Multi-user self-learning adaptive exoskeleton parameter self-tuning control method

Abstract

The present disclosure provides a multi-user self-learning adaptive exoskeleton parameter self-tuning control method, which comprises spatial mapping of the multi-point contact relationship between the exoskeleton and the human body, constructing the distribution relationship of joint driving force in each contact path to obtain an initial force transmission path model; performing actual force transmission attenuation calculation and offset identification in the force transmission path to form a joint level force transmission offset description set; constructing a force transmission path compensation parameter mapping relationship containing torque compensation gain, direction correction amount and stability constraint amount, and establishing a joint dimension compensation mapping table; performing adaptive control parameter tuning and outputting the corrected control instruction; the difference between the actual motion result after the instruction execution and the target motion state, the control effect feedback and the force transmission path model updating are performed, and the continuous self-tuning result is output. The method can identify the real force transmission path and perform control parameter self-tuning to solve the force transmission distortion problem under the multi-joint passive coupling structure.

Description

Technical Field

[0001] This disclosure relates to the field of exoskeleton robot control technology, and in particular to a multi-user self-learning adaptive exoskeleton parameter self-tuning control method. Background Technology

[0002] Exoskeleton robots have broad application prospects in heavy-duty scenarios such as industrial material handling and emergency rescue. Their core function is to distribute the load on human joints by outputting auxiliary torque, thereby improving work efficiency and protecting the operator's health. However, in actual heavy-duty material handling, such as steel structure handling on construction sites, there are multiple flexible contacts and structural gaps between the human body and the exoskeleton. This multi-joint passive coupling structure causes nonlinear attenuation and path deviation of the joint output force during transmission. Specifically, the control parameters obtained based on the dynamic response model cannot be accurately applied to the target joints during actual execution, causing the auxiliary torque of the exoskeleton to fail to be transmitted to the corresponding parts of the human body as expected, resulting in local joint overcompensation or undercompensation. Overcompensation creates antagonistic forces between the human and the robot, increasing the operator's discomfort and additional burden; undercompensation prevents the exoskeleton from providing sufficient assistance, making it difficult to achieve the expected load reduction effect. Most existing control methods assume that the force transmission path is deterministic and linear, failing to fully consider the force transmission distortion caused by flexible connections and structural gaps. Therefore, the control accuracy and adaptability are limited in practical applications.

[0003] Therefore, there is an urgent need for an exoskeleton control method that can identify the true force transmission path and self-tune the control parameters in order to solve the problem of force transmission distortion under multi-joint passive coupling structure and achieve stable following and precise force matching of the human body under heavy load conditions. Summary of the Invention

[0004] In view of this, in order to solve the problems caused by the prior art, this application provides a multi-user self-learning adaptive exoskeleton parameter self-tuning control method.

[0005] In a first aspect, this disclosure provides a multi-user self-learning adaptive exoskeleton parameter self-tuning control method, the method comprising: S1: Spatial mapping of the multi-point contact relationship between the exoskeleton and the human body, constructing the distribution relationship of joint driving force in each contact path, and obtaining the initial force transmission path model; S2: Based on the initial force transmission path model, perform actual force transmission attenuation calculation and offset identification in the force transmission path to form a set of joint-level force transmission offset descriptions; S3: Based on the joint-level force transmission offset description set, construct the force transmission path compensation parameter mapping relationship including torque compensation gain, direction correction amount and stability constraint amount, and establish a joint dimension compensation mapping table; S4: Based on the joint dimension compensation mapping table and the compensation parameter mapping relationship, match it with the current user's real-time motion state, perform adaptive control parameter tuning, and output the corrected control command; S5: Based on the difference between the actual motion result after the execution of the modified control command and the target motion state, perform control effect feedback and force transmission path model update, and output continuous self-tuning results.

[0006] Optionally, S1 includes: Collect data on exoskeleton joint driving torque, human joint reaction force, and pressure, displacement, and spatial position at each contact point to form a unified and correlated sampling set; Based on the unified associated sampling set, the equivalent contact stiffness, torque transmission capacity and contact force transmission capacity index of each contact point are calculated to form a mechanical characterization set of the contact point. Based on the contact point mechanical characterization set, perform multi-point contact space mapping to construct the distribution relationship of joint driving force in each contact path and obtain the corrected distribution coefficient; Based on the corrected allocation coefficients and the displacement response of each contact path, combined with the joint hierarchy numbering rules, a set of joint force transmission path expressions is generated, and the initial force transmission path model is obtained.

[0007] Optionally, S2 includes: The initial force transmission path model is segmented and expanded to form a path segment force transmission calculation sequence; Based on the path segment force transmission calculation sequence, the actual force transmission attenuation and attenuation ratio of each path segment are calculated to form a segment attenuation calculation result set. The transmission direction offset angle of each segment is calculated based on the segmented attenuation calculation result set to form a path direction offset result set; By combining the attenuation ratio in the segmented attenuation calculation result set, the directional offset angle in the path direction offset result set, and the actual human body reaction force, the offset and distortion degree in the path are identified, forming the joint-level force transmission offset description set.

[0008] Optionally, S3 includes: The joint-level force transmission offset description set is decomposed into an intra-joint decomposition to form a path compensation modeling base set; Based on the path compensation modeling base set, the path torque compensation gain, direction correction amount and stability constraint amount are calculated to form a path compensation parameter set; The path compensation parameter set is used to classify and summarize the path compensation parameters for different users and under different working conditions, forming a multi-user shared compensation parameter cluster; The joint dimension compensation mapping table is established based on the multi-user shared compensation parameter cluster.

[0009] Optionally, after classifying and summarizing the path compensation parameters for different users and under different operating conditions to form a multi-user shared compensation parameter cluster, the method further includes: When the user's current operating condition characteristics match multiple existing compensation parameter clusters, calculate the distance between the current operating condition characteristics and the central characteristics of each compensation parameter cluster; Select the N compensation parameter clusters with the smallest distance, where N is 2 or 3; Using the reciprocal of the dispersion of each compensation parameter cluster as the fusion weight, the cluster center moment compensation gain and the cluster center direction correction are weighted and averaged to obtain the fused compensation parameters.

[0010] Optionally, S4 includes: The current user's real-time motion state is matched with the joint dimension compensation mapping table at the joint level to form the current working condition compensation call set; Based on the joint target drive output set and the current working condition compensation call set, calculate the joint-level target compensation drive torque to form the joint target drive output set; Based on the joint compensation control parameter set, continuous constraints and control commands are generated on the corrected control output to form the joint compensation control parameter set. The modified control output corresponding to the joint compensation control parameter set is subjected to continuous constraints and control command generation, and the modified control command is output.

[0011] Optionally, S5 includes: Based on the joint deviation response after the execution of the modified control command, feedback information is extracted to form a feedback evaluation base set; Based on the feedback evaluation base set, the path mismatch intensity is calculated inversely to form a path correction set; The force transmission path model is updated based on the path correction set to form an updated path expression model. The updated path representation model is used to write back the compensation parameters to the joint dimension compensation mapping table, and the call record information in the joint dimension compensation mapping table is updated to output the continuous self-tuning result.

[0012] Optionally, after the output continues to self-tuning, it further includes: Based on the number of times the same user continuously calls the same path compensation record under the same load level and action phase, the calling priority of that record in the compensation mapping table is increased; Based on the degree to which the dispersion of the shared compensation parameter cluster is consistently large, the default call level of the compensation record corresponding to that cluster is reduced.

[0013] In a second aspect, this disclosure provides an electronic device including a memory and at least one processor, the memory storing a computer program, and the processor executing the computer program to implement the method of the first aspect described above.

[0014] Thirdly, this disclosure provides a computer storage medium storing a computer program that, when executed, implements the method described in the first aspect.

[0015] The beneficial effects of the present invention are as follows: Compared with the prior art, the present invention has the following advantages: (1) By constructing a joint force transmission path expression model and performing actual force transmission attenuation calculation and offset identification, the abstract force transmission distortion is transformed into quantifiable segmented attenuation ratio, directional offset angle, and joint-level comprehensive distortion degree, accurately locking the specific path and link of force transmission distortion. Based on this, the establishment of subsequent compensation parameters no longer relies on the idealized linear force transmission assumption, but is based on the measured distortion in the real human-machine coupling structure, fundamentally solving the problem of low control accuracy and poor adaptability caused by neglecting flexible connections and structural gaps in existing control methods.

[0016] (2) By constructing a force transmission path compensation parameter mapping relationship, the compensation experience of different users under similar load levels, action stages, and contact clamping states is classified and summarized into a multi-user shared compensation parameter cluster, and a joint dimension compensation mapping table is established. On this basis, adaptive control parameter tuning based on the compensation mapping is performed, which not only compensates for the torque value, but also corrects the offset of the transmission direction, so that the exoskeleton can quickly match the optimal compensation strategy according to the current user's real-time motion state. At the same time, through control effect feedback and force transmission path model update, a closed-loop self-learning mechanism is formed, so that the model and parameters continuously approach the real human-machine coupling structure as the usage process progresses. Thus, the exoskeleton achieves automatic adaptation and parameter self-tuning under different users, different loads, and different working conditions, significantly improving the following accuracy and assist effect in heavy-duty handling and other scenarios, effectively avoiding local joint over-compensation or under-compensation, and reducing operator discomfort and additional burden. Attached Figure Description

[0017] The accompanying drawings, which are incorporated in and form a part of this specification, illustrate embodiments consistent with this disclosure and, together with the description, serve to explain the principles of this disclosure.

[0018] Figure 1 A flowchart of the multi-user self-learning adaptive exoskeleton parameter self-tuning control method provided in an embodiment of this disclosure is shown; Figure 2 A flowchart illustrating the compensation parameter mapping and shared cluster generation process provided in an embodiment of this disclosure is shown.

[0019] The accompanying drawings have illustrated specific embodiments of this disclosure, which will be described in more detail below. These drawings and descriptions are not intended to limit the scope of the concept in any way, but rather to illustrate the concepts of this disclosure to those skilled in the art through reference to particular embodiments. Detailed Implementation

[0020] The present disclosure will be further described below with reference to the accompanying drawings. The following embodiments are only used to illustrate the technical solutions of the present disclosure more clearly, and should not be used to limit the scope of protection of the present disclosure.

[0021] The components of the embodiments of the invention described and illustrated herein can typically be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention.

[0022] In the following, the terms “comprising,” “having,” and their cognates, which may be used in various embodiments of the invention, are intended only to indicate a particular feature, number, step, operation, element, component, or combination thereof, and should not be construed as excluding, firstly, the presence of one or more other features, numbers, steps, operations, elements, components, or combinations thereof, or adding the possibility of one or more features, numbers, steps, operations, elements, components, or combinations thereof.

[0023] Unless otherwise specified, all terms used herein (including technical and scientific terms) shall have the same meaning as commonly understood by one of ordinary skill in the art to which the various embodiments of the invention pertain. Terms (such as those defined in commonly used dictionaries) shall be interpreted as having the same meaning as in their contextual meaning in the relevant technical field and shall not be interpreted as having an idealized or overly formal meaning, unless clearly defined in the various embodiments of the invention.

[0024] Figure 1 The flowchart of the multi-user self-learning adaptive exoskeleton parameter self-tuning control method provided in the embodiments of this disclosure is as follows: Figure 1 As shown, the process may include the following steps: S1: Spatial mapping of the multi-point contact relationship between the exoskeleton and the human body, construction of the distribution relationship of joint driving force in each contact path, and obtaining the initial force transmission path model.

[0025] Constructing a joint force transmission path representation model under a human-machine coupled structure. This step aims to spatially map the multi-point contact relationship between the exoskeleton and the human body, construct the distribution relationship of joint driving forces in each contact path, form a set of joint force transmission path representations, and obtain an initial force transmission path model. This is specifically achieved through the following sub-steps.

[0026] S1.1: Collect data on the driving torque of exoskeleton joints, the reaction force of human joints, and the pressure, displacement, and spatial position of each contact point to form a unified associated sampling set.

[0027] Torque sensors are deployed at the drive joints of the exoskeleton, such as the hip, knee, and ankle joints. Reaction force detection units are deployed at the corresponding joint connection points on the human body. Pressure sensors and micro-displacement sensors are deployed at contact locations such as straps, shell contact areas, and support pad areas, and the spatial position of each contact point is recorded simultaneously. Subsequently, the driving torque, reaction force, contact pressure, contact displacement, and contact position at the same moment are aligned with a unified sampling period to form a unified associated sampling set. The sampling period is 5 milliseconds to 20 milliseconds, with 8 milliseconds to 12 milliseconds preferred in heavy-duty handling scenarios to balance dynamic response and noise immunity. The number of contact points is 6 to 24, and each contact point needs to be bound to its respective joint level information to facilitate the subsequent construction of force transmission paths along the joint level.

[0028] S1.2: Calculate the equivalent contact stiffness, torque transmission capacity, and contact force transmission capacity index of each contact point based on the unified correlation sampling set, forming a mechanical characterization set of the contact point.

[0029] For each contact point, the ratio of pressure change to displacement change is calculated within adjacent sampling periods to characterize the equivalent contact stiffness of that contact point under the current loading state. Simultaneously, the pressure application area is incorporated into the calculation to obtain the contact force. Combined with the lever arm length from the contact point to the corresponding joint center, the torque transfer capacity of that contact point is calculated. To avoid artificially inflated stiffness due to local indentation of the flexible strap, a lower limit is set for the displacement change, ranging from 0.2 mm to 0.8 mm. When the displacement change is below the lower limit, the lower limit value is used in the calculation.

[0030] Specifically, for the i-th contact point, its equivalent contact stiffness is The value is obtained by multiplying the ratio of the change in contact pressure to the change in displacement between two adjacent sampling times by the equivalent pressure area. The calculation formula is as follows: ; in, and These represent the contact pressure at two adjacent sampling times; This represents the equivalent pressure area of ​​the i-th contact point; and These represent the contact displacements at two adjacent sampling times.

[0031] The torque transmission capability of this contact point The calculation formula is determined by the current contact pressure, the equivalent pressure area, and the lever arm length. ; in, Indicates the current moment of contact with pressure; This indicates the lever arm length from the contact point to the corresponding joint rotation center.

[0032] Furthermore, a contact force transmission capability index is defined. It combines equivalent contact stiffness and torque transmission capacity, and incorporates the attenuation effect of path distance on force transmission capacity. The calculation formula is as follows: ; in, This represents the path distance from the contact point to the center of the drive joint along the force transmission direction. The path distance reference constant, ranging from 0.02 meters to 0.1 meters, is used to characterize the attenuation effect of increasing distance on force transmission capacity.

[0033] The above calculations convert the contact tightness, pressure level, and distance from the joint of the contact point into comparable quantitative results, providing a unified criterion for subsequent contact path selection. Thus, the equivalent contact stiffness, torque transmission capacity, and contact force transmission capacity index of each contact point are obtained, collectively constituting the mechanical characterization set of the contact point.

[0034] S1.3: Perform multi-point contact space mapping based on the contact point mechanical characterization set, construct the distribution relationship of joint driving force in each contact path, and obtain the corrected distribution coefficient.

[0035] Starting from each drive joint, contact points belonging to that joint level or adjacent levels are selected as candidate relay points. Contact paths are established based on spatial proximity, stiffness continuity, and force transmission continuity. Each path contains at least one primary contact point and, if necessary, secondary transmission contact points, such as from the thigh shell area to the strap area and then to the soft tissue area of ​​the thigh. A path allocation coefficient is calculated for each path; a larger coefficient indicates that the driving force is more likely to be transmitted along that path. The spatial distance threshold is set to 20 mm to 80 mm, and the stiffness continuity deviation threshold is set to 10% to 35%. The equivalent stiffness of each candidate path is... Take the minimum equivalent stiffness of all contact points along the path.

[0036] For the i-th drive joint and the j-th candidate contact path, first calculate the initial allocation ratio. It is equal to the overall contact force transmission capacity of that path. The ratio of the total force-transmitting capacity of all candidate paths of the same drive joint to the sum of their combined contact capabilities, i.e.: ; in, It is obtained by summing the contact force transmission capacity index of each contact point along the path. It is the sum of the comprehensive contact force transmission capabilities of all candidate paths for the same drive joint.

[0037] Then, taking into account both spatial distance and stiffness matching, the initial allocation ratio is corrected to obtain the corrected allocation coefficient. : ; in, Let be the spatial distance from the center of the i-th driving joint to the center of the main contact area of ​​the j-th candidate path; The distance attenuation reference constant is taken as 0.02 meters to 0.1 meters; This represents the path equivalent stiffness of the j-th candidate path; This represents the average equivalent stiffness of all candidate paths at this joint level. The above calculation first allocates force based on force transmission capacity, and then corrects for spatial remoteness and stiffness mismatch, thus obtaining a more realistic human-machine coupling force transmission ratio. All corrected allocation coefficients constitute the contact path allocation set.

[0038] S1.4: Combine displacement response and joint hierarchy numbering rules to generate a set of joint force transmission path expressions and obtain an initial force transmission path model.

[0039] For each contact path, the force transmission response coefficient and force transmission stability coefficient are calculated. The force transmission response coefficient reflects whether the path can transmit force with minimal additional deformation after the driving force is applied; the force transmission stability coefficient reflects whether the path will become unstable during continuous transport due to soft tissue compression, strap loosening, or local slippage. Then, a unified numbering system is completed according to the driving joint number, path number, and contact level number. For example, the first contact area of ​​the first main path of the hip joint can be numbered "hip-1-". The allocation coefficient, response coefficient, stability coefficient, and path number are written into the path representation set to obtain the initial force transmission path model.

[0040] Specifically, for the j-th contact path, its force transmission response coefficient By the corrected allocation coefficient The ability of a path to transmit torque Path equivalent stiffness Cumulative compression displacement The decision was made jointly, and the calculation formula is as follows: ; in, This represents the torque transmission capability of the path, measured in Newton-meters. This refers to the corrected allocation coefficients calculated in step S1.3. , is a dimensionless quantity; This represents the cumulative compression displacement during the current loading phase, in meters. This represents the force conversion reference constant, ranging from 50 N to 400 N. A larger coefficient indicates a faster force transmission response along the path. Force transmission stability coefficient. Then, the influence of displacement fluctuation amplitude is introduced into the force transmission response coefficient: ; in, This represents the displacement fluctuation amplitude of the path over several consecutive sampling periods, in meters. The displacement fluctuation reference constant is taken from 0.0005 meters to 0.005 meters. The larger this coefficient is, the more stable the path is during continuous handling. Finally, the correction allocation coefficient, force transmission response coefficient, force transmission stability coefficient, and path number generated according to the rule of drive joint number-path sequence number-contact level sequence number for each path are written into the path expression set, thus obtaining the initial force transmission path model that can be directly called for subsequent steps.

[0041] In the technical solution of this disclosure embodiment, the complex multi-point flexible contact relationship between the exoskeleton and the human body is transformed into a structured set of force transmission paths through multi-point contact spatial mapping and contact point mechanical characterization. A quantitative model that can reflect the force transmission path distribution, equivalent stiffness distribution and contact stability is established, providing a calculable basis for subsequent identification of force transmission distortion and compensation control. This solves the problem that the prior art cannot describe the fuzzy force transmission path under multi-joint passive coupling structure.

[0042] S2: Based on the initial force transmission path model, perform actual force transmission attenuation calculation and offset identification in the force transmission path to form a set of joint-level force transmission offset descriptions.

[0043] Perform actual force transmission attenuation calculation and offset identification under the force transmission path. This step, based on the initial force transmission path model, calculates the force transmission process in each force transmission path, determines the attenuation ratio and transmission direction change of the driving force in the path, and forms a set of joint-level force transmission offset descriptions. This is specifically achieved through the following sub-steps.

[0044] S2.1: Expand the initial force transmission path model into segments to form a path segment force transmission calculation sequence.

[0045] The initial force transmission path model provides overall information for each path, but force transmission distortion often occurs in a local segment of the path. Therefore, it is necessary to segment each joint force transmission path. Specifically, the path is divided into segments in the order of drive joint output segment, structural contact transition segment, and human force application segment. For cases where there are multiple contact zones in the path, it should be further subdivided into local transmission segments between adjacent contact zones. Each segment must retain its path length, contact stiffness, cumulative compressive displacement, displacement fluctuation amplitude, path distribution coefficient, force transmission response coefficient, and force transmission stability coefficient, thus forming a segmented force transmission calculation sequence that can be calculated segment by segment.

[0046] In practice, the minimum length of the path segment is controlled between 5 mm and 30 mm. If the distance between two adjacent contact areas exceeds 80 mm, it is forcibly divided into more than two segments to avoid merging significantly different soft tissue deformation areas with shell force transmission areas. The segmented sequence of each path maintains consistency with the joint hierarchy numbering in step S1, so that the attenuation results obtained in subsequent calculations can be directly traced back to the specific joint, path, and segment. Essentially, this step converts the overall force transmission path expression obtained in step S1 into a segment-by-segment force transmission calculation object, laying the foundation for subsequent calculations of the actual force attenuation and directional shift in each segment.

[0047] S2.2: Calculate the actual force transmission attenuation and attenuation ratio of each path segment based on the path segment force transmission calculation sequence, and form a segment attenuation calculation result set.

[0048] After obtaining the force transmission calculation sequence for each segment of the path, for each segment in the sequence, the force retention of that segment is first calculated based on the input transmitted torque, path length, equivalent stiffness, cumulative compressive displacement, and displacement fluctuation amplitude. Then, the attenuation amount and attenuation ratio of that segment are deduced. Since this step addresses force transmission distortion, the calculation focus is not on simply comparing the numerical difference between the starting and ending points, but on subtracting from each segment one by one to identify which segment experienced significant attenuation.

[0049] When calculating the segmented attenuation, the input transmission torque of the drive joint output segment is directly taken as the output torque of that joint at the current sampling moment; for subsequent segments, the input transmission torque is taken as the output transmission torque of the previous segment, so as to ensure the continuity of the force transmission relationship. For segments with large displacement fluctuations, a stability penalty needs to be introduced to reflect the additional losses caused by strap slack, soft tissue slippage, or local collapse of the contact surface.

[0050] Specifically, for the z-th segment in the j-th force transmission path, its output transmission torque is... Torque is transmitted from the output of the previous segment. The formula is: (Divided by the sum of multiple penalty terms) ; If the current segment is the first segment, then Take the input driving torque of the drive joint corresponding to the path; This indicates the segment length of the segment; The length penalty reference constant is taken from 0.01 meters to 0.1 meters; This indicates the equivalent stiffness of the segment; This represents the cumulative compressive displacement of the segment; This indicates the displacement fluctuation amplitude of this segment; The displacement fluctuation reference constant is taken from 0.0005 meters to 0.005 meters. This formula treats the increase in path length, contact softening, compression, and slip disorder as a uniform attenuation factor that leads to a decrease in force transmission capacity, and recursively derives the actual torque loss process segment by segment.

[0051] After obtaining the output transmission torque, the attenuation of the transmission torque in this segment. This is the difference between the output torque of the previous segment and the output torque of the current segment: ; The corresponding torque attenuation ratio This is the ratio of the attenuation to the output torque of the previous stage: ; Through the above calculations, the output transmission torque, attenuation amount, and attenuation ratio of each segment are obtained, forming a set of segmented attenuation calculation results. Only by calculating the attenuation amount at the segment level can it be determined whether the main distortion is caused by the contact area, transition area, or human body interaction area.

[0052] S2.3: Calculate the directional offset angle of each path segment based on the segmented attenuation calculation result set to form the path directional offset result set.

[0053] In heavy-duty handling scenarios, force transmission distortion is not only manifested as a decrease in torque value, but also often as a deviation in the transmission direction. For example, the driving force output by the hip joint should primarily act on the proximal femoral rotation control, but due to strap misalignment, shell warping, or local soft tissue slippage, the actual force direction may shift forward and upward or outward. Therefore, it is necessary to identify the change in transmission direction for each segment based on the spatial location information and segmental attenuation calculation results in the segmented force transmission calculation sequence. In specific implementation, the spatial coordinates of the start and end points of each path segment are first obtained to form the nominal force transmission direction of that path segment; then, combined with the center coordinates of the actual force-bearing area of ​​the human body after the end of the path segment's action, the actual force direction is formed. Subsequently, the change in the angle between the two is calculated. For segments closer to the force-bearing end of the human body, the influence of directional deviation should be appropriately increased, because end-point deviation is more likely to directly cause force misalignment of the target joint.

[0054] First, calculate the nominal force transmission direction length. That is, the spatial distance between the start and end points of the segment: ; in , , These represent the three-dimensional spatial coordinates of the starting point of the segment. , , These represent the three-dimensional spatial coordinates of the endpoint of the segment.

[0055] Then calculate the length in the actual force direction. That is, the spatial distance between the starting point of the segmented action and the actual center of force on the human body: ; in, , , These represent the three-dimensional spatial coordinates of the starting point of the segment's action. , , These represent the three-dimensional spatial coordinates of the actual center point of force on the human body.

[0056] Furthermore, the components of the nominal force transmission direction on the three coordinate axes are respectively denoted as... , , The coordinates are obtained by subtracting the starting coordinates from the ending coordinates; the components of the actual force direction on the three coordinate axes are denoted as follows: , , The angle of deviation is obtained by subtracting the coordinates of the starting point from the actual coordinates of the force center. It is given by the angle between the two direction vectors: ; This angle, measured in radians, quantifies the difference between the intended direction of force transmission and the actual direction of force transmission. A larger angle indicates a more severe deviation in the force transmission direction for that segment. In specific calculations, if the nominal force transmission direction length or actual force-bearing direction length of a segment is less than 1 mm, it is substituted as 1 mm to prevent calculation anomalies caused by extremely short local segments. The direction deviation angle can be further divided into three levels: slight deviation, moderate deviation, and significant deviation, with cutoff values ​​of 0.09 to 0.17 radians, 0.17 to 0.35 radians, and greater than 0.35 radians, respectively. The calculated direction deviation angles for all segments constitute the path direction deviation result set.

[0057] S2.4: Combine path attenuation and direction offset to identify the offset and distortion in the path to form a set of joint-level force transmission offset descriptions.

[0058] Based on the path direction offset result set generated by S2.3, the direction offset angle of the last segment of each path is extracted as the total direction offset angle of that path. If multiple paths exist for the same joint, the offset angle at the end of the main path with the largest contribution is taken. The segmented attenuation ratios, directional offset angles, and actual human force responses at the ends of all force transmission paths within the same joint level are combined and summarized to construct a comprehensive distortion assessment result at the joint level. The key here is not just to identify a problematic path, but to answer the question of the extent of force attenuation and directional offset ultimately experienced by the joint. Therefore, it is necessary to first accumulate the output transmission torques at the ends of each path within the same joint to obtain the theoretically achievable human force torque for that joint; then, this is compared with the actual joint force torque calculated from the actual human reaction force to calculate the overall offset and the degree of comprehensive distortion.

[0059] In practice, the conversion of the actual reaction force of the human body into the torque of the joint needs to be combined with the lever arm length corresponding to the reaction force. The lever arm length can be directly obtained from the spatial mapping result in step S1, and the recommended range is 0.02 meters to 0.25 meters. If there are more than two main force transmission paths under the same joint, they can be weighted and summarized according to the path correction allocation coefficient.

[0060] Specifically, for the j-th joint level, the total amount of human body action torque that should theoretically be achieved after transmission through each path. It equals the sum of the torques transmitted at the ends of all force transmission paths under this joint: ; in, This represents the output transmitted torque at the end of the m-th force transmission path under the j-th joint. It is the actual joint torque calculated based on the actual reaction forces of the human body. for: ; in, This indicates the actual reaction force of the human body at that joint; This represents the lever arm length corresponding to the reaction force. Then, the overall distortion level of the j-th joint is... for: ; in, This indicates the overall torque mismatch ratio of the joint. This represents the total directional offset angle at the end of the main force transmission path of the joint, taken as the directional offset angle of the last segment of the path. , The normalized reference constant for directional offset is set to 0.17 to 0.52 radians. This comprehensive distortion level unifies the transmission of sufficient torque and directional deviation into a dimensionless index, thus providing direct input for the subsequent step S3 to construct the compensation parameter mapping relationship.

[0061] Finally, the joint number, main distortion segment number, total attenuation ratio, total directional offset angle, and overall distortion level are written into the joint-level force transmission offset description set for reference in step S3.

[0062] In the technical solution of this disclosure, each segment is recursively calculated based on the initial force transmission path model, quantifying the torque attenuation ratio and transmission direction offset angle of the driving force in each segment, and forming a joint-level comprehensive distortion index. This transforms the abstract force transmission distortion problem into measurable attenuation, offset angle, and distortion level, accurately locating the specific path segment where distortion occurs. This provides a direct basis for the precise design of compensation parameters and overcomes the shortcomings of traditional methods that treat force transmission as a linear deterministic process and ignore local distortion.

[0063] S3: Based on the joint-level force transmission offset description set, construct the force transmission path compensation parameter mapping relationship including torque compensation gain, direction correction amount and stability constraint amount, and establish a joint dimension compensation mapping table.

[0064] Construct a force transmission path compensation parameter mapping relationship. This step inputs the joint-level force transmission offset description set into the compensation mapping module, establishes the mapping relationship between the joint drive output and the actual target force, forms a compensation parameter set shared by multiple users, and establishes a compensation mapping table according to the joint dimension. Figure 2 A flowchart illustrating the compensation parameter mapping and shared cluster generation process provided in this disclosure embodiment is shown, such as... Figure 2 As shown, this is achieved through the following sub-steps.

[0065] S3.1: Perform intra-joint decomposition on the joint-level force transmission offset description set to form the path compensation modeling base set.

[0066] The joint-level force transmission offset description set output in step S2 contains comprehensive distortion information for each joint, but compensation needs to be implemented at the specific path and segment level. Therefore, the description set is first split according to the joint dimension, and the main force transmission path number, main distortion segment number, total attenuation ratio, total directional offset angle, actual force torque on the human body, theoretical torque, and comprehensive distortion level are extracted for each joint. Then, using a single joint, single path, and single main distortion segment as the smallest modeling unit, compensation modeling records are established for multiple paths under the same joint. Here, the amount of force loss, directional deviation, and remaining torque on the human body in each path are first separated and organized to form the path compensation modeling base set that can participate in the subsequent compensation mapping calculation.

[0067] For situations where there are more than two main paths for the same joint, the paths should be sorted according to the proportion of the output transmission torque at the end obtained in step S2. The main paths with a cumulative proportion of 85% or more should be retained first, and the remaining paths should be included in the auxiliary compensation record as secondary compensation paths. This avoids diluting the main compensation relationship due to the large number of locally weak paths in heavy-duty handling scenarios. Furthermore, each path needs to be supplemented with a current working condition identifier, including load level, action stage, and contact clamping state. The load level can be determined based on the total driving torque of the exoskeleton; the action stage can be divided into lifting stage, steady-state handling stage, and unloading stage; the contact clamping state can be divided into three levels—low clamping, medium clamping, and high clamping—based on the average contact pressure in step S1. This forms the path compensation modeling base set, which is used in the next sub-step to establish path compensation gain and direction correction. This sub-step transforms the identification results of step S2 into compensation objects that can be directly modeled in step S3, essentially converting the distortion problem from an analytical conclusion into a parameter mapping input.

[0068] S3.2: Calculate the path torque compensation gain, direction correction amount and stability constraint amount based on the path compensation modeling base set to form a path compensation parameter set.

[0069] For each path in the path compensation modeling base set, the torque compensation gain, direction correction, and stability constraint are calculated. The torque compensation gain characterizes how much the original drive output needs to be amplified to still achieve the target force after attenuation through the path; the direction correction characterizes how much the original drive direction needs to be deflected to offset the actual offset in the path; and the stability constraint limits the compensation from overshooting, avoiding overcompensation of local joints due to excessive compensation.

[0070] In practical implementation, the theoretical torque and actual torque in step S2 are first converted into a ratio to obtain the basic compensation multiple. Then, the overall distortion degree and total directional offset angle are introduced as enhancement factors. Finally, the compensation weight of the path is corrected according to the proportion of the output torque transmitted at the end of the path. For paths with high overall distortion degree but low output proportion at the end of the path, it is not advisable to give excessive compensation weight, otherwise it is easy to amplify the weak paths outside the main path abnormally. At the same time, in order to ensure that the compensation parameters can be used for subsequent control tuning, the path compensation results need to be uniformly converted into a form of compensation parameters that can be directly attached to the joint drive output, that is, the output is the torque compensation gain, direction correction factor and compensation upper limit factor for each path. The setting of the compensation upper limit factor is particularly important. In engineering, it is taken between 1.1 and 1.8. It can be relaxed to 2.2 in heavy-load transient lifting conditions, but it should be controlled within 1.5 in the continuous steady-state handling stage.

[0071] Specifically, for the j-th path, let its joint number be J, and its torque compensation gain... The result is obtained by multiplying the ratio of the theoretical torque to the actual torque by the distortion enhancement factor and the orientation offset enhancement factor: ; in, This represents the theoretical torque applied to the joint corresponding to this path. This represents the actual torque applied to the joint. This represents the overall degree of distortion of the joint; This is the total directional offset angle of the main force transmission path of the joint; The normalized reference constant for directional offset is defined as 0.17 radians to 0.52 radians; r is the distortion enhancement exponent, ranging from 0.6 to 1.4; and s is the directional offset enhancement exponent, ranging from 0.4 to 1.2. This formula converts the torque gap, combined distortion, and directional offset into the amount of drive output that needs to be increased to compensate for them.

[0072] Directional correction amount Based on the existing offset angle and the superimposed comprehensive distortion level, the following is obtained: ; This formula amplifies the already identified offset angle to the correction amount required for actual control.

[0073] Compensation cap factor This is used to limit the compensation range based on whether the path itself is the primary carrying path: ; in, The percentage of the torque transmitted at the end of the path; The baseline constant for the proportional balance is taken as 0.1 to 0.4. This formula can prevent weak paths from being mistakenly amplified. In specific implementation, either piecewise interpolation compensation or hierarchical lookup table compensation can be used. Piecewise interpolation compensation is suitable for scenarios with continuously changing path states; hierarchical lookup table compensation is suitable for handling conditions with clear load levels and fixed action phases. If the on-site control cycle is short, hierarchical lookup table compensation is preferred to reduce the amount of online calculation. The compensation gain, direction correction, and compensation upper limit factor for all paths constitute the path compensation parameter set.

[0074] S3.3: The path compensation parameter set is used to classify and summarize the path compensation parameters for different users and under different working conditions to form a multi-user shared compensation parameter cluster.

[0075] The path compensation parameter set is categorized and summarized according to four levels of rules: joint type, load level, action stage, and contact clamping state. For a path compensation parameter to be classified, it is first determined whether it belongs to an existing shared compensation parameter cluster: if the difference between the parameter's characteristics and the central characteristics of an existing cluster is within a preset threshold range (e.g., the difference in torque compensation gain does not exceed 0.15, and the difference in direction correction does not exceed 0.09 radians), then it is assigned to that cluster, and the cluster center and dispersion are updated; if the difference with all existing clusters exceeds the threshold, then a new shared compensation parameter cluster is created for the parameter. The key here is to distinguish whether the force transmission distortion characteristics exhibited by different users under similar human-machine coupling conditions are similar.

[0076] In practice, all path compensation gains under the same joint type are first normalized and sorted, and then similarity aggregation is performed by combining the direction correction and the compensation upper limit factor. Threshold aggregation can be used: when the difference in torque compensation gain between two records does not exceed 0.15, the difference in direction correction does not exceed 0.09 radians, and the difference in compensation upper limit factor does not exceed 0.1, they can be grouped into the same cluster. For each cluster, its central compensation value and dispersion are then calculated. Clusters with large dispersion indicate significant differences between different users under similar working conditions and should be further subdivided; clusters with small dispersion indicate good sharing and can be directly used as multi-user shared compensation parameter clusters. In this process, classification should not be based solely on user body shape or weight. What truly determines compensation behavior is often the contact compression method and load stage, not simply body shape. Therefore, the focus of classification should always revolve around path distortion characteristics, rather than user attributes. For each compensation parameter cluster, the cluster center torque compensation gain, cluster center direction correction, cluster center compensation upper limit factor, number of samples within the cluster, and cluster dispersion index need to be retained. The number of samples within a cluster should be no less than 5; if there are fewer than 5 samples, it can be temporarily listed as a cluster to be expanded and will not be used directly for subsequent shared mapping.

[0077] Specifically, for the k-th compensation parameter cluster, its cluster center moment compensation gain Calculated using a reverse weighted average method, records with greater overall distortion have a smaller impact on cluster centers. ; in, This represents the torque compensation gain of the nth path record falling into this cluster. This indicates the overall distortion level of the nth record. Cluster center direction correction amount. The reverse weighted average is also used: ; in, This represents the direction correction amount for the nth record, in radians. Cluster dispersion. Then, taking into account both gain dispersion and direction dispersion: ; in, Indicates the number of records in this cluster. The direction correction normalization constant is set to 0.09 to 0.35 radians. This dispersion index quantifies whether the cluster is stable or not and whether it can be used as a shared parameter.

[0078] Optionally, for new users or situations where the current operating condition characteristics match multiple existing compensation parameter clusters to a certain extent, a weighted fusion method can be used instead of single-cluster matching. Specifically, the distance between the current operating condition characteristics, including load level, action stage, contact clamping state, etc., and the center characteristics of each compensation parameter cluster is calculated. The top N clusters with the smallest distances are selected, where N is 2 or 3. The reciprocal of the dispersion of each cluster is used as the fusion weight to perform a weighted average of the cluster center torque compensation gain and the cluster center direction correction, resulting in the fused compensation parameters. The dispersion of the fused parameters is the weighted root mean square of the dispersion of each cluster. This method can further improve the compensation smoothness and adaptation speed in multi-user self-learning scenarios.

[0079] This sub-step truly embodies the meaning of multi-user self-learning. It is not simply about piling up all user data together, but rather about precipitating the compensation experience formed by different users under similar coupled operating conditions into a shared set of compensation parameters.

[0080] S3.4: Establish a joint dimension compensation mapping table based on the multi-user shared compensation parameter cluster to form the force transmission path compensation parameter mapping relationship.

[0081] Compensation mapping tables are established according to the joint dimension. Each mapping table includes at least the joint number, path number, load level, motion stage, contact clamping state, cluster center torque compensation gain, cluster center direction correction, cluster center compensation upper limit factor, and intra-cluster dispersion index. Thus, in the subsequent step S4, the compensation parameters of the corresponding joint and path can be directly called based on the current user's real-time motion state and current working condition identifier.

[0082] The mapping table can be established using a primary key matching plus neighborhood correction mechanism. Primary key matching prioritizes complete matching based on joint number, load level, action stage, and contact clamping state. Neighborhood correction allows for the selection of the cluster with the smallest dispersion within adjacent load levels or adjacent clamping states when a complete match is not found, with a reduction factor applied based on the substitution distance. This avoids situations where the table cannot be found on-site.

[0083] Furthermore, to allow the mapping table to directly serve the controller, the cluster center parameters need to be converted into joint-end compensation drive recommendation values. That is, it's necessary to tell the controller not only the gain but also the maximum compensation it can achieve under the current joint target drive torque. Therefore, the current joint target drive torque should be combined with the cluster center torque compensation gain and the cluster center compensation upper limit factor to calculate the upper limit of the recommended compensation drive torque, which should then be written into the mapping table. For clusters with high dispersion, a confidence constraint should be added to reduce their call priority. The dispersion threshold is set between 0.15 and 0.4; clusters exceeding the upper limit are only considered as candidate clusters, not as default clusters.

[0084] Specifically, for the j-th joint, its theoretical compensation driving torque The result is obtained by multiplying the current target driving torque by the matched cluster center torque compensation gain: ; in, The current target driving torque for the j-th joint; The cluster center torque compensation gain for the matched k-th compensation parameter cluster. The upper limit of the recommended compensation driving torque for this joint. Take the smaller value between the theoretical compensation driving torque and the current target driving torque multiplied by the cluster center compensation upper limit factor: ; in, This is the cluster center compensation upper limit factor for the k-th compensation parameter cluster. `min` indicates taking the smaller of the two values ​​to avoid exceeding the safety limit in the compensation output. The final effective suggested compensation driving torque is written into the compensation mapping table. The suggested upper limit is then reduced based on the cluster dispersion: ; in, Let be the dispersion of the k-th compensation parameter cluster. The reference constant for reducing the dispersion is taken as 0.1 to 0.3. The above calculation first obtains the theoretical compensation, then is constrained by the upper limit of compensation, and finally reduces it according to whether the shared parameter cluster itself is stable, so as to obtain the joint compensation recommendation value that can compensate, does not overshoot, and has shared reliability.

[0085] In the technical solution of this disclosure, based on the identified attenuation and offset results, the torque compensation gain, direction correction, and stability constraint are calculated path-by-path. A multi-user shared compensation parameter cluster is then formed through classification and summarization, ultimately establishing a joint-dimensional compensation mapping table. This process precipitates the compensation experience accumulated by different users under similar human-machine coupling conditions into reusable shared parameters, enabling the system to quickly match compensation strategies without relying on individual user characteristics. This achieves a leap from single-user to multi-user self-learning, significantly improving the adaptability and generalization ability of the control method.

[0086] S4: Based on the joint dimension compensation mapping table and the compensation parameter mapping relationship, match it with the current user's real-time motion state, perform adaptive control parameter tuning, and output the corrected control command.

[0087] Perform adaptive control parameter tuning based on compensation mapping. This step matches the set of compensation parameters with the current user's real-time motion state, corrects the control parameters according to the compensation mapping relationship, ensuring that the driving force achieves the target effect after passing through the actual transmission path, and outputs the corrected control command. This is specifically achieved through the following sub-steps.

[0088] S4.1: Perform joint-level matching between the current user's real-time motion state and the joint dimension compensation mapping table to form the current working condition compensation call set.

[0089] For the current exoskeleton operation process of the user, the target motion angle, actual motion angle, angular velocity, target driving torque, load level, action stage, and contact clamping state of each joint are read in real time. This data reflects the specific working condition of the human-machine coupling at the current moment. Then, using the joint number as the main index, the compensation record consistent with the current working condition is retrieved joint by joint in the compensation mapping table output in step S3. The retrieval order prioritizes a complete match, that is, first satisfying the load level, action stage, and contact clamping state simultaneously; if the complete match result is empty, the neighboring correction record is called according to the principle of prioritizing adjacent load levels, followed by adjacent contact clamping states, and ensuring that the action stage does not cross segments for substitution.

[0090] In practical implementation, the load level is divided into three or five levels according to the ratio of the target driving torque to the rated driving torque; the action phase should at least distinguish the lifting phase, steady-state transport phase, and unloading phase; the contact clamping state can be mapped from the average contact pressure in step S1, and it is recommended to divide it into three levels: low clamping, medium clamping, and high clamping. If the same joint corresponds to more than two main force transmission paths, the compensation call set for the current working condition should simultaneously retain the compensation gain, direction correction amount, recommended upper limit of compensation driving torque, and dispersion reduction information of each path, so as to provide input for the next sub-step to calculate the joint-level target drive output.

[0091] Therefore, the static compensation mapping table in step S3 is transformed into a current operating condition compensation call set bound to the current user, current action state, and current load conditions. This sub-step completes the conversion of historical shared compensation experience into compensation objects that can be called in the current control scenario.

[0092] S4.2: Calculate the joint-level target compensation driving torque based on the current working condition compensation call set, and form the joint target driving output set.

[0093] For each joint, first read the current target driving torque, the cluster center torque compensation gain given by the mapping table, the suggested upper limit of the compensation driving torque, and the dispersion reduction. Then, combine these parameters to calculate the theoretical compensation driving torque, effective compensation driving torque, and target driving output of the current joint. If the joint has multiple main force transmission paths, the theoretical compensation driving torque of each path should be calculated separately, and then weighted and summarized according to the proportion of the output transmission torque at the end of the path to obtain the joint-level comprehensive compensation driving torque. The core of this process is that step S3 establishes the path-level compensation mapping relationship, while step S4 outputs the joint-level control command. Therefore, the multi-path compensation results must first be uniformly converted into joint-end output quantities.

[0094] To avoid compensation overshoot, the calculation must be subject to two constraints simultaneously: first, the upper limit of the suggested compensation driving torque given in step S3; and second, the current allowable output limit of the joint, typically taken as 0.8 to 1 times the rated driving torque. If both exist simultaneously, the smaller value is taken as the final effective compensation upper limit. This ensures that even if the human-machine coupling distortion suddenly amplifies at a certain moment during heavy-duty handling, the joint output will not jump abruptly due to an infinite increase in the compensation amount.

[0095] Specifically, for the j-th joint, its theoretical compensation driving torque The current target driving torque is amplified by the compensation gain and reduced according to the shared cluster stability: ; in, This represents the current target driving torque of the j-th joint; This represents the compensation gain obtained by matching the j-th joint; This represents the dispersion of the compensation parameter cluster matched to the j-th joint; A reference constant for reducing the dispersion is set to 0.1 to 0.3. Then, the driving torque is effectively compensated. Take the smaller value between the theoretical compensation driving torque and the suggested upper limit of the compensation driving torque: ; in, This represents the upper limit of the suggested compensation driving torque in the compensation mapping table corresponding to the j-th joint; min indicates taking the smaller of the two values. Finally, the target driving output for this joint is... The sum of the original target driving torque and the effective compensation driving torque: ; If a joint has multiple main paths, the theoretical compensation driving torque can be calculated for each path using the formula above, and then summed according to the path weights to form a unified compensation driving torque for that joint. The path weights directly adopt the path proportions or correction allocation coefficients from step S3 to avoid redefining new meanings. The target driving outputs of all joints constitute the joint target driving output set.

[0096] S4.3: Based on the direction correction amount in the joint target drive output set and the current working condition compensation call set, the joint control parameters are offset and corrected to form a joint compensation control parameter set.

[0097] In multi-joint passively coupled structures, simply increasing the joint driving torque cannot completely solve the problem, as some distortion originates from force transmission direction offset. Therefore, this sub-step requires combining the direction correction obtained in step S3 with the current joint motion state to offset the joint control parameters. Specifically, the direction correction is converted into a control correction factor for the current joint and further applied to the proportional and damping adjustments in the controller, making the joint output not only larger but also more consistent with the true force direction of the current human-machine coupling path. The proportional adjustment is mainly used to enhance the joint's tracking of the target motion; the damping adjustment is mainly used to suppress transient oscillations caused by compensation.

[0098] Regarding parameter settings, the proportional adjustment base value is given according to the joint type, for example, 80 to 200 Nm / radian for the hip joint, 50 to 150 Nm / radian for the knee joint, and 20 to 80 Nm / radian for the ankle joint. The damping adjustment base value is 5 to 40 Nm / s / radian. The larger the directional correction, the more obvious the path deviation. In this case, the proportional adjustment should be increased appropriately, and the damping adjustment should also be increased simultaneously to prevent shock after output correction.

[0099] set up The target drive output for the j-th joint in the target drive output set. Specifically, for the j-th joint, the corrected proportional adjustment amount. The result is obtained by multiplying the proportional adjustment base value by the direction correction enhancement term, and then dividing by the angular velocity suppression term: ; in, This represents the proportional adjustment base value of the j-th joint; This represents the direction correction amount obtained from matching the j-th joint; The direction correction normalization constant is set to 0.09 radians to 0.35 radians; Indicates the current joint angular velocity; The angular velocity balance reference constant is taken from 0.5 to 3 radians per second. This formula improves the proportional adjustment capability according to the directional correction requirements, while appropriately suppressing proportional amplification at higher speeds.

[0100] Corrected damping adjustment The result is obtained by multiplying the damping adjustment base value by the direction correction enhancement term, and then by multiplying it by the angle deviation enhancement term: ; in, This represents the damping adjustment base value of the j-th joint; Indicates the current joint angle deviation; The reference constant for normalizing the angle deviation is taken from 0.03 radians to 0.2 radians. This formula simultaneously improves the damping suppression capability when both directional offset and angle deviation are large.

[0101] Finally, the corrected control output of the j-th joint Add proportional and damping adjustment terms to the target drive output: ; in, The target drive output obtained from sub-step S4.2; This represents the current joint angular velocity deviation. The formula synthesizes the compensated driving torque, joint following correction term, and damping suppression term into the final joint control output. The control tuning process involved here is essentially online adaptive tuning based on mapping parameters. Its execution order remains fixed: first, the mapping table is called; then, the compensated driving torque is calculated; then, the control parameters are corrected; and finally, the control output is synthesized. This avoids confusion in parameter meanings caused by cross-modification within the same cycle. This sub-step directly completes the technical action of correcting the control parameters according to the compensation mapping relationship, so that the compensation no longer stops at the level of torque amplification but enters the level of internal controller parameters, truly possessing engineering control significance.

[0102] S4.4: Based on the joint compensation control parameter set, perform continuous constraints and control command generation on the corrected control output, and output the corrected control command.

[0103] Since sub-steps S4.2 and S4.3 have already compensated and tuned the joint output, directly sending all the calculation results of the current cycle to the driver could easily cause jumps between adjacent control cycles. Therefore, continuity constraint processing is required before finally outputting the control command.

[0104] In practice, the corrected control output for the current cycle of the same joint is compared with the control output executed in the previous cycle. If the difference exceeds the upper limit of allowable variation, the value is truncated according to the upper limit; otherwise, the current value is retained. The upper limit of allowable variation is set according to the joint type: 10 to 40 Nm per cycle for the hip joint, 8 to 30 Nm per cycle for the knee joint, and 3 to 15 Nm per cycle for the ankle joint. For joints that approach the upper limit of variation for three or more consecutive cycles, the proportional adjustment can be automatically reduced by 5% to 15% to suppress the cumulative jump trend.

[0105] Specifically, for the j-th joint, the current cycle correction is calculated first, and then the control output is controlled. Control output executed in the previous cycle The difference magnitude between : ; Then determine the actual allowable change for this period. The difference amplitude and the upper limit of the preset period change The smaller value in: ; Finally, smoothly transition the current output to the new control command value according to the allowed change ratio. ,but ,otherwise: ; in This represents the final control command value issued by the j-th joint. The calculation process first calculates how much to change in the current cycle, then determines the maximum allowable change, and finally smoothly transitions the current output to the new control command value according to the allowable ratio, thus preserving the compensation effect while avoiding control jumps.

[0106] Subsequently, the outputs of each joint after continuity constraints are uniformly encoded into drive control commands, including joint number, target drive torque value, control mode identifier, and compensation call identifier for this cycle, and sent to the actuator. This output is the final result of step S4 and will serve as one of the direct inputs for control effect feedback and force transmission path model update in step S5. In the technical solution of this embodiment, the current user's real-time motion state is matched with the compensation mapping table, the joint-level target compensation drive torque is calculated, and the proportional adjustment amount and damping adjustment amount are offset and corrected according to the direction correction amount, finally outputting the control commands after continuity constraints. This step transforms historical shared compensation experience into the tuning action of the current control cycle in real time, not only compensating for torque value gaps but also actively correcting transmission direction deviations, enabling the exoskeleton drive output to overcome transmission distortion caused by flexible connections and structural gaps, achieving stable following and precise force matching of the human body under heavy load conditions.

[0107] S5: Based on the difference between the actual motion result after the execution of the modified control command and the target motion state, perform control effect feedback and force transmission path model update, and output continuous self-tuning results.

[0108] The process involves feedback on the control effect and updating the force transmission path model. This step dynamically updates the force transmission path model and compensation parameters based on the difference between the actual motion result and the target motion state, achieving continuous self-tuning and optimization of the control parameters. This is implemented through the following sub-steps.

[0109] S5.1: Based on the joint deviation response after the execution of the modified control command, extract feedback information to form a feedback evaluation base set.

[0110] After each control cycle, the target angle, actual angle, target angular velocity, actual angular velocity, target driving torque, actual output driving torque, and actual human reaction force of each joint are collected and aligned with the corrected control command issued in step S4 along the same time axis to form a joint-level feedback evaluation record. Subsequently, the angle following deviation, velocity following deviation, and torque action deviation are calculated for each joint to obtain a feedback evaluation base set that directly reflects the control effect. To avoid the amplified impact of accidental shocks on the update results, the feedback evaluation window is set to 5 to 20 consecutive control cycles; a shorter window is used during the heavy-load lifting stage, and a longer window is used during the steady-state handling stage.

[0111] Specifically, for the j-th joint, its motion deviation index It combines angle following deviation and speed following deviation: ; in, Indicates the target angle; Indicates the actual angle; Indicates the target angular velocity; Indicates the actual angular velocity; The velocity deviation conversion constant is taken as 0.01 seconds to 0.2 seconds to ensure that the dimensions of the second term after conversion are consistent with the square of the angle. Torque deviation. It is then given directly by the absolute value of the difference between the target driving torque and the actual output driving torque: ; in, Driven torque for the target This represents the actual output driving torque. These two basic deviations reflect whether the action kept up and whether the force was delivered properly, respectively, providing feedback for subsequent path model correction. The deviation records of all joints constitute the feedback evaluation base set.

[0112] S5.2: Based on the feedback evaluation base set, the path mismatch intensity is calculated inversely to form the path correction quantity set.

[0113] For each joint's main force transmission path, the degree of mismatch remaining after the current control is determined by combining the main distortion segment number in step S2 and the compensation call record in step S3. If the motion deviation is small but the torque deviation is large, it indicates that the main problem is still in force transmission attenuation; if both the motion deviation and torque deviation are large, it indicates that neither path attenuation nor directional offset has been adequately compensated. Based on this, update values ​​are generated for the path equivalent stiffness, path allocation coefficient, and directional correction, forming a path correction value set. A gradual correction method is used during updates, and the correction amplitude in a single cycle should not be too large; the path equivalent stiffness correction ratio is controlled between 2% and 10%, and the path allocation coefficient correction ratio is controlled between 3% and 12%.

[0114] Specifically, for the main force transmission path corresponding to the j-th joint, its mismatch strength It is determined by both the torque deviation and the motion deviation: ; in, This is the deviation of the torque action; This represents the actual output driving torque. To prevent zero equilibrium torque constant, a value of 1 Nm to 10 Nm is used; As an indicator of motion deviation; The motion deviation normalization constant is taken from 0.02 radians to 0.2 radians. This mismatch intensity reflects the degree of distortion remaining after the current control.

[0115] Furthermore, the mismatch intensity is compressed to the update weight range of approximately 0 to 1 to obtain the update weight. : ; This weight is used to control the correction magnitude during subsequent model updates. The above calculation reinterprets the remaining control error as the degree of path mismatch, making step S5 not a general parameter tuning, but a targeted correction back to the force transmission path model itself. The mismatch intensity of all paths and the update weights constitute the path correction set.

[0116] S5.3: Update the force transmission path model based on the path correction set to form an updated path expression model.

[0117] The path equivalent stiffness, path allocation coefficient, force transmission response coefficient, and force transmission stability coefficient in the initial force transmission path model established in step S1 are recursively updated. For paths with high mismatch strength, their equivalent stiffness and path allocation coefficient are corrected first; for paths with residual directional deviations in multiple consecutive windows, their directional correction parameters are corrected simultaneously. The update adopts a recursive method of retaining old values ​​and injecting new values, which preserves existing working condition experience while incorporating new feedback from current users and under current load conditions.

[0118] Specifically, for the j-th path, the updated equivalent stiffness The result is obtained by multiplying the equivalent stiffness before the update by the stiffness growth factor: ; in, The equivalent stiffness before the update; The stiffness update factor is taken as 0.02 to 0.1. To update the weights. Updated path assignment coefficients. A weighted average approach is then used to strike a balance between the original allocation coefficients and the real-time path allocation estimate obtained by back-calculating based on control feedback in the current window: ; in, Assign coefficients to the paths before the update; The estimated value of the real-time path allocation obtained by back-calculation based on control feedback for the current window is taken as the proportion of the output torque at the end of the path in the current window to the sum of the output torques at the end of all paths under the same joint; The updated coefficients for the allocation coefficients are set to a value between 0.03 and 0.12. To update the weights, the above update, without overturning the original model, gradually writes the current working condition feedback back into the model, so that the path expression model gradually approximates the real human-machine coupling structure.

[0119] Optionally, to avoid frequent model fluctuations under disturbances or noise, a trigger threshold can be set for updates. The update is only executed when the update weight is greater than the preset threshold; otherwise, it is skipped. The preset threshold ranges from 0.1 to 0.3. Furthermore, for paths that have not been updated multiple times consecutively, their equivalent stiffness can be periodically slightly reduced back to the initial nominal value, with each reduction being 2% to 5% of the current value, to prevent the model from becoming rigid and losing its adaptability to new operating conditions.

[0120] The updated path parameters constitute the updated path representation model.

[0121] S5.4: Write back the compensation parameters to the joint dimension compensation mapping table according to the updated path expression model, update the call record information in the joint dimension compensation mapping table, and output the continuous self-tuning result.

[0122] The path equivalent stiffness, path allocation coefficient, and direction correction correlation parameters updated in sub-step S5.3 are re-entered into the compensation mapping logic of step S3. The compensation gain, direction correction amount, and compensation upper limit factor of each joint are recalculated, and the new results overwrite the corresponding records in the original compensation mapping table. At the same time, the call record information of each compensation record in the joint dimension compensation mapping table is updated, including the number of calls and the call timestamp. When the same user calls the same path record more than three times consecutively under the same load level and action stage, the call priority of that record is increased; for shared compensation clusters with persistently high dispersion, their default call level is decreased.

[0123] The specific implementation method is as follows: Call Priority: Each compensation record has a priority value in the mapping table, ranging from 1 to 5, with an initial value of 3. When the same user calls the same path compensation record 3 times consecutively under the same load level and action phase, the priority of that record is increased to 4; when the number of consecutive calls reaches 5 or more, it is increased to 5. During joint-level matching in step S4.1, if multiple records match the current working condition simultaneously, the record with the higher priority is selected first.

[0124] Default Invocation Level: Each compensation record has three default invocation levels: A, B, and C, with level B being the initial level. This applies when the dispersion of the shared compensation parameter set... If the dispersion is greater than 0.3 for more than 10 control cycles, the default call level for the corresponding record in that cluster will be downgraded to level C. Level C records will only be called when there is a perfect match and no other records are available. Once the dispersion falls below 0.3 and remains below 0.3 for 5 cycles, it will revert to level B.

[0125] The final output includes the updated path representation model, the updated set of compensation parameters, and the current cycle self-tuning state flag, which can be directly called in subsequent control cycles.

[0126] Thus, all five main steps of the multi-user self-learning adaptive exoskeleton parameter self-tuning control method have been completed. Step S1 constructs the force transmission path model; step S2 identifies attenuation and offset; step S3 establishes a compensation mapping relationship; step S4 performs adaptive control parameter tuning; and step S5 provides feedback and updates the model, forming a complete closed-loop control system from perception to execution to learning. This method enables the exoskeleton to automatically adapt to changes in the human-machine coupling structure under different users, loads, and working conditions, continuously optimizing control parameters, effectively solving the force transmission distortion problem under multi-joint passive coupling structures, and achieving stable following and precise force matching under heavy-load conditions.

[0127] In the technical solution of this disclosure, the path mismatch strength is calculated by back-calculating the angle following deviation and torque action deviation after control execution. The model parameters, such as the path equivalent stiffness and distribution coefficient, are then recursively updated, and the updated parameters are written back to the compensation mapping relationship. This step forms a complete closed loop from perception to execution to learning, enabling the force transmission path model to gradually approximate the real human-machine coupling structure. Simultaneously, the compensation parameters continuously self-tune and optimize during use, ensuring that the exoskeleton maintains good control performance under different users, loads, and operating conditions.

[0128] According to embodiments of this disclosure, an electronic device is also provided, which may include a processor, a communications interface, a memory, and a communication bus, wherein the processor, the communications interface, and the memory communicate with each other via the communication bus. The processor can invoke logical instructions stored in the memory to execute the methods provided in the above embodiments.

[0129] Furthermore, the logical instructions in the aforementioned memory can be implemented as software functional units and sold or used as independent products, and can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this disclosure, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this disclosure. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0130] On the other hand, this disclosure also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to perform the methods provided in the above embodiments.

[0131] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0132] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0133] It should be understood that the above embodiments are only used to illustrate the technical solutions of this disclosure, and not to limit them; although this disclosure has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or make equivalent substitutions for some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this disclosure.

Claims

1. A multi-user self-learning adaptive exoskeleton parameter self-tuning control method, characterized in that, The method includes: S1: Spatial mapping of the multi-point contact relationship between the exoskeleton and the human body, constructing the distribution relationship of joint driving force in each contact path, and obtaining the initial force transmission path model; S2: Based on the initial force transmission path model, perform actual force transmission attenuation calculation and offset identification in the force transmission path to form a set of joint-level force transmission offset descriptions; S3: Based on the joint-level force transmission offset description set, construct the force transmission path compensation parameter mapping relationship including torque compensation gain, direction correction amount and stability constraint amount, and establish a joint dimension compensation mapping table; S4: Based on the joint dimension compensation mapping table and the compensation parameter mapping relationship, match it with the current user's real-time motion state, perform adaptive control parameter tuning, and output the corrected control command; S5: Based on the difference between the actual motion result after the execution of the modified control command and the target motion state, perform control effect feedback and force transmission path model update, and output continuous self-tuning result; S1 includes: Collect data on exoskeleton joint driving torque, human joint reaction force, and pressure, displacement, and spatial position at each contact point to form a unified and correlated sampling set; Based on the unified associated sampling set, the equivalent contact stiffness, torque transmission capacity and contact force transmission capacity index of each contact point are calculated to form a mechanical characterization set of the contact point. Based on the contact point mechanical characterization set, perform multi-point contact space mapping to construct the distribution relationship of joint driving force in each contact path and obtain the corrected distribution coefficient; Based on the corrected allocation coefficients and the displacement response of each contact path, combined with the joint hierarchy numbering rules, a set of joint force transmission path expressions is generated to obtain the initial force transmission path model. S2 includes: The initial force transmission path model is segmented and expanded to form a path segment force transmission calculation sequence; Based on the path segment force transmission calculation sequence, the actual force transmission attenuation and attenuation ratio of each path segment are calculated to form a segment attenuation calculation result set. The transmission direction offset angle of each segment is calculated based on the segmented attenuation calculation result set to form a path direction offset result set; By combining the attenuation ratio in the segmented attenuation calculation result set, the directional offset angle in the path direction offset result set, and the actual human body reaction force, the offset and distortion degree in the path are identified to form the joint-level force transmission offset description set. S3 includes: The joint-level force transmission offset description set is decomposed into an intra-joint decomposition to form a path compensation modeling base set; Based on the path compensation modeling base set, the path torque compensation gain, direction correction amount and stability constraint amount are calculated to form a path compensation parameter set; The path compensation parameter set is used to classify and summarize the path compensation parameters for different users and under different working conditions, forming a multi-user shared compensation parameter cluster; The joint dimension compensation mapping table is established based on the multi-user shared compensation parameter cluster.

2. The method according to claim 1, characterized in that, After classifying and summarizing the path compensation parameters for different users and under different operating conditions to form a multi-user shared compensation parameter cluster, the method further includes: When the user's current operating condition characteristics match multiple existing compensation parameter clusters, calculate the distance between the current operating condition characteristics and the central characteristics of each compensation parameter cluster; Select the N compensation parameter clusters with the smallest distance, where N is 2 or 3; Using the reciprocal of the dispersion of each compensation parameter cluster as the fusion weight, the cluster center moment compensation gain and the cluster center direction correction are weighted and averaged to obtain the fused compensation parameters.

3. The method according to claim 1, characterized in that, S4 includes: The current user's real-time motion state is matched with the joint dimension compensation mapping table at the joint level to form the current working condition compensation call set; Calculate the joint-level target compensation driving torque based on the current working condition compensation call set, and form a joint target driving output set; Based on the direction correction amount in the joint target drive output set and the current working condition compensation call set, the joint control parameters are offset and corrected to form a joint compensation control parameter set. Based on the joint compensation control parameter set, continuous constraints and control commands are generated for the corrected control output, and the corrected control commands are output.

4. The method according to claim 1, characterized in that, S5 includes: Based on the joint deviation response after the execution of the modified control command, feedback information is extracted to form a feedback evaluation base set; Based on the feedback evaluation base set, the path mismatch intensity is calculated inversely to form a path correction set; The force transmission path model is updated based on the path correction set to form an updated path expression model. The updated path representation model is used to write back the compensation parameters to the joint dimension compensation mapping table, and the call record information in the joint dimension compensation mapping table is updated to output the continuous self-tuning result.

5. The method according to claim 4, characterized in that, After the output continues to self-tuning, it also includes: Based on the number of times the same user continuously calls the same path compensation record under the same load level and action phase, the calling priority of that record in the compensation mapping table is increased; Based on the degree to which the dispersion of the shared compensation parameter cluster is consistently large, the default call level of the compensation record corresponding to that cluster is reduced.

6. An electronic device, characterized in that, The electronic device includes a memory and at least one processor, the memory storing a computer program, and the processor executing the computer program to implement the multi-user self-learning adaptive exoskeleton parameter self-tuning control method according to any one of claims 1-5.

7. A computer storage medium, characterized in that, It stores a computer program, which, when executed, implements the multi-user self-learning adaptive exoskeleton parameter self-tuning control method according to any one of claims 1-5.

Images