Upper limb exoskeleton system cooperative follow-up control method based on active disturbance rejection control strategy

Abstract

The invention discloses an upper limb exoskeleton system cooperative follow-up control method based on an active-disturbance-rejection control strategy, and relates to the technical field of man-machine cooperative control, and the method comprises the following steps: S1, collecting a contact force change between a user and a target object, a tail end weight change and a holding slip signal, fusing to generate load sudden unloading risk early warning information, and storing the information in a database; setting an initial slope parameter of the driving compensation curve; and S2, constructing a load shedding control track based on the early warning information, inputting the load shedding control track to an active-disturbance-rejection control system, dynamically adjusting an original control target, and constraining an acceleration step amplitude. According to the method, multi-source signals are fused to generate unloading early warning, a load shedding track is constructed, driving output is adjusted, mode locking and viscoelastic dissipation are combined to achieve impact absorption, elbow-shoulder cooperative braking and multi-scale modulation are matched to suppress energy transfer, intention control is switched through myoelectricity judgment, and the track is corrected; and the safety, stability and comfort of the exoskeleton in agricultural carrying are improved.

Description

Technical Field

[0001] This invention relates to the field of human-machine collaborative control technology, and more specifically to a collaborative follow-up control method for upper limb exoskeleton systems based on an active disturbance rejection control strategy. Background Technology

[0002] "Collaborative Follow-up Control of Upper Limb Exoskeleton System Based on Active Disturbance Rejection Control Strategy" refers to the application of Active Disturbance Rejection Control (ADRC) disturbance estimation and compensation capabilities to exoskeleton drive control in high-intensity agricultural work scenarios to improve the stability and comfort of the wearer's upper limb operations. This involves using an extended state observer to identify the user's joint movement trends and force states in real time, dynamically correcting the drive output to mitigate the uncertainties caused by uneven terrain, changes in work posture, and load fluctuations. Simultaneously, a collaborative follow-up mechanism is introduced to ensure that the movement trajectory of the exoskeleton joints is highly consistent with the wearer's muscle strength and movement intentions, achieving smooth and natural human-machine interaction. This provides precise auxiliary force to the user during repetitive tasks such as fruit and vegetable harvesting, pruning, and sorting, reducing muscle fatigue and the risk of misoperation, while maintaining stable output during complex movements such as walking, squatting, and turning, thus ensuring efficiency, safety, and comfort during long-term field operations.

[0003] The existing technology has the following shortcomings: In existing technologies, the motion control of exoskeleton systems in agricultural handling operations mostly relies on mechanical compensation mechanisms based on preset models to maintain the stability of joint output. However, in dynamic scenarios with heavy loads, such as handling fruit baskets, if a sudden unloading occurs, such as the basket accidentally slipping or the user unexpectedly and quickly putting it down, the active disturbance rejection control (ADRC) technology in existing technologies often cannot suppress the inertial output in a very short time. Due to certain computational delays and execution lags in the control process, the drive end will continue to output a large auxiliary torque according to the original load state, resulting in significant overshoot of the joint actuator in the absence of actual load feedback. This overshoot not only manifests as exceeding the limits of the motion trajectory but also forms a reverse acceleration impact, causing a sudden mechanical mismatch between the exoskeleton system and the wearer's arm.

[0004] In this situation, the user's arm joints may be subjected to abnormal reverse pulling force, which can easily cause secondary strains or ligament damage. At the same time, since the reverse impact often occurs at the moment of rapid joint movement, the wearer's hand grip will also be disturbed. In extreme cases, it may cause the user to directly lose the stability of gripping the object, which not only increases the risk of the object falling and being damaged, but also poses a serious threat to the user's personal safety.

[0005] The information disclosed in the background section is only intended to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0006] The purpose of this invention is to provide a collaborative follow-up control method for upper limb exoskeleton systems based on an active disturbance rejection control strategy, so as to solve the problems in the background art mentioned above.

[0007] To achieve the above objectives, the present invention provides the following technical solution: a collaborative follow-up control method for an upper limb exoskeleton system based on an active disturbance rejection control strategy, comprising the following steps: S1 collects the changes in contact force between the user and the target object, the changes in end weight, and the grip slip signal, integrates them to generate a load sudden unloading risk warning information, and sets the initial slope parameter of the drive compensation curve accordingly. S2, based on the early warning information, construct the load reduction control trajectory, input it to the active disturbance rejection control system, dynamically adjust the original control target, and constrain the acceleration step amplitude; S3, during the execution of the unloading trajectory, modal locking is activated to restrict the joint degree of freedom response, and at the same time, a viscoelastic energy dissipation device with damping adjustment capability is activated to weaken the sudden reverse kinetic energy; S4 detects elbow joint torque. When a reverse trend occurs, the shoulder joint outputs a counter-flexible support torque to disperse residual impact energy through elbow-shoulder coordinated braking. S5 implements multi-scale acceleration modulation during braking, exponentially decays high-frequency impacts within a short time window, and sets an output upper limit within a long time window to maintain control stability. S6, after the control output stabilizes, performs consistency judgment on electromyographic signals and joint state. When the drive output deviates from the user's intention, it reduces the weight of active disturbance rejection control, switches to intention-driven control, and performs trajectory correction and drive anti-saturation regulation to build a safe closed loop for human-computer interaction.

[0008] Preferably, step S1 includes: Flexible strain sensors are installed on the user's palm, the back of the finger joints, and both sides of the forearm to collect pressure changes, frictional stress changes, and shear strain distribution trends in the contact area in real time, and extract dynamic characteristic parameters of relaxation or sliding in the grip posture. Three sets of force measuring units were installed at the connection between the forearm and wrist to collect data on changes in end load and the decreasing trend of gravitational acceleration, and to filter out high-frequency vibration disturbances to determine whether there is a tendency for rapid load release. The slippage precursor in the grip posture is time-series fused with the end load release trend. When the two are in the same direction and the time difference is less than 100 milliseconds, a load sudden unloading risk warning is generated and the initial slope limit value of the drive compensation curve is set. Based on the risk level of sudden load unloading, the rising slope of the original drive output curve is limited to restrict the rate of change of the output torque in the early stage. A slope limiting manipulator is superimposed in the output path to actively compress the rate of change of control error and reduce the reverse impact caused by inertial output.

[0009] Preferably, step S2 includes: Based on the current joint position, joint angular velocity, and end-load estimate, a load reduction control trajectory is constructed from the current output state to the theoretical no-load state. Electromyographic signals are then used to dynamically adjust the endpoint of this trajectory to match the user's actual movement trend. The load reduction control trajectory is injected into the control process by switching the main control signal channel, and the sudden unloading identification signal is used as the switching trigger condition. The trajectory acceleration direction and change amplitude are kept continuous within the time window before and after the switching, so as to achieve a flexible transition of auxiliary output. During the execution of the load reduction control trajectory, the slope limiter dynamically compresses the expected acceleration change amplitude to ensure that the actual output does not exceed the maximum allowable acceleration threshold, thus preventing the risk of structural overshoot and reverse pull of the driver. After the load reduction control trajectory is completed and the electromyographic activity is confirmed to be below the unloading threshold and the end load is stable, the control signal channel switches back to the task auxiliary trajectory channel and gradually restores the normal operation auxiliary force output state using a slow recovery method.

[0010] Preferably, step S3 includes: Based on the trajectory output direction, sudden unloading level and current joint motion state, it is determined whether there is an inertial impact trend. If so, the electromagnetic actuation structure set on both sides of the toothed ring of the joint shell is activated to trigger the toothed claw to bite the toothed ring to temporarily lock the joint degree of freedom. The viscoelastic energy buffer structure installed between the joint shell and the wearer's upper arm is activated simultaneously. It absorbs sudden inertial kinetic energy through the deformation of polyurethane and silicone composite materials, and achieves dynamic adaptation to different impact levels by adjusting the material compression stroke with the spiral limit ring. After electromagnetic locking is completed, the heating device inside the viscoelastic structure is triggered to raise the surface temperature of the polyurethane material to 40 degrees Celsius, reduce its elastic modulus, enhance its high-strength impact energy absorption capacity, and gradually cool down to restore its initial rigid state after the impact. The impact response is determined based on the integral value of the joint reverse acceleration and the viscoelastic elastic deformation. If the preset release condition is met, the electromagnetic lock is released and the drive output slope is suppressed to ensure structural stability and response continuity during the control signal recovery period.

[0011] Preferably, step S4 includes: The elbow joint torque changes are collected in real time by a metal foil strain gauge set on the inner side of the elbow joint drive shaft, and the dynamic characteristics of the elbow joint entering the reverse acceleration trend are identified. Based on the recognition results, the shoulder joint is controlled to output a flexible support torque in the opposite direction to that of the elbow joint in a neutral posture. The output process limits the upward slope and torque amplitude, and a flexible constraint path between the elbow and shoulder is constructed. By constructing a spatial mechanical closed loop through the inertial release of the elbow joint and the flexible support of the shoulder joint, the residual impact kinetic energy is guided from the elbow joint to the wearer's shoulder strap and back structure, forming a multi-structure collaborative energy absorption mechanism.

[0012] Preferably, step S5 includes: A short-time window acceleration limiting mechanism is activated. Acceleration sensing elements located at the elbow and shoulder joints detect high-frequency impact components and activate the current limiting unit to limit the rate of torque change, thereby quickly weakening high-frequency disturbances. After a short period of adjustment, a medium time window is entered to control the drive output to recover smoothly from the suppressed state, limiting the changes in acceleration slope and angular velocity to avoid discontinuous output caused by control rebound; During the long-term window phase, the upper limit of torque output is dynamically limited by evaluating the trend of total energy output per unit time to prevent the control channel from entering saturation. A unified control cycle synchronization mechanism is set up, which sequentially nests short, medium, and long time windows and sets a transition zone to achieve continuous switching and stable connection of control strategies during execution.

[0013] Preferably, step S5 includes: The system compares the user's electromyography signals with the elbow and shoulder joint movement in real time. By judging the consistency of movement direction and acceleration trend, it identifies whether the drive output deviates from the user's true intention. When human-machine deviation is continuously detected, the signal gain of the disturbance observation channel is immediately reduced and its output influence is frozen. At the same time, the trajectory control mode guided by electromyography trend is activated to generate an intention-driven angular velocity control target. During trajectory control, execution errors are monitored in real time. If the deviation from the intended range continues, the target angle and speed boundary are adjusted to correct the trajectory and improve the following accuracy. During the trajectory correction phase, drive anti-saturation control is activated. If the detected output value exceeds the safety threshold or the rate of change increases abnormally, the amplitude control torque and output slope are immediately limited to ensure long-term stable and safe human-machine interaction.

[0014] The technical effects and advantages provided by the present invention in the above technical solution are as follows: This invention generates real-time unloading risk warnings by fusing multi-source signals and constructs a load reduction control trajectory for feedforward adjustment of the drive output. Supplemented by modal locking and viscoelastic energy dissipation mechanisms, it enables timely physical absorption of impact energy during sudden load loss. Simultaneously, by combining elbow-shoulder coordinated braking and multi-scale acceleration modulation, it achieves efficient suppression of the impact process in both spatial and temporal dimensions. Furthermore, by dynamically switching to a human-intention-driven trajectory control path through electromyographic intent discrimination, and in conjunction with trajectory correction and drive anti-saturation control strategies, it ensures stable, safe, and compliant interactive control during high-frequency dynamic disturbances and long-term complex operations. This significantly improves the safety, reliability, and wearing comfort of the exoskeleton system in practical agricultural handling applications. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0016] Figure 1 This is a flowchart of the method for the coordinated follow-up control of an upper limb exoskeleton system based on an active disturbance rejection control strategy according to the present invention. Detailed Implementation

[0017] Exemplary embodiments will now be described more fully with reference to the accompanying drawings. However, these exemplary embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, they are provided so that the description of this disclosure will be more complete and fully convey the concept of the exemplary embodiments to those skilled in the art.

[0018] This invention provides, for example Figure 1 The illustrated method for coordinated servo control of an upper limb exoskeleton system based on an active disturbance rejection control strategy includes the following steps: By collecting changes in contact force between the user and the target object, changes in end weight, and slip signals in the grip posture, the multi-source data is fused and calculated to generate early warning information for the risk of sudden load unloading. Based on this, the initial slope parameter of the drive compensation curve is set to limit the transient steep changes in drive output. To achieve rapid identification and proactive intervention in sudden unloading events, it is necessary to sense and fuse multiple physical signals during user operation to generate effective early warning information. Based on this, the drive control strategy can be feedforward-limited to avoid overshoot and impact problems caused by inertial drive output delay. The specific implementation method for this step is as follows: During the use of an upper limb exoskeleton device in agricultural scenarios for fruit basket handling, flexible strain sensors located on the back of the user's palm, the first and second joints of the fingers, and the radial and ulnar sides of the forearm continuously collect data on normal pressure changes, tangential frictional stress changes, and local shear strain distribution trends at the contact points with the fruit basket surface. These devices are flexible bodies made of silicone rubber substrate, with an embedded strain bridge array of copper-nickel conductive fibers. They are elastically attached to the skin surface, with a sampling period of 2 milliseconds. After absolute value normalization, the collected signals form a time series containing three physical quantities: pressure change rate, frictional force direction shift rate, and shear stress center drift. Subsequently, by combining the rapid switching characteristics of the applied force direction and the abrupt change characteristics of the contact area in this series, it is determined whether the current grip state exhibits a rapid relaxation or unintentional slippage trend. Specifically, a pressure change rate greater than 20 kPa, a frictional force shift rate exceeding 4 N / s, and a shear stress center drift exceeding 20% ​​of the sensor's length direction are considered precursors to slippage. If the above three parameters exceed the set threshold more than twice consecutively within 250 milliseconds, it can be preliminarily determined that the grip stability is in a critical state.

[0019] The results of the grip stability assessment were verified by fusing them with changes in the end-effector load. To this end, three high-precision force measurement units were installed at the forearm-wrist connection of the exoskeleton device, located on the forearm tendon side, palm side, and ulnar side, respectively. Each force measurement unit was used to collect changes in gravitational acceleration and the total load trend in the direction of the applied force at the end-effector. Each unit contained a strain gauge and an accelerometer to detect the decreasing trend of the vertical load and horizontal vibration interference, respectively. The specific judgment criteria were: when the measured rate of decrease of the vertical load at the end-effector reached 12 Newtons per second per unit time, and the decrease exceeded 35% of the user's initial carrying state within 200 milliseconds, it was considered that there was a clear tendency for load release. During the judgment process, end-effector vibration was simultaneously excluded to ensure that the load fluctuation was not caused by foot tremors or natural shaking. The aforementioned contact force change characteristics and rapid load decrease characteristics were only confirmed as an impending sudden unloading event if they simultaneously met the set thresholds and showed high consistency within the time window—that is, the grip loosening and load release occurred in the same direction and the time difference was less than 100 milliseconds. At this point, a sudden unloading risk warning signal is generated, the warning is marked as high priority and sent to the control strategy processing stage.

[0020] Upon receiving a high-priority sudden unloading warning signal, the control response pre-adjustment mechanism is immediately activated to limit the slope of the time evolution curve during the auxiliary drive force output process. This process first analyzes the initial rising segment of the original drive curve, calculates the linear growth rate of the current output torque over time, and manually limits this rate. For example, if the initial drive output torque rises from 0 to 2.5 N·m in 1 second in the original setting, its linear rise slope is 2.5 N·m per second. If the current warning level is the highest, this rise slope is set to not exceed 1.0 N·m per second, and the output is limited to not exceeding the instantaneous target value corresponding to this linear value within the first 100 milliseconds. This limit is matched to the actual unloading trend direction detected in the previous step, meaning it only takes effect when the grip-relaxation direction is consistent with the torque output direction, avoiding accidental triggering of system drive delays. Furthermore, to improve the coordination between the control output and the user's operational intention, the initial output torque value is also synchronously adjusted to the historical average joint torque value at the start of the corresponding action, achieving a natural transition in the action response. The adjustment amount depends on the median value of the actual output range of the same joint actuator within the past 200 milliseconds, thus ensuring that the actuator does not maintain high force output after a sudden unloading.

[0021] The adjusted initial slope parameter is embedded as a control input constraint into the entire auxiliary force output control process, and the state response channel in the exoskeleton control strategy is updated synchronously. This update process does not change the overall control law structure of the main controller, but only adds a slope limiting manipulator to the drive output path to compress the error rate between the external input and the internal target state to within the set range in real time. At the same time, the estimation mechanism in the disturbance response path of the controller interrupts the delayed identification process of the current inertial state, resets the initial value of the disturbance variable, and enables the disturbance compensation subsystem to quickly converge to the no-load state, preventing the inertial disturbance model from continuing to apply high force output. The intervention of this slope limiting mechanism realizes an active control response soft-start logic, which can quickly reduce the response rate of the controller output channel when the system is still in the initial stage of sudden unloading, avoids the inertial force caused by sudden unloading from continuously acting on the wearer's upper limb joints, and thus prevents the generation of reverse traction force beyond the physiological tolerance range.

[0022] Through the coordinated execution of the above steps, a predictive model of the operational trend was established before the sudden unloading occurred, output regulation was realized when the sudden unloading occurred, response stability was ensured during the output process, and a safe closed loop of control compensation was finally completed, providing a stable, safe, and repeatable control basis for the application of upper limb exoskeletons in high-intensity agricultural operations.

[0023] After obtaining the early warning information of sudden unloading risk, it is constructed into a load reduction control trajectory and input into the active disturbance rejection control system to dynamically adjust the original control target and actively constrain the step amplitude of the acceleration response, so that the control compensation process has continuity and stability. To achieve real-time control response after identifying sudden unloading risks and improve the safety and interaction consistency of the upper limb exoskeleton device in dynamic handling scenarios, dynamic reconstruction of the control target must be performed immediately after the early warning information is generated, and compensated transition control must be performed through trajectory injection. The specific operational steps of this implementation process are as follows: After generating the sudden unloading warning information and obtaining confirmation from three physical quantities—a sudden change in grip posture, a sharp decrease in contact pressure, and rapid release of the vertical weight of the load—that a sudden unloading behavior is about to occur, a load reduction control trajectory is immediately constructed based on three dynamic parameters: the current joint position, the joint angular velocity, and the estimated end-load. The trajectory is constructed using a piecewise displacement interpolation method, with the starting point taken from the current actuator output state and the ending point corresponding to the theoretical support state under no-load conditions. Taking the elbow joint as an example, if the current actuator output torque is 2.3 N·m, and the preset required actuator output under no-load conditions is 0.8 N·m, then the trajectory will gradually reduce the auxiliary force output to 0.8 N·m over a duration of 500 milliseconds, while maintaining an angle output rate of less than 20 degrees per second to avoid physiological discomfort caused by rapid joint contraction. During the path design process, the user's muscle force output trend is considered, and the degree of muscle cooperation is assessed through electromyography (EMG). If the flexor muscles are found to be in a relaxed state, the load reduction endpoint setting is further reduced, making the control strategy more closely match the natural trend of the wearer's actual release action.

[0024] The constructed load reduction control trajectory is input into the current control channel as a continuous curve to cover the original standard auxiliary force output trajectory set based on the task objective. The input method employs a master control signal switching strategy, inserting the load reduction channel into the original trajectory execution sequence. The control command channel switches priorities based on a sudden unloading signal. The switching point is strictly synchronized with the instant of sudden unloading identification in the previous step, with an error not exceeding 10 milliseconds, ensuring timely trajectory coverage response. To avoid nonlinear control response caused by abrupt changes in trajectory commands, this implementation uses a linear transition mechanism for trajectory splicing. Within 50 milliseconds before and after the switch, the two trajectory curves maintain the same acceleration direction and continuous amplitude, ensuring the control output torque curve is continuous, its derivative exists, and it is differentiable, avoiding driver overshoot caused by trajectory discontinuity. During trajectory execution, the attitude adjustment frequency is always maintained at no less than 50 Hz to ensure that trajectory updates maintain high response speed and low latency characteristics during actual operation.

[0025] To avoid mechanical structural impact or drive-end overshoot due to sudden acceleration changes during trajectory switching and load reduction control, this implementation embeds a dynamic acceleration amplitude constraint mechanism into the control execution path. Specifically, before the control signal is transmitted to the driver's front end, a slope limiter compresses the expected acceleration change amplitude to limit the steepness of error adjustment in the actual output. This slope limiter compares the acceleration difference between the current control target output point and the output point of the previous control cycle and limits this change value to no more than a set maximum acceleration limit threshold. Taking shoulder joint drive as an example, if the maximum allowable acceleration is 0.9 meters per second squared, when the acceleration difference caused by the original trajectory change exceeds this threshold, the slope limiter automatically compresses it to the maximum allowable value, maintaining the direction of the compressed acceleration consistent with the original target acceleration direction to avoid structural interference or reverse impact due to reverse pulling. Through this active constraint processing, even in a sudden unloading event with a large trajectory descent slope, it can prevent the driver from following too closely and causing the mechanical end to perform actions beyond the physiological safety range.

[0026] After the load reduction control trajectory is completed, the system needs to revert to the target state to reconnect to the regular task-assisted control logic, ensuring a smooth transition from the unloading and release state to the normal operating state. To this end, control strategy switching conditions are set based on load state determination and electromyography (EMG) activity amplitude monitoring. During execution, if the wrist end load sensor value remains below 0.5 kg and the average EMG value of the biceps brachii and forearm flexor muscles is more than 20% lower than the average EMG value in the unloading state for more than 300 milliseconds, the sudden unloading behavior is considered complete and the user has entered a stable and relaxed state. At this time, the controller automatically releases the load reduction trajectory channel and redirects the main control signal back to the task-assisted trajectory channel. To avoid pulling or mis-assistance caused by immediately entering a strong assist output after the sudden unloading trajectory ends, the controller adopts a "gradual rise recovery mode" for the next 200 milliseconds. This means the assist force output slowly transitions from the unloading end state to the target task-assisted state, and the output curve is set as a cubic gradual rise curve with continuous first and second derivatives to avoid oscillations or sudden force changes during the driver response, allowing the user to maintain a natural movement experience during control strategy switching.

[0027] Through the above steps—generating a load-reducing control trajectory based on multi-source physiological and mechanical states, preferentially injecting this trajectory into the control process, constraining the acceleration response during trajectory execution, and achieving safe back-switch and control recovery after the trajectory ends—a complete dynamic response strategy for sudden unloading risks is constructed. Compared with the existing technology that only passively corrects after the error peak occurs, this invention has the advantages of early warning-driven triggering, active trajectory injection, smooth process control, and output limitation protection, significantly improving the control safety, response speed, and wearer comfort of the upper limb exoskeleton in handling scenarios facing sudden unloading events.

[0028] As the load reduction control trajectory enters the execution phase, a modal locking strategy is initiated to temporarily constrain the joint degree of freedom response, and a viscoelastic energy dissipation device with damping adjustment capability is activated simultaneously to dissipate sudden reverse kinetic energy physically and reduce joint impact effect. To avoid overshooting of the joint structure due to inertial reaction forces during the load reduction control trajectory execution process, which could potentially damage the wearer's upper limbs, a modal locking strategy must be implemented in conjunction with the joint structure status when the actuator load reduction control trajectory enters the actual operation phase. This strategy is then linked to a viscoelastic material energy dissipation structure with damping adjustment capabilities to construct a complete impact mitigation scheme. The specific steps of this implementation method are as follows: Before the load reduction control trajectory is executed, the current motion trend of the joint is evaluated item by item based on the trajectory output direction, the sudden unloading judgment level, and the current joint motion state parameters. Taking the elbow joint as an example, if it is identified that its drive output direction is consistent with the sudden unloading load release direction, that is, it is still providing upward auxiliary force before the controller has fully unloaded the load, and at this time the load has been suddenly unloaded, then the elbow joint is at risk of closing rapidly due to inertia. Three types of known data are used as the judgment criteria during the judgment process: first, the angle between the current driver output torque and the sudden unloading direction is less than 15 degrees; second, the angle change in the predicted trajectory within the next 300 milliseconds exceeds 70% of the current angle movable range; and third, the sudden unloading level is high, representing a strong reverse inertial impact trend. Once any one of the above three conditions is met, the modal locking operation is immediately executed. The modal locking operation is completed by an electromagnetic actuation mechanism set on both sides of the elbow joint body. This mechanism includes a limiting toothed claw nested in the outer shell toothed ring, an aluminum-copper coil winding, and a quick-trigger return spring. When the triggering conditions are met, the electromagnetic device is activated, and the coil generates a magnetic field that attracts the toothed claw and the toothed ring to engage, restricting the axial rotational freedom of the joint and preventing it from continuing to close under inertia, which could cause structural impact or forced compression of the wearer's elbow tendons.

[0029] Simultaneously with modal locking, a viscoelastic energy dissipation structure installed between the joint shell and the wearer's upper arm is immediately activated. This structure consists of two layers of cushioning pads made of different materials: an outer layer of thermoplastic polyurethane elastomer with a thickness of 3 mm and an initial elastic modulus of 0.6 MPa per square millimeter; and an inner layer of silicone composite containing closed-cell microbubbles with a thickness of 4 mm, exhibiting excellent energy absorption performance. After directional processing, this structure possesses high-frequency strain response capability. When the exoskeleton's drive end maintains a certain output torque due to incomplete unloading, and the wearer's arm suddenly stops moving due to a sudden unloading event, the relative displacement between the two may exceed 4 mm within 1 second. The resulting collision energy is absorbed and dispersed by the aforementioned cushioning structure. The viscoelastic material can undergo internal molecular chain slippage within a short time, converting kinetic energy into heat energy and effectively reducing the amplitude of mechanical impact transmitted to the wearer's arm surface. To accommodate the buffering requirements of different levels of sudden unloading, the structure is also equipped with a set of adjustable deformation limits. The compressible stroke of the viscoelastic material is adjusted by a mechanical spiral limit ring, thereby controlling the energy dissipation capacity. Under heavy load conditions, the stroke limit ring can be relaxed to a maximum deformation of 10 mm to improve its kinetic energy absorption efficiency.

[0030] To ensure the modal locking structure and viscoelastic energy buffer structure work synergistically, this embodiment uses a thermal response linkage mechanism to regulate the triggering sequence and physical state synchronization between the two. After the electromagnetic actuation structure completes its locking action, an electric heating element located on the inner surface of the viscoelastic structure automatically begins heating. This heating element consists of a polyimide film carrier and a nickel-chromium alloy heating wire, powered by a 12-volt supply with a peak temperature rise rate of 1.5 degrees Celsius per second. When a sudden unloading event is identified as a high-level event, the heating element raises the surface temperature of the viscoelastic material to approximately 40 degrees Celsius, causing the polyurethane material to transition from a glassy state to a highly elastic state, reducing its elastic modulus by about 20% and providing higher deformation tolerance. This treatment significantly enhances its energy absorption capacity against sudden high-intensity impacts. Simultaneously, within 30 seconds after the impact process ends, the system gradually reduces the heating voltage, allowing the material to return to its initial rigidity and maintaining structural stability. The above-mentioned thermal response synchronization control logic can ensure that the modal locking structure and the energy dissipation structure maintain coordinated and matched physical characteristics throughout the entire process of sudden unloading response, avoiding control abnormalities or structural imbalances caused by insufficient or excessive action of one party.

[0031] To ensure a smooth transition between the natural release of the locked state and subsequent control logic, a delayed release logic must be implemented after the modal locking and the completion of the impact response by the viscoelastic energy-dissipating structure. Smooth management of the lock release process is also required. This step employs a joint reverse acceleration integral determination method. Specifically, by collecting the inertial reverse acceleration change curve of the joint during the locking period, if it remains below 0.3 meters per second squared for 300 milliseconds and the elastic deformation of the viscoelastic structure is less than 20% of the original compression, the inertial impact process is considered complete, allowing the lock to be released. The electromagnetic locking structure disengages from the engaged state via a power-off release mechanism, and a spring reset device restores the pawl to the standby position. The entire process is completed within 100 milliseconds without angular jumps or output oscillations. Simultaneously, the acceleration change slope of the controller's internal output trajectory is synchronously suppressed to within 0.6 meters per second squared to achieve dynamic alignment between the control signal and the physical structure state, ensuring that no second inertial output occurs after the actuator recovers. This step ensures that the control layer response is stable and continuous after the physical layer buffering action is completed, achieving a seamless transition from the impact response state to the normal control state.

[0032] Through the above steps, including joint modal locking triggering, viscoelastic energy buffer structure activation, coordinated thermal response adjustment, and delayed release control of the locked state, effective absorption and dynamic control transition of the reverse impact kinetic energy that may occur during a sudden unloading event are achieved. Compared with existing technologies that rely solely on delayed control logic for emergency compensation or post-processing, this invention not only has significant advantages in structural response speed but also demonstrates a high degree of coupling in the coordination of energy processing methods and control commands, thereby ensuring the wearer's physiological safety, device stability, and continuity of control response during sudden unloading events.

[0033] Subsequently, the elbow joint torque response was monitored. When a reverse acceleration trend appeared, the shoulder joint output a reverse flexible support torque. Through the coordinated braking action of the elbow and shoulder, the residual impact kinetic energy was dispersed among multiple joints, thereby reducing the local structural load. After executing the load reduction control trajectory and completing modal locking and viscoelastic energy dissipation buffering, a certain amount of residual inertial kinetic energy may still exist inside the exoskeleton actuator. This is particularly evident in the elbow joint, which tends to evolve from assist to reverse acceleration within a short period. Without proper guidance and dispersion, this can easily cause instantaneous impact or mechanical damage to the local structure. A cross-joint cooperative braking mechanism is proposed. Specifically, by real-time monitoring of the elbow joint's output torque changes, a flexible support torque in the opposite direction is triggered by the shoulder joint, thereby forming an energy guiding path between the elbow and shoulder. This draws the impact load away from a single joint structure and diffuses it to the overall upper limb structure, effectively reducing local stress concentration. This mechanism includes the following steps: A process for identifying the reverse acceleration trend of the elbow joint is established. This process relies on the axial strain sensing structure of the drive arm inside the joint to continuously collect the dynamic changes in torque generated by the elbow joint during the unloading process after sudden unloading. The sensor used is a metal foil strain gauge, which is closely attached to the inner side of the elbow joint drive shaft, and can acquire the strain response of the output shaft in real time during joint rotation. The sensor sampling frequency is set to 2000 times per second to ensure that signal changes are captured immediately at the beginning of the reverse torque. The judgment criterion is: if, within the continuous sampling period, the direction of the torque output by the elbow joint changes from the assist direction to the suppression direction, and the angular velocity change shows a monotonically increasing trend within 80 milliseconds, and the angle increment exceeds three times the average value of the previous 100 milliseconds, then the elbow joint is considered to have entered a reverse acceleration trend. At this time, a synchronous control command is issued to the shoulder joint drive unit to prepare to output a flexible support torque in the opposite direction to form the subsequent mechanical linkage path. This judgment process is based on the scenario that the reverse kinetic energy has not been completely dissipated in the previous stage of viscoelastic buffering, as a secondary protection mechanism to ensure that the residual impact that has not been dissipated can be guided and transferred.

[0034] Based on the above assessment, the flexible support torque output mechanism of the shoulder joint is activated. The shoulder joint structure employs a mechanical structure driven by a brushless DC motor and connected to the output shaft via a single-stage planetary reduction gear, enabling controllable output in both forward and reverse directions. Upon receiving the reverse acceleration signal from the elbow joint, the motor control system first locks and evaluates the current posture of the shoulder joint, confirming that it is in a neutral torque state, i.e., within a buffer zone with no obvious rotational tendency or a load less than 2 N·m. Then, it initiates the reverse torque application process within 0.1 seconds. The output direction of this torque is opposite to the current inertial motion direction of the elbow joint, not a direct rigid block, but rather set within a flexible control range, ensuring that the slope of its output curve during the rising phase does not exceed 0.05 N·m / ms, and the overall output amplitude does not exceed 60% of the wearer's shoulder joint structure's maximum allowable torque. For example, for an adult male user, whose shoulder joint safety limit torque is 10 N·m, the maximum output during this phase is limited to within 6 N·m, and the drive motor is controlled to output smoothly through the current regulation logic within the torque controller. The torque output by the shoulder joint is not used to drive the movement, but to generate a gentle mechanical constraint in the middle of the wearer's upper arm in the opposite direction to the elbow joint, so as to form a buffer chain opposite to the direction of the elbow's inertial impact, thereby guiding and transferring energy in the proximal direction.

[0035] When the elbow and shoulder joints generate opposing torques, a cross-joint mechanical diffusion path is created throughout the wearer's upper limb structure. This transfers the inertial impact, originally concentrated at the elbow, to the shoulder-arm connection and the shoulder girdle and back attachment areas. The dynamic basis of this process lies in creating a spatial closed loop between the inertial release of the elbow joint and the flexible support of the shoulder joint. This loop prevents some of the impact energy from being concentrated on the elbow structure, instead distributing it rapidly through muscles, soft tissues, joint interfaces, and the skeletal chain to the entire upper limb support system. Specifically, the force distribution path manifests as follows: the opposing torque generated by the elbow joint is transmitted to the mid-upper arm via the skeletal connection, countered by the opposing force of the shoulder joint forming a backward-pulling structure, and finally transmitted to the muscle-fascia area of ​​the wearer's back through the shoulder girdle and backplate. Multiple soft tissues and skeletal structures involved in this path participate in the dissipation of impact energy, not only reducing the impact load on the elbow structure in a very short time but also giving the overall human-machine interface system a certain impact tolerance and dynamic stiffness-flexibility adjustment capability. Ultimately, the shoulder joint maintains a flexible support state for approximately 300 milliseconds. During this period, the controller continuously fine-tunes its output curve to ensure that excessive traction does not cause secondary dynamic interference. After this process, as the inertial force of the elbow gradually disappears, the shoulder joint gradually withdraws its support torque output and returns to a natural assisted state.

[0036] During the operation of the coordinated braking mechanism, a multi-scale acceleration modulation strategy is introduced. In the short time window, the high-frequency impact component is suppressed by exponential decay. In the long time window, a dynamic output upper limit is set to ensure the driving continuity and control stability under different time scales. After the elbow and shoulder joints achieve initial dispersion of the reverse impact energy caused by the sudden unloading event through a coordinated braking strategy, there are still risks of short-term high-frequency disturbances and medium-to-long-term low-frequency inertial output. To ensure the continuity, stability, and smooth output of the entire drive control process during this dynamic adjustment period, an acceleration modulation method based on time scale division is proposed. By setting a staged adjustment strategy with short, medium, and long time windows, the instantaneous peak value of the driver output acceleration, the rate of change of velocity, and the total energy input are intervened in a graded manner, thereby constructing a dynamically stable human-machine interaction state. The specific implementation process includes the following steps: After the elbow joint detects a reverse torque trend and the shoulder joint initiates flexible support output, an acceleration response monitoring and limiting process is immediately initiated within a short time window. The purpose of this process is to suppress instantaneous high-frequency acceleration changes caused by sudden unloading, preventing the actuator from continuing to generate secondary impacts along the inertial output path. The implementation involves placing triaxial acceleration sensing elements (MEMS micromechanical sensors) inside both the elbow and shoulder joints, fixed to the inner side of the drive arm's main shaft, and using metal welding encapsulation to enhance mechanical coupling accuracy. The sampling frequency is 2000 times per second. The sensors continuously acquire linear acceleration data in the actuator's output direction and identify instantaneous peak values. If an acceleration exceeding twice the original average value is detected within 50 milliseconds, and the change period between adjacent peaks is less than 20 milliseconds, it is determined that an impact-type high-frequency disturbance exists. After confirming the disturbance, the control circuit activates the current limiting unit located in the driver power input channel. This limiting unit limits the maximum current output value to 60% of the original output through a variable resistor and a bypass grounding structure, and limits the driver torque change rate to no more than 0.03 N·m per millisecond through an inductor buffer device, thereby achieving rapid physical attenuation of high-frequency disturbances.

[0037] After the initial suppression of impact fluctuations within a short time window, the system enters a medium-time window modulation stage to control the output "rebound" effect caused by excessively rapid high-frequency reduction, preventing oscillating compensation in the controller. This stage lasts between 150 and 600 milliseconds, with the primary goal of constructing a smooth recovery path that allows the driver's torque output to gradually recover from the suppressed state to the task-assisted state, while ensuring a smooth and continuous acceleration change process. The implementation involves acquiring the acceleration trend curve from the previous stage and limiting its unit-time variation to within 80% of the original set value before its upward slope reaches a stable value. For example, if the shoulder joint's upward acceleration slope is set to 1.5 meters per second squared before the sudden unloading, the output is limited to 1.2 meters per second squared by adjusting the drive current growth rate in this stage, and the angular velocity change rate is controlled to not exceed 8 degrees per second. A delay-triggered smooth control mechanism is introduced into the control logic, releasing the limiting segment only after confirming that the acceleration fluctuation has stabilized for 30 milliseconds, ensuring a complete buffering process and avoiding "sudden drops and rises" in control reversal.

[0038] Entering the long-term window control phase, lasting from 600 milliseconds to 1500 milliseconds after the start of the load reduction trajectory, the main objective is to prevent the control system from entering a continuous saturation state due to excessive sudden unloading response. This state often manifests as the actuator output approaching its upper limit for an extended period, potentially leading to increased structural thermal load and control response failure. Therefore, during this phase, an output total assessment mechanism is activated. By dynamically monitoring the total drive torque energy of the previous second, it is determined whether the current output has entered the "compensation saturation" region. Taking the shoulder joint as an example, if its output torque exceeds 85% of the rated value three times consecutively within one second, and the cumulative duration is greater than 600 milliseconds, then its output upper limit is limited to 70% of the rated value within the next 500 milliseconds. For example, if the rated maximum output of the shoulder joint is 12 Newton-meters, then the dynamic limit value is set to 8.4 Newton-meters. Simultaneously, the angular acceleration variation range is limited to no more than 1 meter squared per second to suppress the system from entering a continuous high-dynamic state.

[0039] To ensure that the control strategies at the three time scales do not interfere with each other during execution and maintain dynamic continuity within the overall control structure, this implementation constructs a unified control cycle synchronization mechanism. Specifically, the three time windows are nested sequentially in a fixed order, with start and end times set to 0-150 milliseconds, 150-600 milliseconds, and 600-1500 milliseconds, respectively. Within each stage, the control command switches states based on the state flag of the previous stage, with a 25-millisecond transition zone at the switching point to smoothly process the output signal. Furthermore, during control execution, the controller performs a state check every 100 milliseconds, comprehensively evaluating the amplitude, fluctuation period, and torque stability of the acceleration curve in the feedback signal to automatically determine the current time period and the appropriate modulation strategy. This ensures that the control output maintains physical continuity, dynamic stability, and behavioral consistency throughout the entire process from rapid suppression and high-speed recovery to long-term amplitude limiting.

[0040] After the control output stabilizes, the electromyographic signal and joint motion state are compared in real time to determine whether the drive output deviates from the user's motion intention. Once the deviation exceeds the threshold, the weight of active disturbance rejection control is immediately reduced, and the trajectory control mode dominated by motion intention is switched to implement trajectory correction and drive anti-saturation regulation, thereby constructing a long-term safe and stable human-computer interaction closed loop. To enhance the stability of human-machine interaction during long-term use of exoskeletons in complex agricultural work scenarios, after acceleration modulation and cross-joint coordinated braking, the control system needs to continuously monitor the coordination between the drive output and the user's actual movement intention to prevent human-machine deviations caused by sudden disturbances, perception lag, or control drift. To this end, a control adjustment method based on comparing electromyographic signals with the actual joint movement state is proposed. When the deviation exceeds a threshold, the active disturbance rejection control weight is immediately reduced, switching to an intention-driven trajectory control mode, and trajectory correction and drive anti-saturation regulation are implemented to construct a human-machine interaction closed loop with dynamic adaptability. Specifically, the steps include: After multi-scale modulation is completed in the initial implementation phase and the driver output enters a stable range, the system initiates a real-time comparison process between electromyographic (EMG) signals and joint motion states. To accurately identify the user's true movement intentions, surface EMG electrodes are installed on several key muscle groups in the wearer's upper limb during implementation, specifically including the biceps brachii, triceps brachii, anterior deltoid, levator scapulae, and brachioradialis. Each electrode is made of silver chloride material and is attached to the skin using a reusable adhesive gel. The sampling frequency is 1000Hz to ensure the capture of rapid potential changes during muscle activation. The EMG acquisition circuitry is processed by a low-pass filter and differential amplifier in the hardware circuitry to remove power frequency interference and motion artifacts. The processed EMG waveform is then fed into a signal trend extraction process in real time to identify the currently activated muscle group and movement trend.

[0041] While acquiring electromyographic (EMG) data, high-precision rotary encoders distributed along the axial directions of the elbow and shoulder joints obtain the current joint angle change values, and gyroscopes attached to the drive arm acquire the joint angular velocity data in real time. The time synchronization error of the two data streams is controlled within 2 milliseconds to ensure data comparison accuracy. Throughout the monitoring process, the control flow continuously judges the degree of consistency between the desired movement direction indicated by the EMG signal and the actual joint movement direction. If the dominant EMG activation direction is opposite to the joint movement direction for three consecutive times, or if the EMG signal amplitude gradually increases while the angular velocity shows a continuous decreasing trend, and the duration exceeds 100 milliseconds, the system determines that the current driving behavior has deviated from the user's active intention.

[0042] Once a deviation from the human-machine interface is confirmed, the control strategy is immediately adjusted. To avoid misleading compensation from the disturbance observer due to delay effects in active disturbance rejection control, this step first gradually reduces the signal gain of the disturbance observation channel to below 30% and freezes its output's impact on the main torque channel. Simultaneously, the intention-oriented control path is activated, transforming the aforementioned electromyographic signal trend into the desired joint angular velocity range, which is then used as the updated target for drive control.

[0043] At this point, the controller no longer relies primarily on previous external disturbance observations for control, but instead generates control commands based on electromyographic trends. In implementation, a reference trajectory generator based on electromyographic trends is introduced into the control signal path. This transforms the expected motion in the current muscle-dominant direction into an angularly varying trajectory, while simultaneously limiting the rate of change of angular velocity to ensure that the control output slope does not exceed 30% of the average rate of change in the previous stage, preventing structural oscillations caused by sudden trajectory changes. The switching process lasts approximately 10 milliseconds, during which the last control signal in the original active disturbance rejection output path is used as the transition node for the new trajectory, ensuring continuous output and avoiding abrupt jumps.

[0044] After switching the control strategy and initiating electromyographic (EMG) intent-based trajectory control, a trajectory correction and drive anti-saturation processing mechanism is further activated to ensure that the output behavior remains consistent with the user's intention. The trajectory correction is performed as follows: using the current EMG signal activation level as the dynamic baseline for the control target, a 3-degree error tolerance is set in the joint angle data. If the trajectory execution result deviates from this range for more than 30 milliseconds, fine-tuning of the trajectory is performed. This adjustment process does not regenerate the entire control trajectory; instead, it only adjusts the change range of the current target angle and the upper limit of the angular velocity, keeping the controller output within the intended trend range and preventing the accumulation of small errors from leading to a significant increase in error.

[0045] Simultaneously, an anti-saturation protection mechanism is activated during the execution of trajectory adjustment commands by the driver. This mechanism monitors the torque and angular velocity output by the driver in real time. If the output value exceeds the wearer's safe torque threshold for two consecutive sampling cycles (e.g., 8 N·m for the elbow joint and 12 N·m for the shoulder joint), the driver output is immediately limited to no more than 85% of the current maximum permissible value. Furthermore, by evaluating the integral rate of the output signal, if the output growth rate is found to exceed 1.5 times the average of the previous cycle, indicating a surge in output, output reduction control is immediately activated. In the next control cycle, the output slope is reduced to 50% of the original rate to prevent discomfort caused by control overshoot or physiological response lag.

[0046] Through the above steps—consistency determination of electromyography and joint status, dynamic switching and trajectory correction of control strategies, and drive output limiting protection—this invention constructs a complete human-computer closed-loop interaction method with "active recognition, real-time adjustment, and safety feedback" capabilities at the control logic, perception path, and output behavior levels. This method not only enhances the exoskeleton's responsiveness to user intentions in sudden dynamic scenarios but also significantly reduces the risk of structural impact and user muscle fatigue caused by control deviations.

[0047] The aforementioned collaborative motion control method for upper limb exoskeleton systems based on active disturbance rejection (ADRP) strategies enables rapid dynamic response and energy regulation of the drive output during sudden unloading events, effectively addressing joint impact issues caused by control delays and inertial overshoot in existing technologies. This method generates real-time unloading risk warnings by fusing multi-source signals and constructs a load reduction control trajectory for feedforward adjustment of the drive output. Coupled with modal locking and viscoelastic energy dissipation mechanisms, it achieves timely physical absorption of impact energy during sudden unloading. Simultaneously, the combined elbow-shoulder coordinated braking and multi-scale acceleration modulation techniques efficiently suppress the impact process in both spatial and temporal dimensions. Furthermore, by dynamically switching to a human-intention-driven trajectory control path based on electromyographic intent discrimination, and in conjunction with trajectory correction and drive anti-saturation control strategies, stable, safe, and compliant interactive control is ensured during high-frequency dynamic disturbances and long-term complex operations, significantly improving the safety, reliability, and wearing comfort of the exoskeleton system in practical agricultural handling applications.

[0048] The foregoing has only described certain exemplary embodiments of the present invention by way of illustration. Undoubtedly, those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the foregoing drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

Claims

1. A method for coordinated servo control of an upper limb exoskeleton system based on an active disturbance rejection control strategy, characterized in that, Includes the following steps: S1 collects the changes in contact force between the user and the target object, the changes in end weight, and the grip slip signal, integrates them to generate a load sudden unloading risk warning information, and sets the initial slope parameter of the drive compensation curve accordingly. S2, based on the early warning information, construct the load reduction control trajectory, input it to the active disturbance rejection control system, dynamically adjust the original control target, and constrain the acceleration step amplitude; S3, during the execution of the unloading trajectory, modal locking is activated to restrict the joint degree of freedom response, and at the same time, a viscoelastic energy dissipation device with damping adjustment capability is activated to weaken the sudden reverse kinetic energy; S4 detects elbow joint torque. When a reverse trend occurs, the shoulder joint outputs a counter-flexible support torque to disperse residual impact energy through elbow-shoulder coordinated braking. S5 implements multi-scale acceleration modulation during braking, exponentially decays high-frequency impacts within a short time window, and sets an output upper limit within a long time window to maintain control stability. S6, after the control output stabilizes, performs consistency judgment on electromyographic signals and joint state. When the drive output deviates from the user's intention, it reduces the weight of active disturbance rejection control, switches to intention-driven control, and performs trajectory correction and drive anti-saturation regulation to build a safe closed loop for human-computer interaction.

2. The method for coordinated follow-up control of an upper limb exoskeleton system based on an active disturbance rejection control strategy according to claim 1, characterized in that, Step S1 includes: Flexible strain sensors are installed on the user's palm, the back of the finger joints, and both sides of the forearm to collect pressure changes, frictional stress changes, and shear strain distribution trends in the contact area in real time, and extract dynamic characteristic parameters of relaxation or sliding in the grip posture. Three sets of force measuring units were installed at the connection between the forearm and wrist to collect data on changes in end load and the decreasing trend of gravitational acceleration, and to filter out high-frequency vibration disturbances to determine whether there is a tendency for rapid load release. The slippage precursor in the grip posture is time-series fused with the end load release trend. When the two are in the same direction and the time difference is less than 100 milliseconds, a load sudden unloading risk warning is generated and the initial slope limit value of the drive compensation curve is set. Based on the risk level of sudden load unloading, the rising slope of the original drive output curve is limited to restrict the rate of change of the output torque in the early stage. A slope limiting manipulator is superimposed in the output path to actively compress the rate of change of control error and reduce the reverse impact caused by inertial output.

3. The method for coordinated follow-up control of an upper limb exoskeleton system based on an active disturbance rejection control strategy according to claim 1, characterized in that, Step S2 includes: Based on the current joint position, joint angular velocity, and end-load estimate, a load reduction control trajectory is constructed from the current output state to the theoretical no-load state. Electromyographic signals are then used to dynamically adjust the endpoint of this trajectory to match the user's actual movement trend. The load reduction control trajectory is injected into the control process by switching the main control signal channel, and the sudden unloading identification signal is used as the switching trigger condition. The trajectory acceleration direction and change amplitude are kept continuous within the time window before and after the switching, so as to achieve a flexible transition of auxiliary output. During the execution of the load reduction control trajectory, the slope limiter dynamically compresses the expected acceleration change amplitude to ensure that the actual output does not exceed the maximum allowable acceleration threshold, thus preventing the risk of structural overshoot and reverse pull of the driver. After the load reduction control trajectory is completed and the electromyographic activity is confirmed to be below the unloading threshold and the end load is stable, the control signal channel switches back to the task auxiliary trajectory channel and gradually restores the normal operation auxiliary force output state using a slow recovery method.

4. The method for coordinated servo control of an upper limb exoskeleton system based on an active disturbance rejection control strategy according to claim 1, characterized in that, Step S3 includes: Based on the trajectory output direction, sudden unloading level and current joint motion state, it is determined whether there is an inertial impact trend. If so, the electromagnetic actuation structure set on both sides of the toothed ring of the joint shell is activated to trigger the toothed claw to bite the toothed ring to temporarily lock the joint degree of freedom. The viscoelastic energy buffer structure installed between the joint shell and the wearer's upper arm is activated simultaneously. It absorbs sudden inertial kinetic energy through the deformation of polyurethane and silicone composite materials, and achieves dynamic adaptation to different impact levels by adjusting the material compression stroke with the spiral limit ring. After electromagnetic locking is completed, the heating device inside the viscoelastic structure is triggered to raise the surface temperature of the polyurethane material to 40 degrees Celsius, reduce its elastic modulus, enhance its high-strength impact energy absorption capacity, and gradually cool down to restore its initial rigid state after the impact. The impact response is determined based on the integral value of the joint reverse acceleration and the viscoelastic elastic deformation. If the preset release condition is met, the electromagnetic lock is released and the drive output slope is suppressed to ensure structural stability and response continuity during the control signal recovery period.

5. The method for coordinated servo control of an upper limb exoskeleton system based on an active disturbance rejection control strategy according to claim 1, characterized in that, Step S4 includes: The elbow joint torque changes are collected in real time by a metal foil strain gauge set on the inner side of the elbow joint drive shaft, and the dynamic characteristics of the elbow joint entering the reverse acceleration trend are identified. Based on the recognition results, the shoulder joint is controlled to output a flexible support torque in the opposite direction to that of the elbow joint in a neutral posture. The output process limits the upward slope and torque amplitude, and a flexible constraint path between the elbow and shoulder is constructed. By constructing a spatial mechanical closed loop through the inertial release of the elbow joint and the flexible support of the shoulder joint, the residual impact kinetic energy is guided from the elbow joint to the wearer's shoulder strap and back structure, forming a multi-structure collaborative energy absorption mechanism.

6. The method for coordinated follow-up control of an upper limb exoskeleton system based on an active disturbance rejection control strategy according to claim 1, characterized in that, Step S5 includes: A short-time window acceleration limiting mechanism is activated. Acceleration sensing elements located at the elbow and shoulder joints detect high-frequency impact components and activate the current limiting unit to limit the rate of torque change, thereby quickly weakening high-frequency disturbances. After a short period of adjustment, a medium time window is entered to control the drive output to recover smoothly from the suppressed state, limiting the changes in acceleration slope and angular velocity to avoid discontinuous output caused by control rebound; During the long-term window phase, the upper limit of torque output is dynamically limited by evaluating the trend of total energy output per unit time to prevent the control channel from entering saturation. A unified control cycle synchronization mechanism is set up, which sequentially nests short, medium, and long time windows and sets a transition zone to achieve continuous switching and stable connection of control strategies during execution.

7. The method for coordinated servo control of an upper limb exoskeleton system based on an active disturbance rejection control strategy according to claim 1, characterized in that, Step S6 includes: The system compares the user's electromyography signals with the elbow and shoulder joint movement in real time. By judging the consistency of movement direction and acceleration trend, it identifies whether the drive output deviates from the user's true intention. When human-machine deviation is continuously detected, the signal gain of the disturbance observation channel is immediately reduced and its output influence is frozen. At the same time, the trajectory control mode guided by electromyography trend is activated to generate an intention-driven angular velocity control target. During trajectory control, execution errors are monitored in real time. If the deviation from the intended range continues, the target angle and speed boundary are adjusted to correct the trajectory and improve the following accuracy. During the trajectory correction phase, drive anti-saturation control is activated. If the detected output value exceeds the safety threshold or the rate of change increases abnormally, the amplitude control torque and output slope are immediately limited to ensure long-term stable and safe human-machine interaction.

Images