A dynamic stability control system for a power-assisted robotic arm pneumatic system

Abstract

This invention discloses a dynamic stability control system for a power-assisted robotic arm pneumatic system. The main control module generates the desired motion trajectory. In the core control module, a multi-rate state observation unit fuses displacement and pressure signals with different sampling frequencies to generate a high-frequency, low-latency system state estimate. A dual state observation unit estimates the system model parameters and lumped disturbances online, generating parameter updates and disturbance compensation. A feedforward-feedback control unit integrates the desired trajectory, state estimate, parameter updates, and disturbance compensation, generating a control signal through feedforward, feedback, and disturbance compensation fusion. The pneumatic execution module drives joint movement based on this signal. This application improves the real-time performance of state perception through multi-rate fusion observation. Combining dual observation and adaptive feedforward-feedback control, it effectively overcomes the control challenges caused by the strong nonlinearity and time-varying parameters of the pneumatic system, significantly improving the trajectory tracking accuracy and dynamic stability of the power-assisted robotic arm.

Description

Technical Field

[0001] This invention relates to the field of robot control technology, and in particular to a dynamic stability control system for a power-assisted robotic arm pneumatic system. Background Technology

[0002] In existing technologies, assisted robotic arms typically employ pneumatic drives to achieve smooth, lightweight human-machine interaction. However, the inherent strong nonlinearity of pneumatic systems, the compressibility of gases, and the time-varying characteristics of parameters pose significant challenges to high-precision, high-dynamic-response force / position control. To improve control performance, current common solutions often combine model-based feedforward compensation with traditional PID or fuzzy feedback control.

[0003] However, such solutions still face several bottlenecks in practical applications: First, accurate state feedback is the foundation of advanced control algorithms, but due to cost and size constraints, practical systems often use heterogeneous sensors with significant differences in sampling frequency and communication delay (such as low-frequency bus-type displacement sensors and high-frequency analog pressure sensors). The asynchrony and lag of signals lead to insufficient state observation accuracy and real-time performance, thus limiting the control bandwidth. Second, parameters such as the load and joint friction of the robotic arm change drastically with the wearer's posture and movement state. Fixed feedforward models are difficult to accurately compensate for these changes, and simple online parameter identification algorithms are prone to failure when the system excitation is insufficient. Furthermore, they have slow response speed and limited suppression capabilities for unmodeled dynamics and external disturbances (such as human interaction forces). Finally, traditional control architectures typically design state observation, parameter identification, and disturbance suppression in isolation, lacking coordination. This makes it difficult to simultaneously ensure the system's tracking accuracy, response speed, and robust stability under complex conditions such as model uncertainty, sudden load changes, and sensor limitations. Therefore, existing pneumatic control systems for power-assisted robotic arms have significant shortcomings in terms of real-time state perception, collaborative estimation and compensation of model parameters and unknown disturbances, and adaptive control based on multi-source information fusion, which severely restrict their dynamic performance and practical application.

[0004] Therefore, the industry urgently needs an integrated solution that can deeply integrate high-precision state observation, intelligent parameter identification, rapid disturbance suppression and adaptive control strategies to fundamentally improve the dynamic stability and trajectory tracking accuracy of assistive robotic arms in various operating scenarios. Summary of the Invention

[0005] To improve upon existing methods and systems, this application provides a dynamic stability control system for a power-assisted robotic arm pneumatic system, comprising: Main control module, control core module, and pneumatic actuator module: The main control module is configured to generate the desired motion trajectories of each joint of the robotic arm; The control core module, connected to the main control module, is used to receive the desired motion trajectory and generate control signals; the control core module includes a multi-rate state observation unit, a dual-state observation unit, and a feedforward-feedback control unit. The multi-rate state observation unit is configured to receive displacement and pressure signals based on the joints of the robotic arm with different sampling frequencies and communication delays, perform data fusion and prediction, and generate high-frequency, low-latency system state estimates. The dual-state observation unit, connected to the multi-rate state observation unit, is configured to receive system state estimates and estimate system model parameters and lumped disturbances online, generating parameter updates and disturbance compensations. The feedforward-feedback control unit is connected to the main control module, the multi-rate state observation unit, and the dual state observation unit, respectively. It is configured to receive the desired motion trajectory, the system state estimate, the parameter update, and the disturbance compensation, and calculate the feedforward control quantity based on the desired motion trajectory, the system state estimate, and the parameter update. At the same time, it calculates the feedback control quantity based on the deviation between the desired motion trajectory and the system state estimate. Finally, it fuses the feedforward control quantity, the feedback control quantity, and the disturbance compensation quantity to generate a control signal. The pneumatic actuator module, connected to the control core module, is configured to drive the joint movement of the robotic arm according to control signals.

[0006] In some embodiments, the dual-state observation unit includes an extended state observer and an event-triggered parameter identifier; The extended state observer is configured to uniformly extend the system model uncertainty and external disturbances into new state variables and observe them in real time, outputting an estimate of the lumped disturbance as the disturbance compensation amount. The event-triggered parameter identifier is configured to monitor joint motion acceleration or load changes determined based on system state estimates. It is triggered when the joint motion acceleration or load changes exceed a preset threshold, and based on the system input and output data before and after the triggering time, it recursively identifies key time-varying parameters in the system dynamics model online and outputs parameter update values.

[0007] In some embodiments, the control core module further includes a performance evaluation and switching logic unit, the input of which is connected to the output of the extended state observer and the event trigger parameter identifier, and the output of which is connected to the feedforward-feedback control unit. The performance evaluation and switching logic unit is configured to determine whether the system is in a stable or transient state based on the characteristics of the trajectory tracking error, and accordingly dynamically adjust the fusion weight of the disturbance compensation amount received from the dual-state observation unit and the feedforward control amount based on the parameter update amount in the final control signal of the feedforward-feedback control unit.

[0008] In some embodiments, the dual-state observation unit further includes a coordination module configured to perform the following operations: Receive the lumped disturbance estimate output from the extended state observer; The part of the lumped disturbance estimate that is linearly related to the model parameters is analyzed as an incremental correction term for the parameters to be identified in the event-triggered parameter identifier; When the event-triggered parameter identifier is triggered, the incremental correction term is used as the initial deviation input for parameter identification or as a weighting term in the identification process.

[0009] In some embodiments, the multi-rate state observation unit includes a signal fusion module and a delay prediction compensation module; The signal fusion module is configured to receive low-frequency displacement sampling signals from the displacement sensor and high-frequency pressure signals from the pressure sensor, and to use the high-frequency pressure signals and system dynamics to calculate a high-frequency velocity estimation signal based on the differential of the pressure difference between the two chambers of the cylinder in real time. The delay prediction and compensation module, coupled to the signal fusion module, is configured to establish a system state-space model that includes known communication delays. It uses low-frequency displacement signals as absolute position references and high-frequency velocity estimation signals as high-frequency change quantities. It performs multi-rate data fusion through the Kalman filter algorithm and performs one-step forward prediction on the known communication delays. Finally, it outputs the compensated high-frequency displacement and velocity estimates as system state estimates.

[0010] In some embodiments, a sensor group is also included, which includes a displacement sensor, a first pressure sensor, and a second pressure sensor. The displacement sensor is installed at the joint and configured to measure the actual displacement of the joint. The output signal of the displacement sensor is periodically uploaded via the fieldbus at a first frequency. The first pressure sensor and the second pressure sensor are installed in the rod chamber and rodless chamber of the cylinder, respectively, and are configured to measure the pressure in the two chambers in real time. The output signals of the first pressure sensor and the second pressure sensor are directly acquired and input to the multi-rate state observation unit at a second frequency higher than the first frequency.

[0011] In some embodiments, when the event-triggered parameter identifier performs recursive identification, the system model regression vector used includes a real-time pressure difference term and its real-time differential term calculated based on the cylinder two-chamber pressure signal, which are used to identify the joint equivalent friction coefficient and load mass.

[0012] In some embodiments, the main control module 1 includes a trajectory planning unit and an inverse kinematics unit connected in sequence. The trajectory planning unit is configured to generate an end-effector desired pose sequence according to the task instructions, and the inverse kinematics unit is configured to solve the end-effector desired pose sequence into desired angles, desired angular velocities and desired angular accelerations of each joint.

[0013] In some embodiments, the feedforward-feedback control unit includes a feedforward channel and a feedback channel; The feedforward channel is configured to calculate the feedforward force / torque based on the desired motion trajectory and the parameter update provided by the dual-state observation unit, and use it as the feedforward control quantity through the inverse dynamics model. The feedback channel is configured to calculate the feedback force / torque as the feedback control quantity based on the deviation between the desired motion trajectory and the system state estimate provided by the multi-rate state observation unit using a proportional-derivative control law. The feedforward-feedback control unit also includes a fusion module, which is configured to add the feedforward force / torque, feedback force / torque and disturbance compensation amount to generate the final control force / torque command, and output it as the desired pressure control signal for each cylinder via the force / torque-pressure conversion module, as the control signal.

[0014] In some embodiments, the pneumatic actuator module includes a proportional pressure valve and a cylinder. The electrical interface of the proportional pressure valve is connected to the output of the control core module, and the pneumatic interface of the proportional pressure valve is connected to the rod chamber and rodless chamber of the cylinder, and is configured to adjust the pressure of the two chambers of the cylinder according to the control signal.

[0015] This invention provides a dynamic stability control system for a power-assisted robotic arm's pneumatic system. The core control module of this system includes a multi-rate state observation unit, a dual-state observation unit, and a feedforward-feedback control unit. The multi-rate state observation unit generates a high-precision, low-latency real-time estimate of the system state by fusing asynchronous signals from displacement and pressure sensors. Based on this state estimate, the dual-state observation unit online collaboratively estimates time-varying model parameters and lumped external disturbances, and outputs parameter updates and disturbance compensation amounts respectively. The feedforward-feedback control unit integrates the desired trajectory, real-time state, updated model parameters, and disturbance compensation amounts, and generates the final pneumatic control signal by calculating and fusing feedforward control, feedback control, and disturbance compensation amounts to drive the actuator module's motion. This solution effectively overcomes the control challenges posed by the strong nonlinearity of pneumatic systems, time-varying parameters, and external disturbances through high-precision state reconstruction and collaborative online estimation and compensation of parameters and disturbances. It significantly improves the trajectory tracking accuracy, dynamic response speed, and operational stability and robustness of the power-assisted robotic arm under complex interactive conditions. Attached Figure Description

[0016] Figure 1This is a system architecture diagram of the dynamic stability control system for the pneumatic system of the assisted robotic arm proposed in this invention.

[0017] In the diagram: 1. Main control module; 2. Control core module; 3. Pneumatic actuator module; 21. Multi-rate state observation unit; 22. Dual state observation unit; 23. Feedforward-feedback control unit. Detailed Implementation

[0018] The following description is intended to disclose the invention and enable those skilled in the art to implement it. The preferred embodiments described below are merely examples, and other obvious variations will occur to those skilled in the art.

[0019] This application provides a dynamic stability control system for a power-assisted robotic arm pneumatic system, such as... Figure 1 As shown, it includes: Main control module 1, control core module 2, and pneumatic actuator module 3: Main control module 1 is configured to generate the desired motion trajectories of each joint of the robotic arm; The control core module 2 is connected to the main control module 1 and is used to receive the desired motion trajectory and generate control signals. The control core module 2 includes a multi-rate state observation unit 21, a dual state observation unit 22, and a feedforward-feedback control unit 23. The multi-rate state observation unit 21 is configured to receive displacement and pressure signals based on the joints of the robotic arm with different sampling frequencies and communication delays, perform data fusion and prediction, and generate high-frequency, low-latency system state estimates. The dual-state observation unit 22, connected to the multi-rate state observation unit 21, is configured to receive system state estimates and estimate system model parameters and lumped disturbances online, generating parameter updates and disturbance compensation amounts. The feedforward-feedback control unit 23 is connected to the main control module 1, the multi-rate state observation unit 21, and the dual state observation unit 22, respectively. It is configured to receive the desired motion trajectory, the system state estimate, the parameter update, and the disturbance compensation, and calculate the feedforward control quantity based on the desired motion trajectory, the system state estimate, and the parameter update. At the same time, it calculates the feedback control quantity based on the deviation between the desired motion trajectory and the system state estimate. Finally, it fuses the feedforward control quantity, the feedback control quantity, and the disturbance compensation quantity to generate a control signal. The pneumatic actuator module 3, connected to the control core module 2, is configured to drive the joint movement of the robotic arm according to the control signal.

[0020] In this embodiment, a dynamic stability control system for a power-assisted robotic arm pneumatic system specifically includes a main control module 1, a control core module 2, and a pneumatic execution module 3. The main control module 1 is responsible for trajectory planning and inverse kinematics solution based on high-level task instructions, thereby generating smooth and continuous desired position, velocity, and acceleration trajectories for each joint. This is the command source for the entire control system. The control core module 2 is the core for achieving high-precision dynamic stability, and its design fully considers the multiple challenges faced by pneumatic systems in practical applications. Specifically, the system deploys displacement sensors (such as absolute encoders) installed at the joints and pressure sensors installed in the two chambers of the cylinder. Displacement signals are typically uploaded periodically at a low frequency via a fieldbus, resulting in inherent communication delays; while pressure signals are acquired at high speed in analog form, with extremely low delays but cannot directly reflect position. The multi-rate state observation unit 21 is designed to solve this signal heterogeneity problem. It receives two asynchronous and different frequency original signals, establishes a system state space model containing known communication delays, and uses algorithms such as Kalman filtering to fuse the low-frequency displacement signal and the velocity signal derived from the high-frequency pressure differential and perform one-step forward prediction, thereby generating high-frequency, low-delay joint displacement and velocity estimates. This is equivalent to building a high-performance "virtual sensor" and providing an accurate and timely feedback basis for subsequent advanced control algorithms.

[0021] Based on this, the dual-state observation unit 22 receives the aforementioned high-precision state estimates, aiming to address system model uncertainties and external disturbances. This unit comprises two core components: an extended state observer (ESO) and an event-triggered parameter identifier. The extended state observer (ESO) is designed to treat all dynamics in the system that are difficult to model precisely (such as nonlinear friction and unmodeled dynamics) and external random disturbances (such as unknown interaction forces applied by the wearer) as a unified "total disturbance" and perform real-time observation and estimation, outputting the corresponding disturbance compensation amount. Its advantage lies in its rapid response to disturbances. However, the compensation of the ESO is a generalized compensation. To make the control more forward-looking, it is also necessary to accurately identify key time-varying parameters in the model (such as load mass and friction coefficient). For this purpose, the event-triggered parameter identifier operates synchronously, but it does not work continuously. Instead, it is triggered only when the system state (such as acceleration) is detected to exceed a set threshold, i.e., when the excitation is sufficient, to perform an accurate online parameter identification and update the model parameters. This dual-observation architecture, combining "rapid disturbance observation" with "precise on-demand parameter identification," ensures both the system's rapid disturbance rejection capability and the accuracy and computational efficiency of model parameter updates. Finally, the feedforward-feedback control unit 23 integrates all information. It utilizes the desired trajectory from the main control module 1, the real-time state estimation from the multi-rate state observation unit 21, and the latest model parameters and disturbance compensation from the dual-state observation unit 22 for separate calculations. Specifically, it calculates the feedforward control quantity based on the updated model and desired trajectory to achieve accurate model following; it calculates the feedback control quantity (such as PD control) based on the trajectory tracking error to suppress residual deviation; and finally, it fuses the feedforward control quantity, feedback control quantity, and real-time disturbance compensation quantity to generate the final control force / torque command, which is then converted and output as the desired pressure signal for each cylinder's proportional pressure valve. The pneumatic actuator module 3 receives this control signal and drives the proportional valve to precisely adjust the pressure in both chambers of the cylinder, thereby generating the required joint movement.

[0022] In summary, this system addresses the real-time and accuracy issues at the perception level through the "multi-rate state observation unit 21," resolves the uncertainty issues at the model and disturbance levels through the "dual state observation unit 22," and finally coordinates and makes decisions through the "feedforward-feedback control unit 23," forming a complete closed loop from accurate perception and intelligent estimation to optimized decision-making. Its beneficial effects are significant and direct: high-precision, low-latency state estimation lays a reliable feedback foundation for the entire control system; the synergy between ESO and event-triggered parameter identification enables the system to quickly respond to unmodeled disturbances and accurately follow parameter changes, enhancing adaptability and robustness; ultimately, the control commands integrating precise feedforward, stable feedback, and real-time disturbance compensation enable the assisted robotic arm to achieve high-precision, high-dynamic-response stable trajectory tracking even under conditions of load changes, human interaction force interference, and sensor limitations, significantly improving the compliance and operational efficiency of human-machine collaboration.

[0023] In some embodiments, the dual-state observation unit 22 includes an extended state observer and an event-triggered parameter identifier; The extended state observer is configured to uniformly extend the system model uncertainty and external disturbances into new state variables and observe them in real time, outputting an estimate of the lumped disturbance as the disturbance compensation amount. The event-triggered parameter identifier is configured to monitor joint motion acceleration or load changes determined based on system state estimates. It is triggered when the joint motion acceleration or load changes exceed a preset threshold, and based on the system input and output data before and after the triggering time, it recursively identifies key time-varying parameters in the system dynamics model online and outputs parameter update values.

[0024] In this embodiment, the dual-state observation unit 22 is implemented through the collaborative work of an extended state observer and an event-triggered parameter identifier. The implementation of the extended state observer typically involves constructing state-space equations based on the nominal dynamic model of the controlled system and uniformly expanding all dynamics inconsistent with the model (such as nonlinear friction, unmodeled flexibility, and external load forces) into an additional "disturbance state." By designing appropriate observer gains, this observer can utilize the system's control inputs (such as valve signals) and measurable outputs (such as the state estimates provided by the aforementioned multi-rate state observation unit 21) to estimate this expanded "total disturbance" state in real time. Its output is the disturbance compensation amount that can be directly used for forward channel compensation. The implementation of the event-triggered parameter identifier first requires defining a reasonable triggering condition, such as continuously monitoring the rate of change of joint angular acceleration calculated based on high-precision state observations, or the rate of change of pressure difference between the two chambers of a cylinder. When the absolute value of this rate of change exceeds a preset threshold, it is determined that the system has received significant excitation (such as a sudden load change or the start of rapid motion), at which point the trigger is activated. Once triggered, the identifier immediately collects the system input-output data sequence within a time window before and after the trigger moment. It then employs online identification algorithms such as recursive least squares with a forgetting factor to re-estimate and update key time-varying parameters in the dynamic equations (such as the equivalent viscous friction coefficient of joints and the load mass moment), outputting the latest parameter updates. Furthermore, a more advanced implementation involves establishing a collaborative mechanism between the two. For example, the "total disturbance" signal estimated in real-time by the extended state observer can be analyzed through a pre-designed mapping relationship to extract components linearly related to specific physical parameters (such as friction). These components can then be used as prior information when the event-triggered parameter identifier starts or as weighting terms in the iteration process, thereby guiding the identification process to converge to the true value more quickly.

[0025] This design brings several significant benefits. First, the extended state observer provides a model-insensitive and rapidly responsive general disturbance compensation method. It packages and cancels various complex uncertainties in real time, directly enhancing the system's immediate robustness in the face of unknown changes and external disturbances. Even if the parameter identifier is not working, the system can maintain basic stability. Second, the event-triggered mechanism fundamentally solves the inherent problems of traditional continuous online identification: when the system is running smoothly and the stimulus is insufficient, the identifier automatically goes into sleep mode, avoiding parameter estimation drift or divergence caused by low data signal-to-noise ratio, ensuring the reliability of parameter updates and the economy of computational resources; while when the system state changes drastically and the stimulus is sufficient, it can be woken up in time to provide high-precision model parameter updates. Finally, the synergy of the two constitutes a "fast and slow combination" and "coarse and fine complementarity" estimation system: the extended state observer is responsible for "fast" response and "coarse" compensation, handling high-frequency, unmodeled disturbances; the event-triggered parameter identifier is responsible for "slow" updates and "fine" modeling, correcting low-frequency, time-varying parameters. This synergistic effect enables the feedforward-feedback control unit 23 to obtain a real-time and accurate disturbance compensation amount to deal with sudden situations, and to obtain a dynamic model that gradually approaches the real system for precise feedforward control. As a result, the dynamic stability and steady-state tracking accuracy are improved simultaneously, making the assisted robotic arm both agile and precise in responding to load changes and human-machine interaction forces.

[0026] In some embodiments, the control core module 2 further includes a performance evaluation and switching logic unit, the input of which is connected to the output of the extended state observer and the event trigger parameter identifier, and the output of which is connected to the feedforward-feedback control unit 23. The performance evaluation and switching logic unit is configured to determine whether the system is in a stable or transient state based on the characteristics of the trajectory tracking error, and accordingly dynamically adjust the fusion weight of the disturbance compensation amount received from the dual state observation unit 22 and the feedforward control amount based on the parameter update amount in the final control signal of the feedforward-feedback control unit 23.

[0027] In this embodiment, the performance evaluation and switching logic unit added to the control core module 2 has the core function of intelligently coordinating the contribution ratio of the two core outputs from the dual-state observation unit 22—namely, the fast disturbance compensation amount provided by the extended state observer and the feedforward control amount generated by the model parameters updated by the event-triggered parameter identifier—in the final control signal of the feedforward-feedback control unit 23. The practical application of this unit relies primarily on the accurate evaluation of the system's real-time operating state. A common implementation method is for this unit to continuously monitor the trajectory tracking error generated by the feedforward-feedback control unit 23 and calculate its short-term statistical characteristics, such as the root mean square value of the error, its rate of change (derivative), or its spectral energy distribution. For example, when the system operates smoothly and both the tracking error and its rate of change remain at a low level, the performance evaluation logic can determine that the system is in a "stable operating condition." At this time, it is considered that the model parameters updated based on the event-triggered parameter identifier have high reliability, and the system model is relatively accurate. In this scenario, the unit dynamically increases the fusion weight of the feedforward control quantity based on accurate model parameters, while correspondingly decreasing the weight of the generalized disturbance compensation quantity. This makes the control strategy more focused on "accurate model-driven" to pursue optimal tracking accuracy and energy efficiency. Conversely, when the system detects a sharp increase in tracking error or high-frequency jitter in a short period, such as when a robotic arm suddenly bears an unknown weight or experiences an unexpected impact from the user, the performance evaluation logic immediately determines that the system has entered a "transient condition." At this time, the model may momentarily become inaccurate, and the estimation of the "total disturbance" by the extended state observer becomes more critical. Therefore, the unit quickly switches the weight allocation, significantly increasing the weight of the rapid disturbance compensation quantity, while potentially temporarily reducing the dependence on feedforward quantities based on model parameters that may have lag or bias. This causes the control strategy to quickly shift to a "strong robust disturbance rejection" mode, prioritizing the transient stability of the system.

[0028] The benefits of this design are significant and multi-layered. Its fundamental value lies in its creative solution to the challenge of effectively integrating two different types of control variables—"precise model feedforward" and "rapid disturbance compensation"—in dynamic processes. If fixed-weight fusion is used, accuracy will be lost during stable periods due to over-reliance on potentially noisy disturbance observations, while control will be ineffective during abrupt changes due to untimely model updates. The intelligent switching mechanism in this embodiment allows the system to cruise economically on straight roads (stable conditions) using a precise map (precise model), like an experienced driver, while quickly tightening control (enhanced feedback and disturbance compensation) to ensure safety during sudden bumps or curves (transient conditions). Specifically, under stable conditions, the system can fully utilize high-precision parameter identification results to achieve near-zero steady-state error high-precision tracking through high-weighted model feedforward, while suppressing high-frequency noise that may be introduced by over-reliance on disturbance observations. Under transient or strong disturbance conditions, the system can quickly suppress deviations and maintain stability by relying on high-weight disturbance compensation, thus gaining a time window for the parameter identifier to reconverge and avoiding the risk of runaway due to model mismatch. This condition-based adaptive weight allocation enables the overall control system to have both high-precision steady-state performance and robust dynamic response, realizing automatic optimization of control strategies under different operating scenarios. This comprehensively improves the intelligence, adaptability, and overall control performance of the assistive robotic arm when facing complex and variable tasks and human-machine interaction.

[0029] In some embodiments, the dual-state observation unit 22 further includes a coordination module configured to perform the following operations: Receive the lumped disturbance estimate output from the extended state observer; The part of the lumped disturbance estimate that is linearly related to the model parameters is analyzed as an incremental correction term for the parameters to be identified in the event-triggered parameter identifier; When the event-triggered parameter identifier is triggered, the incremental correction term is used as the initial deviation input for parameter identification or as a weighting term in the identification process.

[0030] In this embodiment, the collaborative module added to the dual-state observation unit 22 is key to achieving deep collaboration between the extended state observer and the event-triggered parameter identifier, thereby improving the efficiency and accuracy of the entire estimation system. The core of its technical implementation lies in establishing an effective correlation between the "total disturbance" and the "model parameter error." A specific implementation method involves the collaborative module pre-setting or learning an "disturbance-parameter" mapping matrix internally. This module receives the lumped disturbance estimate output in real time by the extended state observer. This value essentially includes the equivalent force or torque generated by the combined effects of inaccurate model parameters, unmodeled dynamics, and external disturbances. The collaborative module uses built-in analytical rules (e.g., Jacobian matrix based on the system's nominal model or sensitivity analysis) to separate components from the total disturbance estimate that have an approximately linear relationship with the parameters to be identified (such as load mass and friction coefficient). For example, in the joint dynamics of a assisted robotic arm, disturbances related to load mass are typically manifested as inertial force components proportional to acceleration, while disturbances related to friction are related to the sign of velocity. The collaborative module uses algorithms (such as bandpass filtering and projection calculation) to parse these feature components and quantify them into "incremental correction terms" for specific parameters. These correction terms indicate the direction and approximate magnitude by which the current parameter estimate may deviate from the true value. When the event-triggered parameter identifyer is activated due to sufficient system excitation, the collaborative module does not simply transmit the raw data. Instead, it injects this parsed incremental correction term as high-value prior information into the identification process. Specifically, it can be used as the initial iterative value of the parameter to be estimated in identification algorithms such as recursive least squares, allowing the iteration to start from a point closer to the true solution; or it can be transformed into weighting coefficients of the observation equations in the identification algorithm, giving higher confidence to equations related to these correction terms during solution, thereby guiding the entire identification process to converge to the true parameter value more quickly and accurately.

[0031] The design of this collaborative module brings significant and direct benefits, fundamentally changing the traditional mode of two observers working independently and constructing a closed loop of "observation-identification" collaborative enhancement. Without this module, the extended state observer only performs passive compensation, and its rich perturbation information is discarded after compensation; the event-triggered parameter identifier needs to reconverge "from zero" after triggering, and its response speed is limited by the algorithm's convergence time. The introduction of the collaborative module enables the extended state observer to play the role of a "foreign sentinel" and "information preprocessor," and its rapidly estimated perturbations contain immediate clues about parameter errors. By parsing these clues and transforming them into guiding information input to the parameter identifier, the "cold start" problem of the identification process is greatly optimized. Its effects are reflected in two aspects: First, it accelerates parameter convergence. The identifier no longer needs to start from broad initial values ​​or the previous historical values ​​for lengthy iterations, but instead starts searching from a more accurate neighborhood "hinted" by disturbance information, significantly shortening the time required to converge to the true parameters, enabling the updated model to serve high-precision feedforward control more quickly. Second, it improves the robustness and accuracy of parameter identification. Especially when the excitation signal is not completely continuous and ideal, the incremental correction term from the extended state observer, as an additional constraint derived from the real-time response of the system, can help the identification algorithm effectively resist the interference of measurement noise and avoid converging to an incorrect local optimum, thereby ensuring the reliability of the updated parameters. Ultimately, this collaborative mechanism enables the entire dual-state observation unit 22 to "understand" and adapt to changes in the internal parameters and external environment of the system more quickly and accurately, providing the feedforward-feedback control unit 23 with more timely and accurate disturbance compensation and model updates, which together constitute the intelligent perception and decision-making cornerstone that helps the robotic arm achieve high stability and high precision tracking in dynamic and complex tasks.

[0032] In some embodiments, the multi-rate state observation unit 21 includes a signal fusion module and a delay prediction compensation module; The signal fusion module is configured to receive low-frequency displacement sampling signals from the displacement sensor and high-frequency pressure signals from the pressure sensor, and to use the high-frequency pressure signals and system dynamics to calculate a high-frequency velocity estimation signal based on the differential of the pressure difference between the two chambers of the cylinder in real time. The delay prediction and compensation module, coupled to the signal fusion module, is configured to establish a system state-space model that includes known communication delays. It uses low-frequency displacement signals as absolute position references and high-frequency velocity estimation signals as high-frequency change quantities. It performs multi-rate data fusion through the Kalman filter algorithm and performs one-step forward prediction on the known communication delays. Finally, it outputs the compensated high-frequency displacement and velocity estimates as system state estimates.

[0033] In this embodiment, the multi-rate state observation unit 21 is implemented through the collaborative work of the signal fusion module and the delay prediction compensation module. The implementation of the signal fusion module first requires access to the raw data streams from physical sensors: one path is a displacement sampling signal from displacement sensors such as joint encoders, uploaded via a fieldbus (e.g., CAN bus) at a fixed low-frequency period (e.g., 1kHz). This signal has high absolute accuracy but suffers from a fixed communication delay in the millisecond range and has a limited update frequency. The other path is a high-frequency pressure signal (e.g., 10kHz) directly obtained from cylinder two-chamber pressure sensors (e.g., strain gauge sensors) in analog form or through high-speed AD conversion. This signal has extremely low delay but cannot directly reflect the joint position. The signal fusion module utilizes the dynamic relationship of the pneumatic system, particularly the physical connection between the pressure difference between the cylinder two chambers and the output force. It estimates the force by calculating the pressure difference in real time and combining it with parameters such as the cylinder piston area. Then, based on a simplified dynamic model of the system (e.g., ignoring some nonlinearities), it calculates a high-frequency joint velocity estimation signal in real time. This signal can be called the "pressure differential velocity." The delay prediction compensation module is the core of solving the state synchronization and delay problem; it is coupled after the signal fusion module. Internally, this module establishes a discrete state-space model of the system incorporating a known bus communication delay (e.g., two sampling periods). It then runs a specially designed multi-rate Kalman filter. This filter uses a low-frequency but absolutely accurate displacement sampling signal as an "anchor" for periodic correction to rectify accumulated errors. Simultaneously, within the interval between the arrival of two displacement signals, the aforementioned high-frequency "pressure differential velocity" signal serves as the primary state update, continuously driving state prediction. More importantly, the filter model explicitly includes the delay as part of the state transition, enabling "one-step forward prediction" of a known fixed delay to estimate the true system state at the current moment (rather than at the delayed moment). Finally, the module outputs delay-compensated estimates of joint displacement and velocity at the same frequency as the high-speed pressure signal, which together constitute a high-quality system state estimate.

[0034] The benefits of this design are fundamental, addressing the bottleneck restricting the performance of advanced control algorithms at the information source. Traditional solutions rely on severely lagging displacement signals or unreliable calculated speeds, much like a driver looking in the rearview mirror – inefficient and unrealistic. This embodiment creatively transforms the dynamic information contained in high-frequency pressure signals into continuous velocity observations through signal fusion, effectively filling the gap in low-frequency displacement signal updates and significantly increasing the frequency of state updates, enabling the control loop to respond to faster dynamic changes. The delay prediction compensation module cleverly utilizes the system model to align the lagging absolute position information with the leading dynamic information on the timeline, effectively eliminating the phase lag introduced by fixed communication delays. The combination of these two elements is equivalent to equipping the control system with a "high refresh rate, low latency virtual motion sensor." The direct effect is to provide near real-time, accurate feedback for subsequent dual-state observation and feedforward-feedback control, allowing the extended state observer to detect disturbances earlier, the event-triggered parameter identifier to use more accurate data, and the feedforward-feedback controller to calculate corrections more promptly. This expands the effective bandwidth of the entire control system as a whole, significantly reduces the risk of control overshoot and instability caused by perception delay, and lays an indispensable perception foundation for helping the robotic arm achieve high-speed, high-precision dynamic stable tracking.

[0035] In some embodiments, a sensor group is also included, which includes a displacement sensor, a first pressure sensor, and a second pressure sensor. The displacement sensor is installed at the joint and configured to measure the actual displacement of the joint. The output signal of the displacement sensor is periodically uploaded via the fieldbus at a first frequency. The first pressure sensor and the second pressure sensor are respectively installed in the rod chamber and the rodless chamber of the cylinder and are configured to measure the pressure of the two chambers in real time. The output signals of the first pressure sensor and the second pressure sensor are directly acquired and input to the multi-rate state observation unit 21 at a second frequency higher than the first frequency.

[0036] In this embodiment, the system includes a specially configured sensor group, specifically consisting of a displacement sensor mounted at the joint, and a first pressure sensor and a second pressure sensor mounted respectively in the rod-side and rodless-side chambers of the cylinder. The configuration and data path design of this sensor group form the physical basis for achieving multi-rate fusion observation in this scheme. The displacement sensor (such as a high-precision absolute encoder) is responsible for directly measuring the actual rotational or linear displacement of the joint, and its output signal is typically transmitted via a digital fieldbus (such as CAN or EtherCAT). While this bus transmission method ensures signal anti-interference and absolute accuracy, it inevitably introduces fixed message packaging, transmission, and decoding times, resulting in its signal being periodically uploaded to the controller at a relatively low "first frequency" (e.g., 1 kHz), and exhibiting a fixed, calibrable communication delay of several milliseconds. In stark contrast, the signal paths of the two pressure sensors are designed in a "direct acquisition" mode. The output signals of the first and second pressure sensors (typically fast-response analog or digital pressure transmitters) are acquired in real time with low latency via the analog input port of the controller or a dedicated high-speed synchronous sampling module, without passing through a complex bus protocol stack. This acquisition occurs at a "second frequency" (e.g., 10 kHz) much higher than the displacement signal frequency. This heterogeneous sensing architecture—a low-frequency, high-precision but delayed absolute position signal and two high-frequency, low-latency relative pressure signals—is synchronously input to the aforementioned multi-rate state observation unit 21, providing it with the necessary raw information sources with different spatiotemporal characteristics for data fusion.

[0037] The benefits of this sensor configuration are multi-layered and crucial. First, it rationalizes and supports the "multi-rate fusion" software algorithm at the hardware level. Without this specially designed combination of high and low frequency signals with varying delays, the multi-rate state observation unit 21 would be unable to perform its function, failing to achieve delay prediction and high-frequency compensation. The direct effect is that the system does not need to invest in ultra-high bandwidth, zero-delay displacement sensors (such as high-speed analog encoders, which are expensive and susceptible to noise interference). Instead, it achieves equivalent or even better state perception performance economically through a combination of mature industrial bus sensors and ordinary pressure sensors, supplemented by advanced algorithms, significantly reducing the system's hardware cost and complexity. Second, this clearly defined sensing strategy fully leverages the advantages of various sensors: displacement sensors provide a long-term stable, error-free absolute position reference, ensuring that the state estimation does not drift; while high-frequency pressure signals provide rich information characterizing the system's instantaneous dynamics, enabling the state estimation to keep up with rapid motion changes. The combination of these two factors results in the final system state estimate possessing both high absolute accuracy and fast dynamic response. This fundamentally solves the inherent problem in traditional solutions where the use of a single hysteresis displacement feedback leads to a decrease in the phase margin of the control system and a tendency to cause oscillations. It lays a reliable and timely sensing foundation for the stable and high-performance operation of the entire high-order control loop (including disturbance observation, parameter identification, and precise control), thereby improving the overall dynamic control quality of the assisted robotic arm.

[0038] In some embodiments, when the event-triggered parameter identifier performs recursive identification, the system model regression vector used includes a real-time pressure difference term and its real-time differential term calculated based on the cylinder two-chamber pressure signal, which are used to identify the joint equivalent friction coefficient and load mass.

[0039] In this embodiment, a key technological achievement of the event-triggered parameter identifier lies in its unique construction of the system model regression vector. When the identifier is triggered, it does not rely solely on traditional kinematic quantities such as joint angles, velocities, and accelerations to construct the mathematical model (i.e., the regression vector) for parameter estimation. Its core innovation lies in the inclusion of physical quantities specifically calculated in real-time from the signals of the first and second pressure sensors: one is the real-time pressure difference (ΔP) between the two chambers of the cylinder, and the other is the real-time differential term of this pressure difference (d(ΔP) / dt). Specifically, after the controller acquires the two pressure signals at high speed, it first performs synchronous alignment and filtering, then calculates the difference in real-time; subsequently, it estimates the rate of change of the pressure difference online using numerical differentiation methods (such as first-order backward difference) or observer techniques. These direct force signals originating from the pneumatic drive body, together with traditional acceleration signals obtained based on position differentiation, constitute a richer and more physically meaningful regression vector. This enhanced vector is fed into online estimation algorithms such as recursive least squares with a forgetting factor, specifically designed to decouple and identify key time-varying parameters such as the joint equivalent viscous friction coefficient and load mass.

[0040] This design, which deeply integrates native aerodynamic signals into the regression model, brings significant and direct benefits, fundamentally improving the accuracy and reliability of parameter identification in nonlinear, strongly coupled aerodynamic systems. Traditional methods rely solely on acceleration obtained from the second derivative of the displacement signal. At low speeds or near reversal points, this acceleration signal has an extremely low signal-to-noise ratio and is easily amplified by noise, leading to large fluctuations and inaccuracies in the identification results of inertial parameters (such as load mass). The pressure difference between the two chambers of the cylinder is directly related to the system output force, and its differential term better reflects the trend of force changes. These provide high-confidence information about the forces acting on the system, independent of the kinematic chain. Introducing these two strongly correlated force signals into the regression is equivalent to adding an "internal perspective" and "physical constraints" to the identification algorithm. The effect is twofold: First, it greatly improves the accuracy and speed of load mass identification because the pressure difference and its changes directly reflect the force required to overcome inertia, enabling the algorithm to effectively distinguish inertial forces from other force components even at small accelerations. Secondly, it significantly improves the ability to identify friction coefficients, especially velocity-dependent friction, because pressure signals help to isolate complex frictional forces (such as Coulomb friction and static friction) that dominate in the low-speed region. This more accurate and robust online parameter identification allows the system dynamics model to more realistically reflect the current physical reality, enabling the feedforward control unit, which relies on this model, to generate more precise compensating forces. Ultimately, this results in the robotic arm maintaining excellent trajectory tracking performance and dynamic stability even under load changes and low-speed precision operations.

[0041] In some embodiments, the main control module 1 includes a trajectory planning unit and an inverse kinematics unit connected in sequence. The trajectory planning unit is configured to generate an end-effector desired pose sequence according to the task instructions, and the inverse kinematics unit is configured to solve the end-effector desired pose sequence into desired angles, desired angular velocities and desired angular accelerations of each joint.

[0042] In this embodiment, the main control module 1 is specifically constructed through a trajectory planning unit and a kinematics inverse unit connected in sequence. The trajectory planning unit's technical implementation begins with receiving high-level task instructions from upper-level applications or operators, such as "smoothly move the end effector from point A to point B." Its core function is to generate a continuous, smooth, and physically realizable sequence of desired end effector poses (position and attitude) within the operating space (i.e., Cartesian space) based on specific task requirements and the system's dynamic capabilities. A common implementation method is to use polynomial (e.g., fifth-order polynomial) or spline curve interpolation algorithms to construct a path between a given starting point and target point, ensuring that the planned trajectory is continuous in terms of position, velocity, and even acceleration. This avoids step-like instructions impacting the subsequent pneumatic servo system. Specifically, considering the collaborative work between the assisted robotic arm and the human body, safety must be considered during planning, such as applying soft constraints to the maximum end effector velocity and acceleration. The kinematics inverse unit follows immediately, responsible for mapping from the operating space to the joint space. It receives the end-effector pose, arranged in a time sequence, from the trajectory planning unit, and based on the geometric model and link parameters of the assisted robotic arm, calculates the reference angles of each joint required to drive the end-effector pose in real time and accurately using numerical methods (such as the Newton-Raphson iterative method) or analytical solutions. Furthermore, by numerically differentiating the joint angle sequence, the desired angular velocity and desired angular acceleration of each joint can be obtained synchronously, thus forming a complete time sequence describing the desired motion state.

[0043] The meticulous design of this module brings fundamental and crucial benefits to the entire control system. The existence of the trajectory planning unit ensures that the original commands received by the system are fully "tamed," providing a smooth, abrupt "ideal track" for the nonlinear aerodynamic system with hysteresis. This avoids high-frequency jitter and instability directly caused by discontinuous commands (such as step changes), which is a prerequisite for achieving smooth, anthropomorphic motion. The core role of the inverse kinematics unit is to provide a precise "mapping benchmark." It translates the high-level end-point task objective into the physical quantities (angle, angular velocity, angular acceleration) that the aerodynamic joints need to directly track at the lower level without any errors. This set of calculation results constitutes the precise meaning of the "desired motion trajectory" in the feedforward-feedback control unit 23. Its accuracy directly determines the effectiveness of the model feedforward compensation, because the feedforward control quantity is calculated based on this desired trajectory and the dynamic model. At the same time, it also constitutes the "gold standard" for comparison with the actual state fed back by the multi-rate state observation unit 21, and is the basis for generating the feedback control quantity deviation signal. Therefore, an accurate, real-time, and continuous solution provides a unified and reliable tracking target for all subsequent advanced observation, identification, and control algorithms. It is the fundamental reference system for the entire control loop to achieve high-precision servo tracking, enabling the end effector of the robotic arm to strictly follow the human's intention to complete the predetermined spatial trajectory.

[0044] In some embodiments, the feedforward-feedback control unit 23 includes a feedforward channel and a feedback channel; The feedforward channel is configured to calculate the feedforward force / torque based on the desired motion trajectory and the parameter update provided by the dual-state observation unit 22, and use it as the feedforward control quantity through the inverse dynamics model. The feedback channel is configured to calculate the feedback force / torque as the feedback control quantity based on the deviation between the desired motion trajectory and the system state estimate provided by the multi-rate state observation unit 21 using a proportional-derivative control law. The feedforward-feedback control unit 23 also includes a fusion module, which is configured to add the feedforward force / torque, the feedback force / torque and the disturbance compensation amount to generate the final control force / torque command, and output it as the desired pressure control signal for each cylinder via the force / torque-pressure conversion module, as the control signal.

[0045] In this embodiment, the feedforward-feedback control unit 23 consists of a feedforward channel, a feedback channel, and a fusion module. Its technical implementation begins with a clear division of labor: the core task of the feedforward channel is to perform calculations based on the desired motion trajectory (provided by the main control module 1, including the desired angle, angular velocity, and angular acceleration) and the model parameters updated in real-time by the dual-state observation unit 22 (such as the latest load mass and friction coefficient), using the system's inverse dynamics model. This calculation process essentially "predicts" the theoretically required force or torque for each joint to accurately track the ideal trajectory. For example, when the desired trajectory requires rapid joint acceleration, the inverse dynamics model calculates the feedforward torque required to overcome this inertia based on the updated load parameters; simultaneously, it calculates the feedforward torque to compensate for the expected friction force based on the updated friction model. These calculation results are summarized into feedforward control quantities, the purpose of which is to proactively and forward-lookingly offset most of the known, modelable system dynamics influences. Meanwhile, the feedback channel works in parallel, handling all residual errors not fully compensated by the feedforward and unknown disturbances. It continuously receives high-frequency, low-delay system state estimates (such as real-time joint angles and velocities) from the multi-rate state observation unit 21 and compares them in real time with the desired trajectory provided by the main control module 1 to calculate the position and velocity deviations. Subsequently, a classic but parameter-adjustable proportional-derivative (PD) control law is used to calculate the corresponding feedback force / torque based on this deviation. This feedback control quantity is reactive, and its role is like a "stabilizing hand," continuously pulling the actual state of the system back to the desired trajectory to ensure basic closed-loop stability. The final fusion module is the convergence point of control decisions. It performs a key but intuitive addition operation: algebraically adding the predictive torque provided by the feedforward channel, the corrective torque provided by the feedback channel, and the real-time disturbance compensation provided by the dual state observation unit 22, which is specifically used to offset unmodeled dynamics and external disturbances. The summed "total control force / torque command" is then passed through a force / torque-pressure conversion module (which linearly converts the force command into the desired pressure difference signal between the two chambers of the cylinder based on geometric parameters such as the cylinder piston area, and allocates it to the specific desired pressure value of the two chambers in combination with the current pressure state) to generate the control signal that ultimately drives the proportional pressure valve.

[0046] This clearly structured and collaborative design yields distinct and significant benefits. Its core advantage lies in integrating the strengths of three control concepts: "model prediction feedforward," "error feedback adjustment," and "disturbance feedforward compensation," while mitigating their weaknesses. The feedforward channel utilizes a dynamically updated, accurate model to handle the majority of the "planned" driving tasks, greatly reducing the burden on the feedback channel. This allows the system to achieve near-zero steady-state error and rapid response when tracking a well-planned trajectory, as it essentially reproduces ideal dynamics in an "open-loop" manner, thus significantly reducing tracking errors. The feedback channel focuses on handling residual minor deviations and random disturbances "unplanned," and its PD control ensures the system's basic stability and robustness in the face of any minor uncertainties. The fusion module directly injects disturbance compensation, essentially adding a dedicated rapid-response channel to combat the "unknown" and "sudden changes." This further enhances the system's dynamic stability when dealing with sudden load changes or external shocks, avoiding the oscillations that might result from having to correct these disturbances entirely through a relatively slow feedback loop. Ultimately, through the aggregation and transformation by the fusion module, the system outputs a comprehensive and fully compensated precise pneumatic pressure command, enabling the pneumatic actuator module 3 to obtain just the right driving force. Overall, this unit allows the control system to perform efficient and stable feedforward planning based on a precise map (updating the model), much like autonomous driving, and to make fine adjustments based on real-time road conditions (feedback errors), much like a human driver. It also possesses active suspension compensation capabilities for sudden bumps (lumped disturbances), thus collaboratively ensuring optimal comprehensive control performance of the assisted robotic arm under various working conditions.

[0047] In some embodiments, the pneumatic actuator 3 includes a proportional pressure valve and a cylinder. The electrical interface of the proportional pressure valve is connected to the output of the control core module 2, and the air interface of the proportional pressure valve is connected to the rod chamber and the rodless chamber of the cylinder, and is configured to adjust the pressure of the two chambers of the cylinder according to the control signal.

[0048] In this embodiment, the pneumatic actuator module 3, as the final physical implementer of control commands, is crucial for achieving high dynamic response through its specific structure and working mechanism. This module mainly consists of a proportional pressure valve and a cylinder, connected via a clearly defined interface. The electrical interface of the proportional pressure valve is directly connected to the output of the control core module 2, receiving the precisely calculated desired pressure control signal (usually a voltage or current signal) generated by the feedforward-feedback control unit 23. Its pneumatic interface is connected to the rod-side and rodless-side chambers of the cylinder via pipelines, and then connected to a compressed air source. Technically, the proportional pressure valve typically contains a precision electromagnet and a valve core assembly. When the electrical control signal changes, the magnetic force generated by the electromagnet drives the valve core to produce a displacement proportional to the signal magnitude, thereby continuously and linearly adjusting the intake flow or pressure from the air source to the two chambers of the cylinder, while simultaneously controlling the exhaust from the cylinder chamber. For example, when a control signal instructs the cylinder thrust to increase, the valve core actuates, allowing more high-pressure gas to enter the rodless chamber, while potentially releasing some gas from the rod chamber in a controlled manner, thus establishing the required pressure difference between the two chambers. The cylinder, acting as an actuator, generates a linear force or torque under the direct action of this pressure difference, pushing the connecting rod, which is ultimately converted into the rotational or linear motion of the robotic arm joint. The entire pneumatic circuit typically also includes necessary filters, pressure reducing valves, and quick-release valves to ensure the cleanliness of the air source, stable pressure, and rapid pressure relief in emergencies, but the core drive and precise pressure regulation functions are handled by a proportional pressure valve.

[0049] This modular execution scheme, with a proportional pressure valve at its core, directly drives the cylinder, bringing direct and efficient benefits. First, it achieves extremely high dynamic response speed and force control precision. The proportional pressure valve can respond to electrical signals continuously at the millisecond level, allowing complex force commands calculated based on advanced algorithms, including feedforward, feedback, and disturbance compensation, to be converted into pneumatic commands and applied to the cylinder piston with almost no distortion. This ensures that the control core's "intelligent decision-making" can be quickly and accurately translated into "physical action," eliminating control performance bottlenecks caused by slow actuator response and providing the hardware foundation for the system's high dynamic stability. Second, through independent and precise coordinated control of the pressure in both chambers of the cylinder, the system can achieve smooth start-stop, fine force adjustment, and effective damping control, which is crucial for assistive robotic arms that require human-assisted operation and extremely high motion compliance. Finally, the module's structure is relatively simple, reliable, and has a high power density. The overload resistance of the pneumatic components themselves provides a natural safety buffer for the entire system, enabling the assisted robotic arm to output powerful assistance while maintaining the system's durability and safety when performing tasks such as material handling. Therefore, this efficient and precise execution terminal, combined with advanced sensing, estimation, and control algorithms, constitutes a complete high-performance control system that seamlessly connects "decision-making" to "execution."

[0050] It should be noted that the order of the above embodiments of the present invention is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, the above description focuses on specific embodiments of this specification. Additionally, the processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some embodiments, multitasking and parallel processing are possible or may be advantageous.

[0051] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

[0052] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A dynamic stability control system for a pneumatic system of a power-assisted robotic arm, characterized in that, It includes a main control module (1), a control core module (2), and a pneumatic actuator module (3): The main control module (1) is configured to generate the desired motion trajectory of each joint of the robotic arm; The control core module (2) is connected to the main control module (1) and is used to receive the desired motion trajectory and generate control signals; the control core module (2) includes a multi-rate state observation unit (21), a dual state observation unit (22) and a feedforward-feedback control unit (23); The multi-rate state observation unit (21) is configured to receive displacement and pressure signals based on each joint of the robotic arm with different sampling frequencies and communication delays, perform data fusion and prediction, and generate high-frequency, low-latency system state estimates. The dual-state observation unit (22) is connected to the multi-rate state observation unit (21) and is configured to receive the system state estimate, estimate the system model parameters and lumped disturbances online, and generate parameter update amount and disturbance compensation amount. The feedforward-feedback control unit (23) is connected to the main control module (1), the multi-rate state observation unit (21), and the dual state observation unit (22) respectively. It is configured to receive the desired motion trajectory, the system state estimate, the parameter update amount, and the disturbance compensation amount, and calculate the feedforward control amount based on the desired motion trajectory, the system state estimate, and the parameter update amount. At the same time, it calculates the feedback control amount based on the deviation between the desired motion trajectory and the system state estimate. Finally, it fuses the feedforward control amount, the feedback control amount, and the disturbance compensation amount to generate the control signal. The pneumatic actuator module (3) is connected to the control core module (2) and is configured to drive the joint movement of the robotic arm according to the control signal.

2. The system according to claim 1, characterized in that, The dual-state observation unit (22) includes an extended state observer and an event trigger parameter identifier; The extended state observer is configured to uniformly extend the system model uncertainty and external disturbances into new state variables and observe them in real time, and output the estimated value of the lumped disturbance as the disturbance compensation amount. The event-triggered parameter identifier is configured to monitor the joint motion acceleration or load change determined based on the system state estimate. When the joint motion acceleration or load change exceeds a preset threshold, it is triggered and, based on the system input and output data before and after the triggering time, it recursively identifies key time-varying parameters in the system dynamics model online and outputs the parameter update amount.

3. The system according to claim 2, characterized in that, The control core module (2) further includes a performance evaluation and switching logic unit. The input of the performance evaluation and switching logic unit is connected to the output of the extended state observer and the event trigger parameter identifier. The output of the performance evaluation and switching logic unit is connected to the feedforward-feedback control unit (23). The performance evaluation and switching logic unit is configured to determine whether the system is in a stable or transient state based on the characteristics of the trajectory tracking error, and accordingly dynamically adjust the fusion weight of the disturbance compensation amount received from the dual state observation unit (22) and the feedforward control amount based on the parameter update amount in the final control signal of the feedforward-feedback control unit (23).

4. The system according to claim 3, characterized in that, The dual-state observation unit (22) further includes a coordination module configured to perform the following operations: Receive the lumped disturbance estimate output by the extended state observer; The portion of the lumped disturbance estimate that is linearly related to the model parameters is analyzed as an incremental correction term for the parameter to be identified in the event trigger parameter identifier; When the event-triggered parameter identifier is triggered, the incremental correction term is used as the initial deviation input for parameter identification or as a weighting term in the identification process.

5. The system according to claim 4, characterized in that, The multi-rate state observation unit (21) includes a signal fusion module and a delay prediction compensation module; The signal fusion module is configured to receive a low-frequency displacement sampling signal from a displacement sensor and a high-frequency pressure signal from a pressure sensor, and to use the high-frequency pressure signal and the system dynamics relationship to calculate a high-frequency velocity estimation signal based on the differential of the pressure difference between the two chambers of the cylinder in real time. The delay prediction and compensation module, coupled to the signal fusion module, is configured to establish a system state space model containing known communication delays. Using the low-frequency displacement signal as the absolute position reference and the high-frequency velocity estimation signal as the high-frequency change quantity, it performs multi-rate data fusion through the Kalman filter algorithm and performs one-step forward prediction on the known communication delay. Finally, it outputs the compensated high-frequency displacement and velocity estimates as the system state estimates.

6. The system according to claim 5, characterized in that, It also includes a sensor group, which includes a displacement sensor, a first pressure sensor, and a second pressure sensor; The displacement sensor is installed at the joint and configured to measure the actual displacement of the joint. The output signal of the displacement sensor is periodically uploaded via a fieldbus at a first frequency. The first pressure sensor and the second pressure sensor are respectively installed in the rod chamber and the rodless chamber of the cylinder and are configured to measure the pressure of the two chambers in real time. The output signals of the first pressure sensor and the second pressure sensor are directly acquired and input to the multi-rate state observation unit (21) at a second frequency higher than the first frequency.

7. The system according to claim 6, characterized in that, When the event trigger parameter identifier performs recursive identification, the system model regression vector used includes a real-time pressure difference term and its real-time differential term calculated based on the pressure signals of the two chambers of the cylinder, which are used to identify the joint equivalent friction coefficient and load mass.

8. The system according to claim 7, characterized in that, The main control module (1) includes a trajectory planning unit and an inverse kinematics unit connected in sequence. The trajectory planning unit is configured to generate an end-effector desired pose sequence according to the task instructions. The inverse kinematics unit is configured to solve the end-effector desired pose sequence into the desired angle, desired angular velocity and desired angular acceleration of each joint.

9. The system according to claim 8, characterized in that, The feedforward-feedback control unit (23) includes a feedforward channel and a feedback channel; The feedforward channel is configured to calculate the feedforward force / torque through an inverse dynamics model based on the desired motion trajectory and the parameter update provided by the dual-state observation unit (22), and use it as the feedforward control quantity. The feedback channel is configured to calculate the feedback force / torque using a proportional-derivative control law based on the deviation between the desired motion trajectory and the system state estimate provided by the multi-rate state observation unit (21), and use it as the feedback control quantity. The feedforward-feedback control unit (23) further includes a fusion module, which is configured to add the feedforward force / torque, the feedback force / torque and the disturbance compensation amount to generate the final control force / torque command, and output it as the desired pressure control signal for each cylinder via the force / torque-pressure conversion module, as the control signal.

10. The system according to claim 9, characterized in that, The pneumatic actuator module (3) includes a proportional pressure valve and a cylinder. The electrical interface of the proportional pressure valve is connected to the output end of the control core module (2). The air circuit interface of the proportional pressure valve is connected to the rod chamber and rodless chamber of the cylinder and is configured to adjust the pressure of the two chambers of the cylinder according to the control signal.

Images