As one embodiment, when event criterion (2) is in When the conditions are met, the relevant information is immediately broadcast to the actuator. Using previously stored parameters and the latest control input, the actuator decodes the data via the following protocol. Decode to update Initial value set to . and The selection strictly follows the definitions in equations (3)–(6) and utilizes the actuator's local memory. Combining the triggering condition in equation (2) with the decoding principle (7), it can be deduced that... The overall architecture of DCEM is as follows: Figure 1 As shown.
[0104] To visually demonstrate DCEM, consider and This configuration produces a standard binary system, which transmits a compact 3-bit string on each trigger, where each bit is... Under these settings, whenever equation (2) is triggered, the detailed encoding and decoding process is summarized in Table I.
[0105] Table I: Encoder–Decoder Map
[0106] Note 3. The proposed DCEM aims to simultaneously reduce triggering frequency and data load, thereby improving communication efficiency. Although traditional coding schemes can effectively compress transmitted data into a compact format... Bit string This is intended to save bandwidth and provide basic network security, but its fixed threshold can lead to frequent triggering when handling transient errors caused by handover. By dynamically evolving parameters... By introducing triggering conditions, the proposed collaborative design achieves resilient tolerance, which helps buffer these sudden transients. This mechanism reduces the risk of triggering dense bursts while maintaining the required tracking accuracy, thus providing a feasible communication-saving solution for bandwidth-constrained multi-stage robotic arm systems.
[0107] Step S3 includes: The method for constructing the preset performance function of the switching monotonic tube includes: Definition of the first Performance function of joke task for:
[0108] in, For the first The initial boundary magnitude at the start of the joke task. To the maximum allowable steady-state tracking error, The rate of convergence is exponential. and The first The start and end times of the joke-writing task; At the switching time The performance boundaries are reset according to the following rules:
[0109] in, The left limit value of the boundary before the switching time. For the extent of boundary expansion, This is the proportionality coefficient. To adjust the adjustable coefficients for boundary asymmetry, The instantaneous value of the tracking error at the switching moment; As one example, to adapt to the inherent instantaneous changes in multi-task trajectory tracking, the core parameters... Designed as SMT-PPF. Let the discrete switching time sequence be... At each switching moment The boundary amplitude is immediately reset to strictly track error jumps within the envelope. In this recursive mechanism, the boundary expansion amplitude... It is directly triggered by the actual tracking error at the switching moment.
[0110] Linkage position tracking error Constrained as:
[0111] in, Let be the upper boundary function. It is the lower boundary function; and The basic performance funnel curve, and This is an auxiliary transient function used to achieve finite-time convergence after the switching time. The basic performance funnel curve and The expression is:
[0112] in, For symbolic functions, This is an adjustable coefficient used to adjust the degree of asymmetry of the performance boundary relative to the initial sign of the tracking error; As one example, such as Figure 2 As shown, The maximum allowable steady-state tracking error is specified.
[0113] The auxiliary transient function and The expression is:
[0114]
[0115] in, For scaling parameters, The initial value for tracking error, and These are the initial values for the upper and lower boundaries, respectively. A finite convergence time is set for the user; when hour, and The error tends to zero uniformly, ensuring that the error is minimized within a finite time. Return to the basic performance funnel.
[0116] As one example, to meet the performance requirements of a flexible joint robotic arm during task transitions, a Switching Monotonic Pipe Preset Performance Function (SMT-PPF) was constructed. The framework was built starting with the definition of an auxiliary transient function.
[0117] Definition 1. Let and It is a scalar auxiliary function and naturally satisfies the following prerequisites: 1) Functions and their maximum values derivative of order and All are known, continuous, and globally bounded; 2) Its initial state satisfies the condition and ; 3) When time approaches a steady state When the function and its first derivative both tend to zero, that is, when hour , They represent Regarding time The first derivative; 4) Relationship For all All are true, and for any Both functions are always strictly zero.
[0118] Although all the various mathematical candidate forms conform to the properties in Definition 1, the following explicit auxiliary transient function is adopted for subsequent control synthesis. and ,in Acts as a positive scaling parameter. (Item) and Corresponding to the basic performance boundaries and The initial value.
[0119] As one embodiment, switching the preset performance function of the monotonic pipe. It can be explicitly expressed as a piecewise continuous function as follows:
[0120] This architecture ensures SMT-PPF Smoothly decays to steady-state tolerance during stable operation. Meanwhile, whenever the robotic arm encounters a sudden trajectory change, the performance monotonicity tube expands instantaneously. This flexible reset capability effectively avoids the singularity problem of error transformation and ensures that control execution is feasible in all consecutive subtasks.
[0121] Step S4 includes: Define state variables and Equation (1) can be transformed into the following form:
[0122] in, This represents the angular velocity vector of the connecting rod; This represents the angular acceleration vector of the link; Representing the inertia matrix The inverse matrix; Represents the joint stiffness matrix; The friction term relates to the angular velocity of the connecting rod. The function; The gravity vector is given by the angular position of the link. The function; Represents the matrix of Coriolis force and centrifugal force; Represents the rotor angular velocity vector; This represents the rotor angular acceleration vector; Represents the actuator inertia matrix; This represents the actuator input torque vector; Represents the actuator damping matrix; Represents the system output vector; Based on the state equation of the flexible robotic arm system described in equation (9), a state feedback controller is designed as follows: Define tracking error
[0123]
[0124] in, and For system status, To track the desired trajectory, This indicates the virtual control law that is about to be defined; The derivative of equation (10) is
[0125] Based on the state function (9) and the tracking error (11), equation (12) can be transformed into
[0126] As one embodiment, after defining the preset performance function of the preset switching monotonic transistor, in order to achieve the corresponding transient control effect, the following error transformation function is used to combine the preset performance function of the preset switching monotonic transistor with the control tracking error: Combine the preset performance function of the preset switching monotonic tube with the control tracking error:
[0127]
[0128] in, This represents the upper normalized error vector, containing the tracking errors of all joints relative to the upper boundary. The normalized ratio; Indicates the first The tracking error of each joint, i.e., the difference between the link angular position and the desired trajectory; Indicates the first The upper performance boundary of each joint is determined by the preset performance function of the switching monotonic tube; This represents the lower normalized error vector, containing the tracking errors of all joints relative to the lower boundary. The normalized ratio; Indicates the first Tracking error of each joint; This represents the joint index, with a value range of [value range missing]. to ; This indicates the number of degrees of freedom of the robotic arm, i.e., the number of joints. Indicates the first The performance boundary center offset of each joint is used to convert the asymmetric upper and lower boundaries into a symmetric form. Indicates the first The performance boundary half-width of each joint is used to normalize and scale the tracking error. For robotic arms, the following asymmetric time-varying barrier Lyapunov function is proposed:
[0129] but, The time derivative is
[0130] Substituting equation (13) into equation (14) yields
[0131] Design virtual control law To ensure the stability of the first-order subsystem:
[0132] in, Indicates the first The first virtual control law for each joint is used to stabilize the first-order error subsystem and ensure that the tracking error converges within the preset performance boundary. Indicates the first Desired trajectory of each joint The first derivative with respect to time, i.e., the desired angular velocity; Indicates the first Joint performance boundary center offset The first derivative with respect to time; Indicates the first Half-width of joint performance boundary The first derivative with respect to time; Indicates the first The positive definite matrices of each joint in the first virtual control law are used to adjust the error convergence rate. Substituting equation (16) into equation (15) yields
[0133] From the above, we can see that The time derivative is
[0134] Introducing another virtual control law Construct error variables as follows:
[0135] Consider Lyapunov functions :
[0136] Differentiating equation (18) and utilizing the oblique symmetry property of the robotic arm, i.e. Coriolis item Can be rewritten as ; item Completely offset, resulting in
[0137]
[0138]
[0139] According to equation (19), the virtual control law is set. To ensure the stability of the second-order subsystem; Define the lumped unknown continuous function as
[0140] in, It represents a lumped unknown continuous function, including the Coriolis force and centrifugal force terms, gravity terms, friction terms, and the coupling terms of the inertial matrix and the virtual control law derivative, and is used to uniformly describe all modelable but parameter-unknown nonlinear dynamics in a robotic arm system; The definition is that the left-hand symbol is defined by the right-hand expression, and it is an assignment operator; The inertia matrix is given by the angular position of the link. A symmetric positive definite matrix, describing the inertial coupling characteristics of each joint of the robotic arm; Represents the first virtual control law The first derivative with respect to time is used to compensate for dynamic changes in the virtual control law in subsequent controller design; Using radial basis function neural networks in compact sets Approximating with arbitrary desired accuracy :
[0141] in, This represents a Gaussian function vector, where each component is a radial basis function used to transform the input vector. Mapping to a high-dimensional feature space to approximate unknown nonlinear functions; It represents the bounded approximation error, reflecting the residual between the ideal approximation of the neural network and the true function; The estimated value of the weight matrix is adjusted online by an adaptive update law; The ideal value of the weight matrix is the optimal weight matrix that minimizes the approximation error. The weight estimation error matrix is defined as follows: That is, the difference between the estimated value and the ideal value, which is used to characterize the convergence of neural network adaptive learning in stability analysis; The transpose operation represents a matrix or vector; The input vector of the neural network is represented by the link angle position. Angular velocity of the connecting rod Expected trajectory Expected speed and expected acceleration Together they form; The approximation function is
[0142] Virtual control law Designed for
[0143] in ; Substituting (20) into (19), we get The time derivative is:
[0144]
[0145] From the above, we can see that The time derivative is
[0146] Introducing another virtual control law Construct error variables for
[0147] Thus obtain
[0148] Consider Lyapunov functions for
[0149] Differentiating it and using (23), we get
[0150] To eliminate cross terms and ensure the stability of the third-order subsystem, Designed for
[0151] in, This represents the positive definite gain matrix of the third virtual control law. For scalar design parameters, for 3D identity matrix; Substituting (21) and (25) into (24), we get
[0152]
[0153] in This represents the Lyapunov function designed in the third step, used to analyze the stability of the first three error subsystems; The time derivative is:
[0154] in The fourth error variable is defined as the difference between the rotor angular velocity and the third virtual control law. ; Represents the fourth error variable The first derivative with respect to time; Represents the rotor angular velocity vector The first derivative with respect to time, i.e., the rotor angular acceleration; Represents the third virtual control law The first derivative with respect to time; Represents the actuator inertia matrix The inverse matrix; This represents the actual control input torque vector, which is the final control signal applied to each joint actuator; This represents the actuator damping matrix, which is a positive definite diagonal matrix. This represents the joint stiffness matrix, which is a positive definite diagonal matrix. Represents the rotor angular position vector; Represents the angular position vector of the link; Consider Lyapunov functions :
[0155] Then, The time derivative is
[0156] Ideal continuous control law Defined as
[0157] in This represents the positive definite gain matrix of the fourth virtual control law. , Scalar design parameters; By introducing a dynamic coding event triggering mechanism, the actual control signal applied to the system is ; Define the trigger error as , can be obtained
[0158] Substituting (27) and (28) into (26), we get The time derivative is (29) The weight update law of the radial basis function neural network is designed as follows:
[0159] in Represents the estimated value of the neural network weight matrix The first derivative with respect to time is the weight adaptive update law; This represents the positive definite adaptive gain matrix, used to adjust the learning rate of the neural network weights; express - Correct parameters, Scalar design parameters are used to enhance the robustness of weight updates; Consider the following global composite Lyapunov function:
[0160] Differentiating (31) yields
[0161] Combining (30) and (32), and noting , can be obtained
[0162] Substituting (29) into (33), we get
[0163] in The trace operation represents the sum of the elements on the main diagonal of a matrix. Represents the positive definite adaptive gain matrix The inverse matrix; Represents the overall composite Lyapunov function The first derivative with respect to time; Representation matrix traces; This indicates that the scalar expression obtained by expanding the cross terms is equal to the result of its trace operation; This represents the trace of the product of the weight estimation error matrix and the weight estimation matrix; Represents the third error variable The square of the Euclidean norm; Represents the fourth error variable The square of the Euclidean norm; Indicates the second error variable The square of the Euclidean norm; ; Using Young's inequality, the cross term can be defined as
[0164] Substituting inequality (35) into (34), we obtain the final continuous dynamics:
[0165] in This represents the neural network approximation error vector; Indicates triggering error; Represents the weight estimation error matrix The square of the Frobenius norm is used to characterize the overall estimation bias of all elements of the weight matrix; The square of the Euclidean norm of the neural network approximation error vector is represented. The square of the Euclidean norm of the trigger error vector is represented. Represents the ideal weight matrix The square of the Frobenius norm.
[0166] The method also includes an output feedback control mode: when the speed state of the robotic arm link end cannot be directly measured, the unmeasurable state is estimated by designing a state observer, and the estimated value is used to replace the actual state value. Output feedback control is realized based on the adaptive state feedback controller framework; the state observer adopts a high-gain observer or a dimension-reduced observer structure.
[0167] A dynamic coded event-triggered adaptive control system for a flexible joint robotic arm, the system comprising: The signal acquisition module is used to acquire the angular position signals of each joint link and the rotor in real time; The storage module is used to store the desired tracking trajectory, the parameter configuration of the preset performance function of the switching monotonic tube, the encoding parameters of the dynamic encoding event triggering mechanism, and the initial weights of the radial basis function neural network. The module for calculating the preset performance function of the switching monotonic tube is used to calculate the time-varying performance boundary of each joint based on the current running stage and real-time tracking error, and to adaptively reset the boundary amplitude based on the error jump amplitude at the trajectory switching moment. The dynamic encoding event trigger judgment module is used to calculate the trigger error between the ideal continuous control signal and the actual control signal in real time, compare the absolute value of the trigger error with the dynamic trigger threshold, and generate a trigger signal and determine the encoding string at the current moment when the trigger error reaches or exceeds the dynamic trigger threshold. The encoding and decoding module is used to encode the symbol information and amplitude level of the trigger error into a discrete string of finite bit length for transmission at the trigger time, and decode it into the actual control torque command at the actuator end according to the received encoded string and the local storage parameters. The adaptive state feedback controller module includes a cascaded error control law based on backstepping and a radial basis function neural network adaptive approximator, which is used to generate an ideal continuous control torque signal based on the connecting rod angular position signal, the rotor angular position signal and the desired trajectory. The actual control signal output module is used to keep the control torque unchanged at the previous triggering time when it is not triggered, update it to the currently decoded control torque at the triggering time, and output the control torque to each joint actuator. The output of the preset performance function calculation module for the switching monotonic tube is connected to the input of the adaptive state feedback controller module, providing real-time error constraint boundaries for the controller; the output of the adaptive state feedback controller module is connected to the inputs of the dynamic encoding event trigger judgment module and the encoding and decoding module; the output of the encoding and decoding module is connected to the input of the actual control signal output module; the output of the actual control signal output module is connected to each joint actuator, forming a closed-loop control. In one embodiment, extensive simulations and experiments were conducted to verify the effectiveness of the proposed dynamic event-triggered adaptive control with switching preset performance schemes. A six-DOF flexible articulated robotic arm was considered. Simulations were performed.
[0168] State feedback control simulation: The parameter settings for the flexible joint robotic arm are as follows: The initial configuration of the rotor angular position is set as follows: This allows for the dynamic calculation of the inertia matrix using the inertia() method in RoboticsToolbox. The gain parameter of the backstepping controller is set to...
[0169] For RBFNNs, considering the dimension of the input vector, this neural network contains 324 nodes, and its basis function width is... The adaptive update parameter is selected as follows: and All initial states of the robotic arm and neural network weights are initialized to zero.
[0170] The parameter design for switching the monotonic tubular preset performance function (SMT-PPF) is as follows: and To implement the Dynamic Coded Event Triggering Scheme (DCES), the encoding cardinality and length are set to... and The dynamic threshold parameter is selected as follows: and Reference trajectory Designed specifically for A mutation occurs at one of the locations, in which the first mutation occurs at the second location. element Defined as:
[0171] Accordingly, performance boundaries are allowed to be dynamically reset at these moments to avoid singularity issues. The total simulation time is set to... Step size is .
[0172] Simulation results for joints 1 and 2 are shown in the following figures. Figure 3 and Figure 4 The upper subplot compares the position tracking performance of the proposed adaptive state feedback control with that of a PID benchmark controller. The parameters of the PID controller are set to... and .exist and The non-smooth abrupt change occurs at this point, which is due to the reference trajectory. This is caused by the PID controller exhibiting severe position oscillations with amplitudes exceeding 10 rad at joint 1, failing to meet tracking requirements. The proposed controller suppresses physical resonance and controls the maximum position deviation to within 0.5 rad during jump transients, thereby achieving tracking of the reference trajectory.
[0173] The intermediate subplot depicts the tracking errors of each joint. The changes are as shown in the trajectory tracking results in the top subplot. The PID controller produced a large tracking error, with fluctuations exceeding [the expected range]. The proposed controller limits the tracking error within specified performance boundaries. At the reset time, a sudden change in the reference signal introduces an instantaneous tracking error of approximately 0.9 rad. SMT-PPF extends the boundary range to 2.0 rad to encompass the instantaneous error, thus avoiding the singularity problem of the error transformation function. An auxiliary function guarantees that the trajectory error remains within a specified finite time. It returns to the basic exponential channel. This is determined by the decay rate. Driven by this, the error continues to converge exponentially to the steady-state boundary. .
[0174] The bottom subplot shows the actual control inputs and event trigger communication moments. In practical applications, the robotic arm's actuators have absolute maximum torque constraints; therefore, the continuous control torque and trigger inputs in the simulation are strictly limited to [specific values]. Nm. Simulation results show that the actual control input remains within these physical saturation limits throughout the entire task. Furthermore, the piecewise constant characteristic of the control torque reflects the discrete execution of the DCES.
[0175] The control input is updated only when the dynamic threshold condition is violated. The trigger ratio is... This reduces the communication burden compared to continuous time-triggered control.
[0176] The above are merely exemplary embodiments of this disclosure and should not be construed as limiting the scope of this disclosure. Any equivalent changes and modifications made in accordance with the teachings of this disclosure shall still fall within the scope of this disclosure.
[0177] This application is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not described in this disclosure. The specification and embodiments are to be considered exemplary only, and the scope and spirit of this disclosure are defined by the claims.
Claims
1. A dynamic coding event-triggered adaptive control method for a flexible joint robotic arm, characterized in that, The method includes the following steps: Step S1: Establish a dynamic model of the flexible joint manipulator described by the coupling of link dynamics and rotor dynamics; Step S2: Design a dynamic encoding event triggering mechanism to reduce the communication burden between the controller and the actuator; encode the control signal to be transmitted into a discrete string of finite bit length, the discrete string containing both the sign and amplitude of the trigger error; introduce a dynamic threshold parameter into the triggering condition; Step S3: Design a preset performance function for switching monotonic tubes to constrain link position tracking error; for the case where the robotic arm executes multiple discontinuous reference trajectories, construct a monotonically decaying performance boundary for each continuous sub-task, and adaptively expand the performance boundary at the switching moment when the reference trajectory changes. Step S4: Based on the preset performance function of the switching monotonic tube and the dynamic coding event triggering mechanism, the state feedback controller is designed using the backstepping method. The radial basis function neural network is used to approximate the lumped unknown nonlinear dynamics of the system online to obtain the actual control torque of each joint. Step S5: Based on the actual control torque of each joint, the control signal is encoded and transmitted through a dynamic coding event triggering mechanism. When the event triggering condition is met, the actual control signal of the actuator is updated, and the dynamic threshold parameter is updated at the same time. In each control cycle, it is determined whether the triggering condition is met. If it is met, steps S4 to S5 are repeated; if it is not met, the current control input is kept unchanged to realize the dynamic closed-loop control of the flexible joint robot arm.
2. The method for dynamic coding event-triggered adaptive control of a flexible joint robotic arm according to claim 1, characterized in that, Step S1 includes: Dynamic model of flexible joint robotic arm: in, These represent the angular position, angular velocity, and angular acceleration vectors of the link, respectively. This represents the angular position, angular velocity, and angular acceleration vectors on the rotor side. The inertia matrix; The vectors of the Coriolis force and the centrifugal force; It is the gravity vector; This is the friction term; These are the joint stiffness matrix, actuator inertia matrix, and damping matrix, respectively. The input torque vector for each actuator; denoted as the number of degrees of freedom of the robotic arm.
3. The method for dynamic coding event-triggered adaptive control of a flexible joint robotic arm according to claim 1, characterized in that, Step S2 includes: The encoding and triggering logic of the dynamic encoding event triggering mechanism includes: Continuous control signal With the actual output of the actuator The difference between them is defined as the trigger error. ; Event trigger time sequence Determined by the following triggering conditions: in, Indicates the sequence number of the trigger time; Represents the continuous control signal to be designed; even integer. The base of the encoding number system is an even number. Indicates the first The encoding base of the power; For the first Weighting coefficients for joke-related tasks; This is a dynamic threshold parameter; Gain is adjusted using a dynamic threshold. The variables in equation (2) are governed by the following update law: in, This represents the amplitude level index, used to determine the quantization range to which the current control input amplitude belongs; This represents a range index, with values ranging from... arrive An integer used to identify the interval number where the control input amplitude is located; Indicates the total number of intervals; Indicates the first The lower boundary of each amplitude quantization interval and Together constitute the first A range of intervals; Indicates the bit length of the transmitted string; Indicates the first Each trigger time base The first in the encoded string The value of the bit takes the value of arrive Integers between; Indicates the encoding radix of The power is used to calculate the weighted value of each bit in the encoded string; The base of the encoded number system is an even number. Representation of range index The corresponding trigger threshold weight coefficient is a piecewise constant positive scalar; This represents the dynamic threshold attenuation coefficient, which is a design parameter. Represents dynamic threshold parameter The first derivative with respect to time is used to describe the continuous evolution of the dynamic threshold; When the trigger condition (2) is satisfied, the sign and magnitude of the trigger error are encoded into a string of length [length missing]. discrete strings Transmission is performed, with the most significant bit... The symbol representing the trigger error, the rest. The bit indicates the magnitude level of the trigger error.
4. The method for dynamic coding event-triggered adaptive control of a flexible joint robotic arm according to claim 3, characterized in that, Step S2 also includes: The decoding rule of the actuator is as follows: in, This is the actual control input after decoding. The value is the control signal value from the previous trigger moment; the actuator then determines the value based on the received discrete string. and local storage , Parameters, combined with dynamic threshold parameters Reconstruct control signals.
5. The method for dynamic coding event-triggered adaptive control of a flexible joint robotic arm according to claim 1, characterized in that, Step S3 includes: The method for constructing the preset performance function of the switching monotonic tube includes: Definition of the first Performance function of joke task for: in, For the first The initial boundary magnitude at the start of the joke task. To the maximum allowable steady-state tracking error, The rate of convergence is exponential. and The first The start and end times of the joke task; At the switching time The performance boundaries are reset according to the following rules: in, The left limit value of the boundary before the switching time. For the extent of boundary expansion, This is the proportionality coefficient. To adjust the adjustable coefficients for boundary asymmetry, The instantaneous value of the tracking error at the switching moment; Linkage position tracking error Constrained as: in, Let be the upper boundary function. It is the lower boundary function; and The basic performance funnel curve, and This is an auxiliary transient function used to achieve finite-time convergence after the switching time. The basic performance funnel curve and The expression is: in, For symbolic functions, This is an adjustable coefficient used to adjust the degree of asymmetry of the performance boundary relative to the initial sign of the tracking error; The auxiliary transient function and The expression is: in, For scaling parameters, As the initial value for tracking error, and These are the initial values for the upper and lower boundaries, respectively. A finite convergence time is set for the user; when hour, and The error tends to zero uniformly, ensuring that the error is minimized within a finite time. Return to the basic performance funnel.
6. The method for dynamic coding event-triggered adaptive control of a flexible joint robotic arm according to claim 2, characterized in that, Step S4 includes: Define state variables and Equation (1) can be transformed into the following form: in, This represents the angular velocity vector of the connecting rod; This represents the angular acceleration vector of the link; Representing the inertia matrix The inverse matrix; Represents the joint stiffness matrix; The friction term relates to the angular velocity of the connecting rod. The function; The gravity vector is given by the angular position of the link. The function; Represents the matrix of Coriolis force and centrifugal force; Represents the rotor angular velocity vector; This represents the rotor angular acceleration vector; Represents the actuator inertia matrix; This represents the actuator input torque vector; Represents the actuator damping matrix; Represents the system output vector; Based on the state equation of the flexible robotic arm system described in equation (9), a state feedback controller is designed as follows: Define tracking error in, and For system status, To track the desired trajectory, This indicates the virtual control law that is about to be defined; The derivative of equation (10) is Based on the state function (9) and the tracking error (11), equation (12) can be transformed into Combine the preset performance function of the preset switching monotonic tube with the control tracking error: in, This represents the upper normalized error vector, containing the tracking errors of all joints relative to the upper boundary. The normalized ratio; Indicates the first The tracking error of each joint, i.e., the difference between the link angular position and the desired trajectory; Indicates the first The upper performance boundary of each joint is determined by the preset performance function of the switching monotonic tube; This represents the lower normalized error vector, containing the tracking errors of all joints relative to the lower boundary. The normalized ratio; Indicates the first Tracking error of each joint; This represents the joint index, with a value range of [value range missing]. to ; This indicates the number of degrees of freedom of the robotic arm, i.e., the number of joints. Indicates the first The performance boundary center offset of each joint is used to convert the asymmetric upper and lower boundaries into a symmetric form. Indicates the first The performance boundary half-width of each joint is used to normalize and scale the tracking error. For robotic arms, the following asymmetric time-varying barrier Lyapunov function is proposed: but, The time derivative is Substituting equation (13) into equation (14) yields Design virtual control law To ensure the stability of the first-order subsystem: in, Indicates the first The first virtual control law for each joint is used to stabilize the first-order error subsystem and ensure that the tracking error converges within the preset performance boundary. Indicates the first Desired trajectory of each joint The first derivative with respect to time, i.e., the desired angular velocity; Indicates the first Joint performance boundary center offset The first derivative with respect to time; Indicates the first Half-width of joint performance boundary The first derivative with respect to time; Indicates the first The positive definite matrices of each joint in the first virtual control law are used to adjust the error convergence rate. Substituting equation (16) into equation (15) yields From the above, we can see that The time derivative is Introducing another virtual control law Construct error variables as follows: Consider Lyapunov functions : Differentiating equation (18) and utilizing the oblique symmetry property of the robotic arm, i.e. Coriolis item Can be rewritten as ; item Completely offset, resulting in According to equation (19), the virtual control law is set. To ensure the stability of the second-order subsystem; Define the lumped unknown continuous function as in, It represents a lumped unknown continuous function, including the Coriolis force and centrifugal force terms, gravity terms, friction terms, and the coupling terms of the inertial matrix and the virtual control law derivative, and is used to uniformly describe all modelable but parameter-unknown nonlinear dynamics in a robotic arm system; The definition is that the left-hand symbol is defined by the right-hand expression, and it is an assignment operator; The inertia matrix is given by the angular position of the link. A symmetric positive definite matrix, describing the inertial coupling characteristics of each joint of the robotic arm; Represents the first virtual control law The first derivative with respect to time is used to compensate for dynamic changes in the virtual control law in subsequent controller design; Using radial basis function neural networks in compact sets Approximating with arbitrary desired accuracy : in, This represents a Gaussian function vector, where each component is a radial basis function used to transform the input vector. Mapping to a high-dimensional feature space to approximate unknown nonlinear functions; It represents the bounded approximation error, reflecting the residual between the ideal approximation of the neural network and the true function; The estimated value of the weight matrix is adjusted online by an adaptive update law; The ideal value of the weight matrix is the optimal weight matrix that minimizes the approximation error. The weight estimation error matrix is defined as follows: That is, the difference between the estimated value and the ideal value, which is used to characterize the convergence of neural network adaptive learning in stability analysis; The transpose operation represents a matrix or vector; The input vector of the neural network is represented by the link angle position. Angular velocity of the connecting rod Expected trajectory Expected speed and expected acceleration Together they form; The approximation function is Virtual control law Designed for in ; Substituting (20) into (19), we get The time derivative is: From the above, we can see that The time derivative is Introducing another virtual control law Construct error variables for Thus obtain Consider Lyapunov functions for Differentiating it and using (23), we get To eliminate cross terms and ensure the stability of the third-order subsystem, Designed for in, This represents the positive definite gain matrix of the third virtual control law. For scalar design parameters, for 3D identity matrix; Substituting (21) and (25) into (24), we get in This represents the Lyapunov function designed in the third step, used to analyze the stability of the first three error subsystems; The time derivative is: in The fourth error variable is defined as the difference between the rotor angular velocity and the third virtual control law. ; Represents the fourth error variable The first derivative with respect to time; Represents the rotor angular velocity vector The first derivative with respect to time, i.e., the rotor angular acceleration; Represents the third virtual control law The first derivative with respect to time; Represents the actuator inertia matrix The inverse matrix; This represents the actual control input torque vector, which is the final control signal applied to each joint actuator; This represents the actuator damping matrix, which is a positive definite diagonal matrix. This represents the joint stiffness matrix, which is a positive definite diagonal matrix. Represents the rotor angular position vector; Represents the angular position vector of the link; Consider Lyapunov functions : Then, The time derivative is Ideal continuous control law Defined as in This represents the positive definite gain matrix of the fourth virtual control law. , Scalar design parameters; By introducing a dynamic encoded event triggering mechanism, the actual control signal applied to the system is ; Define the trigger error as , can be obtained Substituting (27) and (28) into (26), we get The time derivative is (29) The weight update law of the radial basis function neural network is designed as follows: in Represents the estimated value of the neural network weight matrix The first derivative with respect to time is the weight adaptive update law; This represents the positive definite adaptive gain matrix, used to adjust the learning rate of the neural network weights; express - Correct parameters, Scalar design parameters are used to enhance the robustness of weight updates; Consider the following global composite Lyapunov function: Differentiating (31) yields Combining (30) and (32), and noting , can be obtained Substituting (29) into (33), we get in The trace operation represents the sum of the elements on the main diagonal of a matrix. Represents the positive definite adaptive gain matrix The inverse matrix; Represents the overall composite Lyapunov function The first derivative with respect to time; Representation matrix traces; This indicates that the scalar expression obtained by expanding the cross terms is equal to the result of its trace operation; This represents the trace of the product of the weight estimation error matrix and the weight estimation matrix; Represents the third error variable The square of the Euclidean norm; Represents the fourth error variable The square of the Euclidean norm; Indicates the second error variable The square of the Euclidean norm; ; Using Young's inequality, the cross term can be defined as Substituting inequality (35) into (34), we obtain the final continuous dynamics: in This represents the neural network approximation error vector; Indicates triggering error; Represents the weight estimation error matrix The square of the Frobenius norm is used to characterize the overall estimation bias of all elements of the weight matrix; The square of the Euclidean norm of the neural network approximation error vector is represented. The square of the Euclidean norm of the trigger error vector is represented. Represents the ideal weight matrix The square of the Frobenius norm.
7. The method for dynamic coding event-triggered adaptive control of a flexible joint robotic arm according to claim 1, characterized in that, The method also includes an output feedback control mode: when the speed state of the robotic arm link end cannot be directly measured, the unmeasurable state is estimated by designing a state observer, and the estimated value is used to replace the actual state value. Output feedback control is realized based on the adaptive state feedback controller framework; the state observer adopts a high-gain observer or a dimension-reduced observer structure.
8. A dynamic coded event-triggered adaptive control system for a flexible joint robotic arm, used to implement the dynamic coded event-triggered adaptive control method for a flexible joint robotic arm as described in any one of claims 1-7, characterized in that, The system includes: The signal acquisition module is used to acquire the angular position signals of each joint link and the rotor in real time; The storage module is used to store the desired tracking trajectory, the parameter configuration of the preset performance function of the switching monotonic tube, the encoding parameters of the dynamic encoding event triggering mechanism, and the initial weights of the radial basis function neural network. The module for calculating the preset performance function of the switching monotonic tube is used to calculate the time-varying performance boundary of each joint based on the current running stage and real-time tracking error, and to adaptively reset the boundary amplitude based on the error jump amplitude at the time of trajectory switching. The dynamic encoding event trigger judgment module is used to calculate the trigger error between the ideal continuous control signal and the actual control signal in real time, compare the absolute value of the trigger error with the dynamic trigger threshold, and generate a trigger signal and determine the encoding string at the current moment when the trigger error reaches or exceeds the dynamic trigger threshold. The encoding and decoding module is used to encode the symbol information and amplitude level of the trigger error into a discrete string of finite bit length for transmission at the trigger time, and decode it into the actual control torque command at the actuator end according to the received encoded string and the local storage parameters. The adaptive state feedback controller module includes a cascaded error control law based on backstepping and a radial basis function neural network adaptive approximator, which is used to generate an ideal continuous control torque signal based on the connecting rod angular position signal, the rotor angular position signal and the desired trajectory. The actual control signal output module is used to keep the control torque of the previous trigger moment unchanged when it is not triggered, update it to the currently decoded control torque when it is triggered, and output the control torque to each joint actuator. The output of the switching monotonic tube preset performance function calculation module is connected to the input of the adaptive state feedback controller module, providing the controller with real-time error constraint boundaries; the output of the adaptive state feedback controller module is connected to the input of the dynamic encoding event trigger judgment module and the encoding and decoding module; the output of the encoding and decoding module is connected to the input of the actual control signal output module; the output of the actual control signal output module is connected to each joint actuator, forming a closed-loop control.
