An exoskeleton upper limb robot joint motor optimization control method and system

Abstract

The application provides an exoskeleton upper limb robot joint motor optimization control method and system, relates to the technical field of robot control, and solves the technical problem that the trajectory tracking accuracy and motion smoothness are difficult to be considered under the existing variable load working condition. The method specifically comprises the following steps: obtaining current running state data and human-computer interaction state data of a joint motor of an exoskeleton upper limb robot; performing global optimization on the control parameters of the joint motor based on a multi-strategy collaborative optimization mechanism to obtain target control parameters; determining the fitness value of the candidate control parameters based on a fitness evaluation model to guide the optimization direction of the multi-strategy collaborative optimization mechanism; and generating a driving control signal of the joint motor based on the target control parameters and the current running state data. The application is used for controlling the joint motor of the exoskeleton upper limb robot.

Description

Technical Field

[0001] This application relates to the field of robot control technology, and in particular to an optimized control method and system for joint motors of an exoskeleton upper limb robot. Background Technology

[0002] In the application of exoskeleton upper limb rehabilitation robots, joint motors need to perform precise trajectory tracking tasks under complex human-machine interaction conditions such as time-varying patient limb impedance and switching between active and passive modes. Existing technologies typically use fixed-parameter PID control or conventional intelligent optimization algorithms for parameter tuning. However, when faced with nonlinear disturbances in human body load and the stringent safety requirements for motion smoothness in rehabilitation training, it is often difficult to simultaneously guarantee dynamic response speed and steady-state motion smoothness. This can easily lead to trajectory tracking overshoot or sudden changes in abrupt changes in motion, thereby causing patient discomfort or even secondary injury. Therefore, existing technologies suffer from the technical problem of being unable to simultaneously achieve both trajectory tracking accuracy and motion smoothness under variable load conditions. Summary of the Invention

[0003] This application provides a method and system for optimizing the control of joint motors in an exoskeleton upper limb robot, which solves the technical problem in the prior art that it is difficult to balance trajectory tracking accuracy and motion smoothness under variable load conditions.

[0004] To achieve the above objectives, this application adopts the following technical solution: Firstly, a method for optimizing the control of joint motors in an exoskeleton upper limb robot is provided, comprising: acquiring current operating state data and human-machine interaction state data of the joint motors of the exoskeleton upper limb robot; globally optimizing the control parameters of the joint motors based on a multi-strategy collaborative optimization mechanism to obtain target control parameters; wherein, the multi-strategy collaborative optimization mechanism includes a population collaboration strategy that divides the population into a first subpopulation focusing on dynamic response speed and a second subpopulation focusing on steady-state smoothness for parallel search, and a search balance strategy that adjusts inertia weights based on iteration progress; determining the fitness values ​​of candidate control parameters based on a fitness evaluation model to guide the optimization direction of the multi-strategy collaborative optimization mechanism; wherein, the fitness evaluation model includes a time-multiplied absolute error integral term and a safety penalty term for motion smoothness; and generating drive control signals for the joint motors based on the target control parameters and the current operating state data.

[0005] The above scheme decouples the optimization population into two subgroups, dynamic response and steady-state smoothness, according to the physical control objective, and combines them with a fitness evaluation model that incorporates safety constraints. This allows the optimization process of control parameters to directly serve the dual performance requirements in the rehabilitation scenario, thereby simultaneously improving tracking accuracy and motion compliance under variable load conditions.

[0006] In conjunction with the first aspect mentioned above, in one possible implementation, the control parameters of the joint motor are globally optimized based on a multi-strategy collaborative optimization mechanism to obtain the target control parameters. This includes: dividing the initial population into a first subpopulation and a second subpopulation; in each iteration, performing position updates on the first subpopulation and the second subpopulation respectively to obtain a first updated subpopulation and a second updated subpopulation; exchanging optimal individual information between the first updated subpopulation and the second updated subpopulation based on a preset information exchange mechanism to update the global optimal solution; and when a preset iteration termination condition is met, using the control parameters corresponding to the global optimal solution as the target control parameters.

[0007] In conjunction with the first aspect mentioned above, in one possible implementation, in each iteration, position updates are performed on the first subpopulation and the second subpopulation, including: introducing a mutation operator to perturb the current population centroid and generating a back-learning individual; determining the updated individual position based on the fitness values ​​of the back-learning individual and the current individual; and nonlinearly adjusting the inertia weight in the search balancing strategy based on the ratio of the current iteration number to the maximum iteration number.

[0008] In conjunction with the first aspect mentioned above, in one possible implementation, determining the fitness value of the candidate control parameters based on the fitness evaluation model includes: obtaining the joint angular velocity error sequence of the joint motor when the candidate control parameters are applied; calculating the time-per-absolute error integral term based on the joint angular velocity error sequence and the corresponding sampling time; obtaining the joint angular acceleration change rate of the joint motor when the candidate control parameters are applied; calculating a safety penalty term based on the excess amount when the joint angular acceleration change rate exceeds a preset rehabilitation safety threshold; and determining the fitness value based on the weighted sum of the time-per-absolute error integral term and the safety penalty term.

[0009] In conjunction with the first aspect mentioned above, in one possible implementation, the method further includes: introducing a torque penalty term for human-computer interaction torque into the fitness evaluation model; predicting the human-computer interaction torque for the next sampling period based on the current operating state data and candidate control parameters; when the predicted human-computer interaction torque exceeds a preset torque safety threshold, calculating the torque penalty term based on the torque excess, and incorporating the torque penalty term into the weighted calculation of the fitness value.

[0010] In conjunction with the first aspect mentioned above, in one possible implementation, a drive control signal for the joint motor is generated based on the target control parameters and the current operating state data, including: mapping the target control parameters to proportional gain, integral gain, and derivative gain; generating a basic drive signal based on the proportional gain, integral gain, derivative gain, and the state error between the current operating state data and the desired state data; and performing low-pass filtering on the derivative term in the basic drive signal to generate the drive control signal.

[0011] In conjunction with the first aspect mentioned above, in one possible implementation, low-pass filtering is performed on the differential term in the basic driving signal, including: obtaining the state error difference between the current sampling period and the previous sampling period; weighting and fusing the state error difference with a preset filtering time constant to obtain the filtered differential compensation amount; and using the filtered differential compensation amount to replace the original differential term in the basic driving signal.

[0012] In conjunction with the first aspect mentioned above, in one possible implementation, the method further includes: monitoring human-computer interaction status data in real time during rehabilitation training; when the human-computer interaction status data is detected to meet preset mode switching conditions, triggering a multi-strategy collaborative optimization mechanism to perform local re-optimization of control parameters in order to update the target control parameters.

[0013] In conjunction with the first aspect above, in one possible implementation, the human-computer interaction state data includes interaction torque data and surface electromyography (EMG) signals; the mode switching conditions include at least one of the following: the detected interaction torque data exceeds the passive training torque threshold; the amplitude of the collected EMG signals exceeds the active intention trigger threshold.

[0014] Secondly, a joint motor optimization control system for an exoskeleton upper limb robot is provided, comprising: a state acquisition module for acquiring current operating state data and human-machine interaction state data of the joint motors of the exoskeleton upper limb robot; a collaborative optimization module for globally optimizing the control parameters of the joint motors based on a multi-strategy collaborative optimization mechanism to obtain target control parameters; wherein, the multi-strategy collaborative optimization mechanism includes a population collaboration strategy of dividing the population into a first subpopulation focusing on dynamic response speed and a second subpopulation focusing on steady-state smoothness for parallel search, and a search balance strategy of adjusting inertia weights based on iteration progress; a fitness evaluation module for determining the fitness values ​​of candidate control parameters based on a fitness evaluation model to guide the optimization direction of the multi-strategy collaborative optimization mechanism; wherein, the fitness evaluation model includes a time multiplication absolute error integral term and a safety penalty term for motion smoothness; and a drive control module for generating drive control signals for the joint motors based on the target control parameters and current operating state data.

[0015] This application decouples the population in a multi-strategy collaborative optimization mechanism from physical objectives based on dynamic response speed and steady-state smoothness. It then combines this with a fitness evaluation model that integrates the time-time multiplied by the absolute error integral and the motion smoothness safety penalty term. This deeply couples the control parameter optimization process with the variable load characteristics and safety constraints of the exoskeleton rehabilitation scenario. The design principle transforms abstract algorithmic optimization into targeted control of specific physical performance indicators. This ensures trajectory tracking accuracy while suppressing sudden changes in abrupt changes in motion and abnormal interaction forces, achieving high-precision, low-overshoot, and compliant and safe joint motor control, effectively improving the effectiveness of rehabilitation training and patient comfort.

[0016] It should be understood that the descriptions of technical features, technical solutions, beneficial effects, or similar language in this application do not imply that all features and advantages can be achieved in any single embodiment. Rather, it is understood that the description of a feature or beneficial effect means that a specific technical feature, technical solution, or beneficial effect is included in at least one embodiment. Therefore, the descriptions of technical features, technical solutions, or beneficial effects in this specification do not necessarily refer to the same embodiment. Furthermore, the technical features, technical solutions, and beneficial effects described in this embodiment can be combined in any suitable manner. Those skilled in the art will understand that embodiments can be implemented without one or more specific technical features, technical solutions, or beneficial effects of a particular embodiment. In other embodiments, additional technical features and beneficial effects may be found in specific embodiments that do not include all the features of the embodiments. Attached Figure Description

[0017] Figure 1 A flowchart illustrating an optimized control method for joint motors of an exoskeleton upper limb robot provided in this application embodiment; Figure 2 A system architecture diagram of an upper limb rehabilitation robot hardware system provided in this application embodiment; Figure 3 A flowchart illustrating another method for optimizing the joint motor control of an exoskeleton upper limb robot provided in this application embodiment; Figure 4 A flowchart illustrating another method for optimizing the joint motor control of an exoskeleton upper limb robot provided in this application embodiment; Figure 5 A flowchart illustrating another method for optimizing the joint motor control of an exoskeleton upper limb robot provided in this application embodiment; Figure 6 This application provides a system architecture diagram for an optimized control system for joint motors of an exoskeleton upper limb robot. Detailed Implementation

[0018] In the description of this application, unless otherwise stated, " / " means "or," for example, A / B can mean A or B. The "and / or" in this document is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, and B alone. Furthermore, "at least one" means one or more, and "multiple" means two or more. The terms "first," "second," etc., do not limit the quantity or order of execution, and "first," "second," etc., do not necessarily imply differences.

[0019] It should be noted that, in this application, the terms "exemplary" or "for example" are used to indicate that something is being described as an example, illustration, or illustration. Any embodiment or design described as "exemplary" or "for example" in this application should not be construed as being more preferred or advantageous than other embodiments or design solutions. Specifically, the use of terms such as "exemplary" or "for example" is intended to present the relevant concepts in a concrete manner.

[0020] Example 1: like Figure 1 As shown in the figure, this application provides an optimized control method for joint motors of an exoskeleton upper limb robot. This method is applied to the control system of the exoskeleton upper limb robot and aims to solve the problem of balancing trajectory tracking accuracy and motion smoothness caused by time-varying human load during rehabilitation training. The method mainly includes the following steps: S101. Obtain the current operating status data of the joint motors of the exoskeleton upper limb robot and the human-computer interaction status data.

[0021] Among them, the current operating status data refers to the physical quantity that can characterize the real-time kinematic characteristics of the joint motor, while the human-computer interaction status data is a set of signals that reflect the mechanical coupling and physiological intentions between the patient and the robot.

[0022] In the embodiments of this application, combined with Figure 2 The hardware system architecture shown uses angle sensors to collect joint angles and angular velocities in real time as current operating status data. Simultaneously, it utilizes wireless physiological data acquisition devices to obtain surface electromyography (EMG) signals and combines this with a three-dimensional force sensor to collect human-machine interaction torque as human-machine interaction status data. The raw signals collected by these sensors are transmitted to the controller via a wireless module or CAN bus, where they undergo preprocessing operations such as digital filtering and normalization to eliminate the influence of environmental noise and electromagnetic interference on subsequent control decisions.

[0023] Based on the above steps, high-precision, low-latency multimodal data sensing provides a reliable physical feedback benchmark for subsequent adaptive parameter optimization, ensuring that the control strategy can respond promptly to minute changes in the patient's limb status.

[0024] S102. Based on the multi-strategy collaborative optimization mechanism, the control parameters of the joint motor are globally optimized to obtain the target control parameters.

[0025] Among them, the multi-strategy collaborative optimization mechanism includes a population collaboration strategy that divides the population into a first subpopulation that focuses on dynamic response speed and a second subpopulation that focuses on steady-state smoothness for parallel search, as well as a search balancing strategy that adjusts inertia weight based on iteration progress. The multi-strategy collaborative optimization mechanism is a parameter tuning strategy for exoskeleton variable load conditions. Its core is to meet the dual requirements of rapid response and smooth movement in rehabilitation training through functionally decoupled population collaboration.

[0026] In this embodiment, the mechanism encodes the control parameters to be optimized (such as PID gain) into individuals and initializes two subpopulations with different search preferences. The first subpopulation is configured to prioritize exploring parameter regions that enable the system to quickly track sudden changes in patient intent, focusing on improving dynamic response speed; the second subpopulation is guided to the parameter space that can suppress mechanical jitter and overshoot during long-term training, focusing on ensuring steady-state smoothness. Both subpopulations perform position updates in parallel during the iteration process and dynamically adjust the inertia weights according to the current optimization progress to achieve a balance between broadly exploring the feasible region and finely developing local optima.

[0027] Based on the above steps, adaptive tuning of control parameters in complex human-computer interaction environments is achieved, enabling the joint motor to not only sensitively follow the patient's movement intentions but also maintain a smooth and compliant output characteristic, effectively improving the safety and comfort of rehabilitation training.

[0028] S103. Determine the fitness values ​​of candidate control parameters based on the fitness evaluation model to guide the optimization direction of the multi-strategy collaborative optimization mechanism.

[0029] The fitness evaluation model includes a time-to-absolute-error integral term and a safety penalty term for motion smoothness. The fitness evaluation model is a bridge connecting numerical optimization algorithms and physical world safety constraints, used to quantify the comprehensive performance of each set of candidate control parameters in actual or simulation operation.

[0030] In this embodiment, the model consists of two weighted components: the time-to-absolute-error integral term is used to measure the dynamic tracking quality of the system on the desired trajectory. The smaller the integral value, the faster the response and the smaller the steady-state error. The safety penalty term is directly related to safety constraints. For example, when the rate of change of joint angular acceleration (jumping) exceeds the preset human tolerance threshold, the value of this term will increase sharply, thereby significantly reducing the fitness score of the corresponding parameter and forcing the optimization process to stay away from the dangerous parameter area that may cause joint impact.

[0031] Based on the above steps, a comprehensive evaluation system that takes into account both tracking accuracy and physiological safety was constructed, ensuring that the final output control parameters are not only theoretically optimal, but also safe and reliable in application.

[0032] S104. Based on the target control parameters and current operating status data, generate the drive control signal for the joint motor.

[0033] Among them, generating the drive control signal for the joint motor is the key step in transforming the abstract optimization result into physical instructions that can be executed by the hardware, thus completing the closed loop from the algorithm layer to the execution layer.

[0034] In this embodiment, the controller maps the optimized target control parameters to the proportional, integral, and derivative gain register values ​​of the underlying driver in real time, and calculates the basic drive quantity using a PID control law, combined with the currently acquired joint state error. Subsequently, this drive quantity undergoes digital-to-analog conversion or pulse width modulation processing to generate the voltage or PWM signal required for power amplification, and is sent to the corresponding joint drive device via the CAN bus to drive the motor to output precise auxiliary torque.

[0035] Based on the above steps, a complete closed-loop control from perception and decision-making to execution is achieved, effectively transforming the results of intelligent optimization algorithms into a precise and compliant rehabilitation assistance experience that patients can perceive.

[0036] Based on the above technical solution, the multi-strategy collaborative optimization mechanism and the fitness evaluation model with integrated safety constraints are deeply embedded in the control loop, which effectively solves the technical problem of balancing trajectory tracking accuracy and motion smoothness in exoskeleton upper limb robots under variable load human-computer interaction conditions, and significantly improves the safety and effectiveness of rehabilitation training.

[0037] Example 2: In one possible implementation of the embodiments of this application, combined with Figure 1 ,like Figure 3 As shown, the above S102 can be specifically implemented through the following S301, S302, S303 and S304, which are explained in detail below: S301. Divide the initial population into a first subpopulation and a second subpopulation.

[0038] The first subpopulation is configured to focus on optimizing dynamic response speed, aiming to enable the joint motors to quickly track changes in the patient's movement intentions; the second subpopulation is configured to focus on optimizing steady-state smoothness, aiming to suppress mechanical jitter or overshoot that may occur during long-term rehabilitation training.

[0039] In this embodiment, the initialization process generates two groups of individuals according to a preset ratio, based on the differentiated performance requirements of rehabilitation training at different stages. For example, the total population size is set to 30, with the first subpopulation containing 16 individuals and the second subpopulation containing 14 individuals. The individuals in the first subpopulation are initially guided to explore regions with high gain ratios to prioritize the transient response capability of the system; the individuals in the second subpopulation are initially guided to a combination region of low integral gain and appropriate differential gain to prioritize the smoothness of motion. This population partitioning strategy based on physical control objectives gives the algorithm a clear functional orientation from the initial search stage.

[0040] It should be noted that the proportion of the subgroup is not fixed, but can be adjusted according to the specific type of rehabilitation task. For example, in passive training for patients with spasticity, the proportion of the second subgroup can be appropriately increased to enhance safety; while in active assisted training for the muscle strength recovery period, the proportion of the first subgroup can be increased to improve following sensitivity.

[0041] As an example, in the initialization phase of elbow flexion rehabilitation training, the 16 individuals of the first subpopulation are randomly distributed in the high-gain interval Kp∈[50,100], and the 14 individuals of the second subpopulation are randomly distributed in the low-integral interval Ki∈[0,20], thus constructing an initial search space that covers both dynamic and steady-state performance requirements.

[0042] Based on the above steps, the population initialization with functional decoupling avoids the problem of easily falling into local optimum traps under a single optimization objective, laying a diversified foundation for subsequent parallel search.

[0043] S302. In each iteration, position updates are performed on the first subpopulation and the second subpopulation respectively to obtain the first updated subpopulation and the second updated subpopulation.

[0044] The location update process integrates a reverse learning strategy and an adaptive inertial weight adjustment mechanism to balance global exploration and local exploitation capabilities while maintaining population diversity.

[0045] In this embodiment, for each subpopulation, a mutation operator is first introduced to perturb the current population's centroid, generating back-learning individuals. Based on the fitness values ​​of the back-learning individuals and the current individuals, the updated individual positions are determined, thus escaping local extreme value regions. Simultaneously, the inertia weight in the search balancing strategy is nonlinearly adjusted based on the ratio of the current iteration count to the maximum iteration count. Specifically, a larger inertia weight is assigned in the early stages of iteration, enabling the first subpopulation to conduct global exploration in a broad parameter space with large step sizes, quickly locating potential advantageous regions with response times less than 0.1s. As iterations progress, the inertia weight decreases nonlinearly, prompting the second subpopulation to switch to a small-step, refined development mode, further exploring precise parameter combinations that suppress jitter within the discovered high-quality regions. The entire update process is guided by fitness values, ensuring that the search direction always points to a physical parameter space with superior overall performance.

[0046] It should be noted that the generation of back-learned individuals is not a simple mathematical transformation, but rather incorporates constraints from the physical boundaries of the parameters. When the parameters generated by back-learning exceed the safety limits of the motor driver, they are automatically truncated to the boundary values ​​to prevent invalid calculations. Furthermore, the adjustment curve for the nonlinear inertia weights can be calibrated based on actual convergence to adapt to the dynamic characteristics of motors with different joints.

[0047] As an example, in the 20th iteration, the first subpopulation perturbs the centroid using the Cauchy mutation operator, generating several reverse-learning individuals located in the combination region of high Kp and low Kd. After fitness evaluation, three new solutions that are better than the current individuals are retained. At the same time, the inertia weight is nonlinearly reduced from 0.9 to 0.6, guiding the second subpopulation to perform a refined search in the narrow band of Ki∈[5,15].

[0048] Based on the above steps, the simultaneous optimization of dynamic response and steady-state smoothness is achieved, which not only ensures the algorithm's global search capability in complex parameter space, but also improves the later convergence accuracy, effectively overcoming the shortcomings of traditional algorithms that are prone to premature convergence.

[0049] S303. Based on a preset information exchange mechanism, the optimal individual information is exchanged between the first update subpopulation and the second update subpopulation to update the global optimal solution.

[0050] Among them, the information exchange mechanism refers to the two subpopulations sharing their respective optimal parameter configurations according to predetermined rules during independent search processes, thereby achieving complementary advantages.

[0051] In this embodiment, a periodic elite migration strategy is employed as the specific implementation of information exchange. At preset generation intervals, the fitness values ​​of the current best individuals in the two subpopulations are compared. If the first subpopulation discovers a solution with higher fitness, it is synchronized to the second subpopulation to replace its current worst individual, and vice versa. This exchange frequency can be dynamically reduced with the iteration progress; for example, exchanging every 10 generations in the early stages to promote rapid information dissemination, and exchanging every 20 generations in the later stages to maintain the independent development capabilities of each subpopulation and avoid premature convergence leading to loss of diversity.

[0052] It should be noted that information exchange is not an indiscriminate replacement of the entire population, but rather a targeted migration only to elite individuals. This ensures that high-quality genes can spread throughout the entire population, accelerating global convergence, while preserving the original search rules of each subpopulation, preventing the loss of optimization capabilities for specific physical indicators due to excessive homogenization.

[0053] As an example, at the end of the 30th generation iteration, the first subpopulation found a parameter solution with a response time of only 0.08s but slightly poor smoothness. After this solution was transferred to the second subpopulation, it inspired the second subpopulation to search for a new set of solutions with a response time of 0.09s and significantly improved smoothness in its neighborhood, thereby updating the global optimal record.

[0054] Based on the above steps, controlled information interaction breaks down information silos between subpopulations, effectively accelerating the discovery of the global optimum while maintaining population diversity, and ensuring that the final parameters have both the excellent characteristics of rapid response and smooth movement.

[0055] S304. When the preset iteration termination condition is met, the control parameters corresponding to the global optimal solution are used as the target control parameters.

[0056] The iteration termination condition can be reaching the preset maximum number of iterations, or the global optimal fitness value not changing significantly for several consecutive generations, or the fitness value being lower than the preset performance threshold.

[0057] In this embodiment, when any of the above termination conditions are met, the system decodes the currently recorded globally optimal individual into a triplet of proportional gain, integral gain, and derivative gain. This triplet is the final output target control parameter. For example, if the maximum number of iterations is set to 80, the search automatically stops and the current optimal PID parameter combination is output when the algorithm reaches the 80th generation or the fitness value stabilizes below 0.05 for more than 10 generations. This parameter directly reflects the optimal control configuration under the dual constraints of dynamic response speed and steady-state smoothness.

[0058] It should be noted that the output target control parameters must undergo hardware safety verification before being loaded into the underlying driver to ensure that each gain value is within the safe operating range of the motor and reducer, preventing equipment damage caused by the algorithm optimization result exceeding physical limits.

[0059] Based on the above steps, the abstract optimization process is transformed into precise control parameters that can be directly applied to the physical system, completing the final closed loop from algorithm optimization to engineering implementation. This ensures that the exoskeleton upper limb robot can achieve high-precision, low-overshoot, and compliant and safe trajectory tracking control in actual rehabilitation training.

[0060] Example 3: In one possible implementation of the embodiments of this application, combined with Figure 1 ,like Figure 4 As shown, the above S103 can be specifically implemented through the following S401 to S404, which are explained in detail below: S401. Obtain the joint angular velocity error sequence of the joint motor when applying candidate control parameters.

[0061] The joint angular velocity error sequence refers to the set of absolute values ​​of the difference between the expected angular velocity and the actual angular velocity recorded at a fixed sampling period within a single evaluation duration. The sequence length is equal to the evaluation duration divided by the sampling period.

[0062] In this embodiment, during simulation or actual testing, the controller compares the expected angular velocity generated by trajectory planning with the actual angular velocity fed back by the angle sensor in real time, calculates the absolute value of the difference between the two, and stores this absolute value along with the corresponding timestamp. For example, within a single evaluation window set at 3 seconds, 3000 error data points are continuously collected at a sampling frequency of 1kHz, forming a complete angular velocity error sequence. This sequence not only contains steady-state error information but also completely records the transient deviation distribution throughout the entire process of starting, constant speed, and braking.

[0063] It is important to note that the acquisition of the error sequence must ensure strict synchronization between the expected signal and the feedback signal on the time axis to avoid distortion in error calculation due to communication delays or clock drift. Furthermore, the sequence length should cover at least one complete rehabilitation training cycle to ensure that the evaluation results reflect the dynamic performance throughout the entire process.

[0064] Based on the above steps, by constructing a high temporal resolution error sequence, a raw data foundation is provided for the subsequent accurate quantification of the time-cumulative effect of dynamic tracking deviation, avoiding the one-sidedness of evaluation caused by relying solely on error values ​​at a single moment.

[0065] S402. Based on the joint angular velocity error sequence and the corresponding sampling time, calculate the integral term of time multiplied by absolute error.

[0066] Among them, the time-to-absolute-error integral term is the numerical integral of the product of the absolute value of the angular velocity error and time, which is used to comprehensively characterize the overall quality of the system in terms of response speed and steady-state accuracy.

[0067] In this embodiment, the trapezoidal integral method or Simpson's integral method is used to process the angular velocity error sequence. Specifically, the absolute value of the error at each sampling point is multiplied by the corresponding timestamp to obtain a weighted error value, and then all weighted error values ​​are summed. This integration mechanism assigns higher weight to later errors, meaning that if the system still has a large deviation in the later stages of adjustment, it will be subject to a more severe penalty, thereby forcing the optimization algorithm to prioritize the elimination of steady-state errors and shorten the adjustment time. For example, when the calculated ITAE value (time multiplied by the absolute error integral term) is 0.05, it indicates that the system has excellent fast tracking and zero steady-state error characteristics.

[0068] As an example, in the evaluation of a certain set of candidate parameters, although its maximum overshoot was small, the error persisted in the second half due to the long settling time, and the final calculated ITAE value was as high as 0.18, which was judged as a suboptimal solution; while another set of parameters had a slight overshoot but converged very quickly, with an ITAE value of only 0.04, which was judged as a high-quality solution.

[0069] Based on the above steps, the complex dynamic response process is compressed into a single scalar index by using a time-weighted integral mechanism, which effectively guides the optimization algorithm to pursue fast and zero steady-state error tracking performance, meeting the requirements of rehabilitation training for the timeliness of movements.

[0070] S403. Obtain the joint angular acceleration change rate of the joint motor when applying candidate control parameters. When the joint angular acceleration change rate exceeds the preset rehabilitation safety threshold, calculate the safety penalty item based on the excess amount.

[0071] Among them, the joint angular acceleration change rate, or jerkiness, characterizes the smoothness of the motor output torque; the safety penalty is the cost imposed when the jerkiness exceeds the human body's tolerance limit, used to forcibly constrain the compliance of movement.

[0072] In this embodiment, angular acceleration is obtained by performing a second difference on the angular velocity error sequence or by directly differentiating the encoder data, and then performing a first difference to obtain the rate of change of angular acceleration. When the rate of change is detected to exceed a preset recovery safety threshold, a penalty value is accumulated according to the amount exceeded. Optionally, a safety penalty item... Satisfy the following formula:

[0073] in, To measure the rate of change of joint angular acceleration, The preset rehabilitation safety threshold, The penalty coefficient is... The penalty order is usually 1 or 2.

[0074] It should be noted that the introduction of a safety penalty term changes the traditional optimization approach's single-minded focus on tracking accuracy, transforming "physiological safety" into a mathematical constraint understandable to the algorithm. The order of the penalty function... The approach can be adjusted according to the recovery stage. Higher-order penalties can be used in the acute phase to suppress the risk of exceeding limits, while linear penalties can be used in the recovery phase to allow for a moderate dynamic response.

[0075] Based on the above steps, by explicitly embedding the rehabilitation safety threshold into the evaluation system, the combination of high-amplitude parameters that may cause secondary injury to patients is avoided, ensuring the inherent safety of the optimization results in clinical applications.

[0076] S404. Determine the fitness value based on the weighted sum of the time multiplication absolute error integral term and the safety penalty term.

[0077] Among them, the fitness value is the final scalar basis for guiding the search direction of the multi-strategy collaborative optimization mechanism. Its composition may also include a torque penalty term for human-computer interaction torque, so as to achieve a comprehensive trade-off of multiple physical objectives.

[0078] In this embodiment, the total fitness value is obtained by weighted summation of the ITAE term, the safety penalty term, and an optional torque penalty term. For the torque penalty term, based on the current operating state data and candidate control parameters, the human-machine interaction torque for the next sampling period is predicted using a dynamic model. When the predicted value exceeds a preset torque safety threshold, this term is calculated based on the excess torque and included in the sum. The weights of each sub-term are not fixed but dynamically adjusted according to the rehabilitation training stage.

[0079] Based on the above steps, a multi-dimensional evaluation system integrating tracking accuracy, motion smoothness, and human-computer interaction safety was constructed. Through a dynamic weighting mechanism, the physical interpretability and scenario adaptability of multi-objective optimization were achieved, ensuring that the final output control parameters achieve the best balance among various rehabilitation indicators.

[0080] Example 4: In one possible implementation of the embodiments of this application, combined with Figure 1 ,like Figure 5 As shown, the above S104 can be specifically implemented through the following S501 to S504, which are explained in detail below: S501, Map the target control parameters to proportional gain, integral gain and derivative gain.

[0081] Mapping refers to the process of converting the dimensionless or normalized optimal solution output by the multi-strategy collaborative optimization mechanism into physical register values ​​that can be recognized by the underlying motor driver.

[0082] In this embodiment, after receiving the target control parameters, the controller assigns them to the proportional gain Kp, integral gain Ki, and derivative gain Kd registers of the PID controller through linear transformation or a preset lookup table method. To ensure hardware safety, the mapping process incorporates hardware limiting protection logic, which automatically truncates the optimized parameter values ​​to boundary values ​​when they exceed the safe operating range allowed by the motor driver.

[0083] Based on the above steps, a seamless conversion from the abstract optimization space to specific physical control parameters is achieved, and the inherent safety of the system during the adaptive tuning process is ensured through a hardware-level limiting mechanism.

[0084] S502 generates a basic drive signal based on the proportional gain, integral gain, derivative gain, and the state error between the current operating state data and the desired state data.

[0085] The basic drive signal is the raw PID control output without filtering, which is a linear superposition of the proportional term, integral term and the original derivative term.

[0086] In this embodiment, the controller calculates the state error e between the desired trajectory and the actual feedback in real time during each sampling period and generates the basic driving quantity according to the classic PID control law. Specifically, the proportional term is obtained by multiplying Kp by the current error and is used for rapid response to deviation; the integral term is obtained by multiplying Ki by the time accumulation of the error and adopts an anti-integral saturation design, which automatically stops the integral accumulation when the output reaches the limit to prevent excessive overshoot during system recovery; the derivative term is obtained by multiplying Kd by the rate of change of the error and is used to predict trends and provide damping. However, since the sensor signals of the rehabilitation robot inevitably contain high-frequency measurement noise, and the derivative component inherently has high-pass filtering characteristics, directly using the original derivative term will significantly amplify these high-frequency noises, resulting in audible noise or high-frequency jitter in the motor.

[0087] It should be noted that although the basic drive signal can theoretically provide correction torque, in practical engineering applications, if the differential components are not processed, the noise amplification effect will severely reduce the patient's wearing comfort and may even induce muscle spasms. Therefore, the signal generated in this step is only an intermediate quantity and must be further conditioned before it can be used as the final command.

[0088] Based on the above steps, a preliminary control quantity containing complete correction information was constructed. At the same time, the inherent defects of the original differential term in noise-sensitive environments were exposed, providing a clear engineering requirement guide for the introduction of targeted filtering processing.

[0089] S503. Perform low-pass filtering on the differential term in the basic drive signal to generate the drive control signal.

[0090] Low-pass filtering aims to preserve the corrective effect of the differential term on the effective motion trend at low frequencies, while attenuating its amplification effect on high-frequency sensor noise. It is a key step in balancing dynamic response and control stability.

[0091] In this embodiment, a first-order inertial filtering algorithm in the discrete domain is used to reconstruct the differential term. Specifically, the difference in state error between the current sampling period and the previous sampling period is first obtained. Then, the difference is compared with the preset filter time constant. Weighted fusion is performed to obtain the filtered differential compensation amount. Finally, this compensation amount is used to replace the original differential term in the basic drive signal. Optionally, the filtered differential compensation amount satisfies the following formula:

[0092] in, This is the differential compensation amount after filtering at the current moment. This represents the state error difference at the current moment. The sampling period is This is the filtered output from the previous time step. Let be the filter coefficients and satisfy... , This is the preset filtering time constant.

[0093] It should be noted that the filtering time constant The choice of [aspect name] directly determines the system's bandwidth and smoothness. If [aspect name] is selected... If the value is too large, although it can completely filter out noise, it will introduce significant phase lag, weakening the differential term's ability to suppress sudden disturbances; if... If the size is too small, the noise reduction effect will be insignificant. In the rehabilitation robot application of this application, The optimal value range is 0.001s to 0.01s, so as to preserve the effective frequency band of human motion (usually below 10Hz) while sufficiently attenuating power frequency interference above 50Hz and sensor quantization noise. This discrete implementation based on differential fusion avoids the numerical instability problems that may be caused by continuous domain transfer functions in digital controllers, and has higher engineering robustness.

[0094] Based on the above steps, without sacrificing dynamic correction performance, the high-frequency jitter and noise of the actuator are effectively eliminated, significantly improving the motion compliance and patient comfort during rehabilitation training, and solving the problem of "precision and smoothness being difficult to achieve simultaneously" in the application of traditional PID control in high-precision rehabilitation robots.

[0095] S504 outputs the final drive signal to the joint motor driver.

[0096] The final drive signal is a physical execution command that has been filtered and converted from digital to analog or pulse width modulation, and it directly determines the torque output characteristics of the joint motor.

[0097] In this embodiment, the drive control signal, after low-pass filtering, is sent to the controller's DAC module or PWM generation unit, where it is converted into an analog voltage signal or a square wave signal with an adjustable duty cycle. Before output, the signal undergoes bus voltage clamping to ensure its amplitude never exceeds the rated supply voltage, preventing overvoltage drive. Subsequently, the signal is sent to the shoulder joint drive device or elbow joint drive device via the CAN bus or a dedicated drive interface, causing the drive motor to operate smoothly according to the expected torque curve.

[0098] Based on the above steps, the final transformation from digital control algorithm to physical mechanical motion was completed, realizing high-precision, low-noise, and stable drive of the exoskeleton upper limb robot joint motor in complex rehabilitation scenarios.

[0099] Example 5: In one possible implementation of this application, to adapt to the dynamic changes in the patient's state during rehabilitation training, the exoskeleton upper limb robot joint motor optimization control method further includes a mode switching triggering and parameter adaptive adjustment mechanism. Specifically, during rehabilitation training, human-computer interaction status data is monitored in real time.

[0100] Among them, human-computer interaction status data are direct physical quantities that reflect the patient's movement intentions and limb force conditions, mainly including interactive torque data and surface electromyography signals.

[0101] In this embodiment, a three-dimensional force sensor installed at the end of the robotic arm continuously collects the human-machine interaction torque, while a wireless physiological data acquisition device simultaneously acquires the surface electromyographic signals of the target muscle group. To ensure the accuracy of state recognition, the acquired raw signals need to be smoothed through a sliding window to eliminate transient interference.

[0102] Based on the above steps, a high-fidelity real-time perception of the patient's physiological state is achieved, providing a reliable data foundation for accurately identifying rehabilitation mode transition points.

[0103] When the human-computer interaction status data is detected to meet the preset mode switching conditions, a multi-strategy collaborative optimization mechanism is triggered to perform local re-optimization of the control parameters in order to update the target control parameters.

[0104] Among them, the mode switching condition is a physical criterion for determining whether a patient has shifted from passively receiving training to actively participating or exhibiting an abnormal antagonistic state. It includes at least one of the following: the detected interactive torque data exceeds the passive training torque threshold, or the amplitude of the collected surface electromyography signal exceeds the active intention trigger threshold.

[0105] In this embodiment, the passive training torque threshold is set according to clinical safety standards, for example, 5 N·m, to prevent the robotic arm from excessively stretching the patient's limbs; the active intention trigger threshold is calibrated based on the patient's muscle strength assessment results, for example, 20 μV RMS, to capture weak active movement attempts. Once either condition is met, the system determines that the current control parameters are no longer suitable for the new state and immediately triggers local re-optimization. This process does not restart the global search, but rather reduces the population size and search radius within the neighborhood of the current optimal solution, only fine-tuning the proportional gain and integral gain to ensure that the re-optimization time is controlled within 50 ms, meeting the real-time requirements of rehabilitation control.

[0106] Based on the above steps, the control system achieves adaptive and seamless transitions between rehabilitation stages, ensuring safety during passive training and providing timely and appropriate assistance when the patient actively exerts effort, significantly improving the human-machine integration experience.

[0107] After completing the local re-optimization, the system immediately applies the newly obtained target control parameters and overwrites the old values, thus completing the parameter update.

[0108] Among them, parameter update refers to the process of writing the optimized PID gain into the underlying driver, and the smoothness of the switching process must be ensured.

[0109] In this embodiment, the new parameters undergo a first-order low-pass smoothing process before being loaded into the shoulder or elbow joint drive device to prevent sudden changes in motor torque caused by abrupt parameter variations. Simultaneously, the system automatically records the timestamp, trigger reason, and parameter change log for each mode switch, providing objective data support for subsequent clinical efficacy evaluation. For example, within 0.1 seconds of detecting a mode switch, the controller gradually transitions Kp from the old value to the new value over 10 cycles, ensuring continuous and shock-free output torque.

[0110] Based on the above steps, the physical execution safety and continuity of the control parameters during the dynamic adjustment process are ensured, enabling the exoskeleton upper limb robot to maintain a high degree of adaptation to the patient's condition throughout the entire rehabilitation cycle.

[0111] Example 6: like Figure 6As shown in the figure, this application embodiment also provides an exoskeleton upper limb robot joint motor optimization control system, including a state acquisition module 601, a collaborative optimization module 602, an fitness evaluation module 603, and a drive control module 604.

[0112] The status acquisition module 601 is used to acquire the current operating status data of the joint motors of the exoskeleton upper limb robot and the human-computer interaction status data.

[0113] In this embodiment, the state acquisition module 601 establishes a physical communication connection with the angle sensor, the wireless physiological data acquisition device, and the wireless module. Specifically, this module reads the joint angle and angular velocity signals acquired by the angle sensor through a serial port or a dedicated I / O interface, and simultaneously receives surface electromyography signals and three-dimensional force sensor data transmitted by the wireless physiological data acquisition device via the wireless module. To ensure the efficiency of subsequent processing, the module also integrates a data preprocessing unit, which converts the multi-source heterogeneous raw signals into structured data frames aligned with a unified timestamp and outputs them to the collaborative optimization module. This structured data frame typically includes a frame header, timestamp, sensor values ​​for each channel, and check bits, facilitating rapid parsing and retrieval by downstream modules.

[0114] The collaborative optimization module 602 is used to perform global optimization of the control parameters of the joint motor based on a multi-strategy collaborative optimization mechanism to obtain the target control parameters. The multi-strategy collaborative optimization mechanism includes a population collaboration strategy that divides the population into a first subpopulation that focuses on dynamic response speed and a second subpopulation that focuses on steady-state smoothness for parallel search, as well as a search balancing strategy that adjusts the inertia weight based on the iteration progress.

[0115] In this embodiment, the collaborative optimization module 602 is deployed within the core processing unit of the controller and serves as the intelligent decision-making center of the system. This module receives real-time status data and human-computer interaction data from the status acquisition module and internally instantiates and runs the aforementioned multi-strategy collaborative optimization mechanism. Specifically, this module maintains two independent population buffers in memory, storing the individual parameter vectors of the first and second sub-populations respectively, and performs position updates and information exchange operations in parallel according to a preset iterative logic. After each iteration, the module outputs the currently found optimal control parameters (such as the Kp, Ki, Kd triplet) to the drive control module, and dynamically adjusts the search strategy for the next round based on the evaluation results fed back by the fitness evaluation module.

[0116] The fitness evaluation module 603 is used to determine the fitness values ​​of candidate control parameters based on the fitness evaluation model, so as to guide the optimization direction of the multi-strategy collaborative optimization mechanism; wherein, the fitness evaluation model includes a time multiplication absolute error integral term and a safety penalty term for motion smoothness.

[0117] In this embodiment, the fitness evaluation module 603 is logically embedded within or tightly coupled to the collaborative optimization module 602, acting as a "value judge" in the optimization process. This module receives candidate control parameters generated by the collaborative optimization module and, in conjunction with measured or simulated feedback data provided by the state acquisition module 601, calculates the comprehensive fitness value for each set of candidate parameters. Specifically, the module integrates an ITAE calculator and a safety penalty calculator. The former performs time-weighted integration on the angular velocity error sequence, while the latter monitors the rate of change of joint angular acceleration and the predicted value of human-machine interaction torque in real time, automatically accumulating penalty terms when exceeding limits is detected. The final fitness value is fed back in real time to the optimization engine of the collaborative optimization module 602, serving as the sole basis for individual performance ranking and global optimal update.

[0118] The drive control module 604 is used to generate drive control signals for the joint motor based on the target control parameters and the current operating status data.

[0119] In this embodiment, the drive control module 604 serves as a bridge connecting the intelligent decision-making layer and the physical execution layer. Its physical carrier includes a PID calculation unit, a low-pass filter unit, and a CAN communication interface within the controller. This module receives the target control parameters output by the collaborative optimization module 602 in real time and loads them into the gain register of the PID controller. Subsequently, based on the current state error and the updated gain parameters, the module calculates the basic drive quantity and performs noise suppression processing on the differential term using a built-in low-pass filter, ultimately generating a smooth PWM or voltage command. This command is sent to the shoulder joint drive device or elbow joint drive device via the CAN bus, driving the motor to output precise auxiliary torque.

[0120] Although this application has been described herein in conjunction with various embodiments, those skilled in the art can understand and implement other variations of the disclosed embodiments by reviewing the accompanying drawings, the disclosure, and the appended claims when implementing the technical solutions claimed herein. In the claims, the word "comprising" does not exclude other components or steps, and "a" or "an" does not exclude multiple components. A single processor or other unit can implement several functions listed in the claims. While different dependent claims may describe certain measures, this does not mean that these measures cannot be combined to produce good results.

[0121] Although this application has been described in conjunction with specific features and embodiments, it is obvious that various modifications and combinations can be made thereto without departing from the spirit and scope of this application. Accordingly, this specification and drawings are merely exemplary illustrations of this application as defined by the appended claims, and are considered to cover any and all modifications, variations, combinations, or equivalents within the scope of this application. Clearly, those skilled in the art can make various alterations and modifications to this application without departing from the spirit and scope of this application. Thus, if such modifications and modifications of this application fall within the scope of the claims of this application and their equivalents, this application is also intended to include such modifications and modifications.

Claims

1. A method for optimizing the control of joint motors in an exoskeleton upper limb robot, characterized in that, include: Acquire the current operating status data of the joint motors of the exoskeleton upper limb robot and the human-computer interaction status data; The control parameters of the joint motor are globally optimized based on a multi-strategy collaborative optimization mechanism to obtain the target control parameters. The multi-strategy collaborative optimization mechanism includes a population collaboration strategy that divides the population into a first subpopulation that focuses on dynamic response speed and a second subpopulation that focuses on steady-state smoothness for parallel search, and a search balance strategy that adjusts the inertia weight based on the iteration progress. The fitness values ​​of candidate control parameters are determined based on the fitness evaluation model to guide the optimization direction of the multi-strategy collaborative optimization mechanism; wherein, the fitness evaluation model includes a time multiplication absolute error integral term and a safety penalty term for motion smoothness. Based on the target control parameters and the current operating status data, the drive control signal for the joint motor is generated.

2. The method according to claim 1, characterized in that, The multi-strategy collaborative optimization mechanism is used to globally optimize the control parameters of the joint motor to obtain the target control parameters, including: The initial population is divided into the first subpopulation and the second subpopulation; In each iteration, position updates are performed on the first subpopulation and the second subpopulation respectively to obtain the first updated subpopulation and the second updated subpopulation; Based on a preset information exchange mechanism, the optimal individual information is exchanged between the first update subpopulation and the second update subpopulation in order to update the global optimal solution; When the preset iteration termination condition is met, the control parameters corresponding to the global optimal solution are used as the target control parameters.

3. The method according to claim 2, characterized in that, The step of performing position updates on the first subpopulation and the second subpopulation in each iteration includes: A mutation operator is introduced to perturb the current population centroid, generating reverse-learning individuals; Based on the fitness values ​​of the reverse-learned individual and the current individual, the updated individual position is determined; The inertia weights in the search balancing strategy are nonlinearly adjusted based on the ratio of the current iteration count to the maximum iteration count.

4. The method according to claim 1, characterized in that, The determination of the fitness values ​​of candidate control parameters based on the fitness evaluation model includes: Obtain the joint angular velocity error sequence of the joint motor when the candidate control parameters are applied; Based on the joint angular velocity error sequence and the corresponding sampling time, the time-to-absolute-error integral term is calculated; Obtain the rate of change of joint angular acceleration of the joint motor when the candidate control parameters are applied; When the rate of change of the joint angle acceleration exceeds the preset rehabilitation safety threshold, the safety penalty item is calculated based on the excess amount; The fitness value is determined based on the weighted sum of the time multiplication absolute error integral term and the safety penalty term.

5. The method according to claim 4, characterized in that, The method further includes: A torque penalty term for human-computer interaction torque is introduced into the fitness evaluation model; Based on the current operating status data and the candidate control parameters, predict the human-machine interaction torque for the next sampling period; When the predicted human-computer interaction torque exceeds the preset torque safety threshold, the torque penalty term is calculated based on the amount of torque excess, and the torque penalty term is included in the weighted calculation of the fitness value.

6. The method according to claim 1, characterized in that, The step of generating the drive control signal for the joint motor based on the target control parameters and the current operating status data includes: The target control parameters are mapped to proportional gain, integral gain, and derivative gain; Based on the proportional gain, the integral gain, the differential gain, and the state error between the current operating state data and the desired state data, a basic driving signal is generated; The differential term in the basic drive signal is subjected to low-pass filtering to generate the drive control signal.

7. The method according to claim 6, characterized in that, The step of performing low-pass filtering on the differential term in the basic driving signal includes: Obtain the state error difference between the current sampling period and the previous sampling period; The state error difference is weighted and fused with a preset filtering time constant to obtain the filtered differential compensation amount. The original differential term in the basic driving signal is replaced by the filtered differential compensation amount.

8. The method according to claim 1, characterized in that, The method further includes: During rehabilitation training, the human-computer interaction status data is monitored in real time; When the human-computer interaction status data is detected to meet the preset mode switching conditions, the multi-strategy collaborative optimization mechanism is triggered to perform local re-optimization of the control parameters in order to update the target control parameters.

9. The method according to claim 8, characterized in that, The human-computer interaction status data includes interaction torque data and surface electromyography signals; The mode switching conditions include at least one of the following: The detected interactive torque data exceeded the passive training torque threshold; The amplitude of the collected surface electromyography signal exceeded the active intent trigger threshold.

10. An optimized control system for joint motors of an exoskeleton upper limb robot, characterized in that, The system includes: The status acquisition module is used to acquire the current operating status data of the joint motors of the exoskeleton upper limb robot and the human-computer interaction status data; The collaborative optimization module is used to perform global optimization of the control parameters of the joint motor based on a multi-strategy collaborative optimization mechanism to obtain the target control parameters. The multi-strategy collaborative optimization mechanism includes a population collaboration strategy that divides the population into a first subpopulation that focuses on dynamic response speed and a second subpopulation that focuses on steady-state smoothness for parallel search, and a search balance strategy that adjusts the inertia weight based on the iteration progress. The fitness evaluation module is used to determine the fitness values ​​of candidate control parameters based on the fitness evaluation model, so as to guide the optimization direction of the multi-strategy collaborative optimization mechanism; wherein, the fitness evaluation model includes a time multiplication absolute error integral term and a safety penalty term for motion smoothness; The drive control module is used to generate drive control signals for the joint motor based on the target control parameters and the current operating status data.

Images