Adaptive control method, system and medium for posture correction

Abstract

The application provides an adaptive control method, system and medium for posture correction, relating to the technical field of posture capture, the method comprising: collecting the motion posture of a target user and making a correction decision based on a posture correction component according to a multi-source sensing array of an anti-gravity treadmill assembly; determining a first correction posture based on a posture correction sequence, triggering a control decision component, executing a decomposition of a posture error component, a lower three-element single-thread decision and a physical consensus protocol coordination, and determining a first posture correction parameter; controlling and driving the anti-gravity treadmill, executing multi-stage posture correction regulation, and terminating the posture correction control process. The application solves the technical problem that, in the prior art, the interaction between multiple driving mechanisms is not considered in posture error decomposition, resulting in intervention conflicts and space-time mismatches, thereby affecting the accuracy and efficiency of posture correction. By optimizing the collaborative work of multiple driving mechanisms, the accuracy and efficiency of posture correction are improved.

Description

Technical Field

[0001] This application relates to the field of attitude capture technology, and specifically to adaptive control methods, systems and media for attitude correction. Background Technology

[0002] Existing anti-gravity treadmills typically employ independent closed-loop or simple rule-based collaborative control strategies for each actuator during posture correction. However, this fragmented control architecture makes it difficult to precisely match the forces and torques applied to the human body by each mechanism in time and space. This leads to conflicting intervention commands and mismatched control signals during posture correction, affecting the accuracy of posture correction and impacting the efficiency of the anti-gravity treadmill. Because the errors cannot be precisely decomposed and there is a lack of adaptive and intelligent adjustment mechanisms, traditional systems often cannot flexibly adjust their control strategies when faced with the personalized needs of different users or complex motion states, resulting in a lengthy and unstable posture correction process.

[0003] In summary, existing technologies suffer from technical problems such as intervention conflicts and spatiotemporal mismatches caused by the failure to consider the interaction between multiple driving mechanisms in attitude error decomposition, which affects the accuracy and efficiency of attitude correction. Summary of the Invention

[0004] The purpose of this application is to provide an adaptive control method, system, and medium for attitude correction, in order to solve the technical problem in the prior art that intervention conflicts and spatiotemporal mismatches are caused by the failure to consider the interaction between multiple drive mechanisms in attitude error decomposition, thereby affecting the accuracy and efficiency of attitude correction.

[0005] To achieve the above objectives, this application provides an adaptive control method, system, and medium for attitude correction.

[0006] In a first aspect, this application provides an adaptive control method for posture correction, which is implemented through an adaptive control system for posture correction. The adaptive control method for posture correction includes: acquiring the motion posture of a target user based on a multi-source sensor array mounted on an anti-gravity treadmill and making correction decisions based on a posture correction component to determine a posture correction sequence, wherein a developmental posture tree embedded in the posture correction component is used as the decision basis; determining a first correction posture based on the posture correction sequence, triggering a control decision component to execute a ternary single-thread decision and physical consensus protocol coordination under the decomposition of posture error components to determine a first posture correction parameter, wherein the posture error components and the ternary single-thread are defined based on a pneumatic weight reduction device, a balance auxiliary arm, and a treadmill drive mechanism; and controlling and driving the anti-gravity treadmill according to the first posture correction parameter, terminating the posture correction control process by executing multi-stage posture correction regulation under posture correction sequence polling, wherein the posture correction component and the control decision component are embedded plug-ins of the anti-gravity treadmill central control system.

[0007] Optionally, a basic standing balance pattern is determined, wherein the basic standing balance pattern is characterized as a first support phase and a first swing phase based on the posture point cloud under the gait cycle; taking the basic standing balance pattern as the root node, the support phase and swing phase are developed according to a first motion complexity to determine a first leaf node layer, wherein the first leaf node layer contains at least one posture pattern that satisfies the first motion complexity, and each posture pattern is identified by a support phase and a swing phase based on the posture point cloud under the gait cycle; taking the first leaf node layer, the support phase and swing phase are developed according to a second motion complexity, and the developmental posture tree is generated through hierarchical development; and a posture correction component is constructed based on the developmental posture tree.

[0008] Optionally, the motion posture signal of the target user is collected based on the multi-source sensor array assembled on the anti-gravity treadmill; the data platform receives the motion posture signal, performs preprocessing and feature extraction, and determines the posture features, wherein the posture features include real-time posture point cloud, gait phase, symmetry and stability indices; the posture correction component is triggered to perform similarity matching and pruning of the posture features in the developmental posture tree to determine the posture correction sequence.

[0009] Optionally, a target pose node is determined by performing similarity matching on the pose features in the developmental pose tree; the target pose node is then used to perform hierarchical forward and backward propagation pruning in the developmental pose tree as a pose correction sequence; wherein, in the pose correction sequence, the sequence node weights are reduced in order with the target pose node as the center.

[0010] Optionally, for the pre-control components of the anti-gravity treadmill, the adjustment dimensions and scope of each pre-control component are defined, and an adjustment decision module is built through parallel training. The adjustment decision module includes a support force field modulator, a spatial torque guide, and a ground dynamic guide corresponding to the pre-control components. A first decomposition layer is deployed based on attitude error component decomposition, a second decision layer is deployed based on independent component decisions of the adjustment decision module, and a third consensus layer is deployed by introducing a physical consensus protocol to build the control decision component.

[0011] Optionally, for the air pressure weight reduction device, a first control dimension and a first range of action are defined, and a support force field modulator is built. The first control dimension includes the vertical support force distribution, pressure center offset, and weight reduction gradient, and the first range of action is a gravity scalar field. For the balance auxiliary arm, a second control dimension and a second range of action are defined, and a spatial torque guide is built. The second control dimension includes six-dimensional torque, contact point position, and compliance stiffness, and the second range of action is a spatial pose vector field. For the treadmill drive mechanism, a third control dimension and a third range of action are defined, and a ground dynamic guide is built. The third control dimension includes the surface velocity field, local deformation, and friction characteristics, and the third range of action is the tensor field of the foot-ground interaction interface.

[0012] Optionally, based on the attitude correction sequence, a first corrected attitude is determined using a first neighborhood of the target attitude node; the first corrected attitude is input into the first decomposition layer to determine a first attitude error component; wherein, the decomposition of the attitude error component includes: cross-mapping the attitude features with the first corrected attitude, measuring the vertical component, lateral balance component, and propulsion dynamics component based on the cross-mapping error; associating the vertical component with the air pressure weight reduction device, associating the lateral balance component with the balance auxiliary arm, and associating the propulsion dynamics component with the treadmill drive mechanism.

[0013] Optionally, the first attitude error component is transferred to the control decision module in the second decision layer to perform directional input decision and determine the ternary control parameters; according to the third consensus layer, the ternary control parameters are conflict-resolved and spatiotemporally phase-coordinated to determine the first attitude correction parameters.

[0014] Secondly, this application also provides an adaptive control system for posture correction, used to execute the adaptive control method for posture correction as described in the first aspect, wherein the adaptive control system for posture correction includes: a correction decision module, used to collect the motion posture of a target user and make correction decisions based on a posture correction component according to a multi-source sensor array mounted on an anti-gravity treadmill, and determine a posture correction sequence, wherein a developmental posture tree embedded in the posture correction component is used as the decision basis; and a parameter determination module, used to determine a first correction posture based on the posture correction sequence, and trigger... A control decision component is established to perform ternary single-thread decision-making and physical consensus protocol coordination under the decomposition of attitude error components to determine the first attitude correction parameter. The attitude error component and the ternary single-thread are defined based on the air pressure weight reduction device, the balance auxiliary arm, and the treadmill drive mechanism. A multi-stage attitude correction control module is used to control and drive the anti-gravity treadmill according to the first attitude correction parameter. By executing multi-stage attitude correction control under attitude correction sequence polling, the attitude correction control process is terminated. The attitude correction component and the control decision component are embedded plug-ins of the anti-gravity treadmill central control.

[0015] Thirdly, a computer-readable storage medium storing a computer program that, when executed, implements the steps of the adaptive control method for attitude correction described in any of the first aspects above.

[0016] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0017] By collecting motion posture data from the multi-source sensor array mounted on the anti-gravity treadmill and making correction decisions based on the posture correction component, a posture correction sequence is determined. The developmental posture tree embedded in the posture correction component serves as the decision-making basis. A first corrected posture based on the posture correction sequence is determined, triggering the control decision component to execute a ternary single-thread decision-making process based on the decomposition of posture error components and coordination with a physical consensus protocol. This determines the first posture correction parameter, where the posture error components and the ternary single-thread decision-making process are defined based on the air pressure weight reduction device, the balance auxiliary arm, and the treadmill drive mechanism. Based on the first posture correction parameter, the anti-gravity treadmill is controlled and driven. Through multi-stage posture correction regulation under posture correction sequence polling, the posture correction control process is terminated. The posture correction component and the control decision component are embedded plug-ins in the central control unit of the anti-gravity treadmill. In other words, the attitude of the target user is captured by a multi-source sensor array, the attitude correction sequence is determined by the developmental attitude tree embedded in the attitude correction component, the first correction attitude is determined, and the control decision component is triggered to perform the ternary single-thread decision and physical consensus protocol coordination under the decomposition of attitude error components to determine the first attitude correction parameters. The anti-gravity treadmill is controlled and driven, and multi-stage attitude correction regulation is performed until the attitude correction control process is terminated. The collaborative work between multiple drive mechanisms is optimized to improve the accuracy and efficiency of attitude correction.

[0018] The above description is merely an overview of the technical solution of this application. To better understand the technical means of this application and to facilitate its implementation according to the description, and to make the above and other objects, features, and advantages of this application more apparent, specific embodiments of this application are described below. It should be understood that the content described in this section is not intended to identify key or important features of the embodiments of this application, nor is it intended to limit the scope of this application. Other features of this application will become readily apparent through the following description. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely exemplary. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0020] Figure 1 This is a flowchart illustrating the adaptive control method for attitude correction used in this application.

[0021] Figure 2 This is a schematic diagram of the adaptive control system used for attitude correction in this application.

[0022] Figure labeling: Correction decision module 11, parameter determination module 12, multi-stage posture correction and control module 13. Detailed Implementation

[0023] This application addresses the technical problem in existing technologies where the interaction between multiple drive mechanisms is not considered in the attitude error decomposition, leading to intervention conflicts and spatiotemporal mismatches that affect the accuracy and efficiency of attitude correction. It provides an adaptive control method, system, and medium for attitude correction, solving these issues. The method involves capturing the target user's attitude using a multi-source sensor array, determining the attitude correction sequence using a developmental attitude tree embedded in the attitude correction component, identifying the first correction attitude, and triggering a control decision component. This component performs ternary single-threaded decision-making based on the decomposition of attitude error components and coordinates with a physical consensus protocol to determine the first attitude correction parameters. The anti-gravity treadmill is then controlled and managed, undergoing multi-stage attitude correction regulation until the attitude correction control process is terminated. This optimizes the collaborative work between multiple drive mechanisms, improving both the accuracy and efficiency of attitude correction.

[0024] The technical solutions of this application will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. It should be understood that this application is not limited to the exemplary embodiments described herein. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application. It should also be noted that, for ease of description, only the parts related to this application are shown in the accompanying drawings, not all of them.

[0025] Example 1, please refer to the appendix. Figure 1 This application provides an adaptive control method for attitude correction, wherein the adaptive control method for attitude correction is executed by an adaptive control system for attitude correction, and the adaptive control method for attitude correction specifically includes the following steps: Based on the multi-source sensor array assembled on the anti-gravity treadmill, the motion posture of the target user is collected and the correction decision is made based on the posture correction component to determine the posture correction sequence. The developmental posture tree embedded in the posture correction component is used as the decision basis.

[0026] Furthermore, this application also includes the following steps: determining a basic standing balance pattern, wherein the basic standing balance pattern is characterized as a first support phase and a first swing phase based on the posture point cloud under the gait cycle; using the basic standing balance pattern as the root node, developing the support phase and swing phase according to a first motion complexity to determine a first leaf node layer, wherein the first leaf node layer contains at least one posture pattern that satisfies the first motion complexity, and each posture pattern is identified by a support phase and a swing phase based on the posture point cloud under the gait cycle; using the first leaf node layer, developing the support phase and swing phase according to a second motion complexity, and generating the developmental posture tree through hierarchical development; and constructing a posture correction component based on the developmental posture tree.

[0027] Specifically, the basic standing balance pattern is determined, which is the balance posture pattern of the human body when standing within a specific gait cycle. It is characterized by the first support phase and the first swing phase based on the posture point cloud of the gait cycle. During the gait cycle, the human body maintains standing balance during the support phase, while during the swing phase, the other foot leaves the ground and swings. The basic standing balance pattern forms the basis for subsequent movement patterns and is determined by acquiring and analyzing the human body's posture point cloud. The support phase is the stage during gait where one foot contacts the ground and bears the body weight; the swing phase is another stage in the gait cycle, referring to the movement process when the other foot leaves the ground and prepares for the next step.

[0028] The standardized basic standing balance pattern is set as the root node of the entire posture tree. It is further refined based on the first kinematic complexity, and physical simulation and biomechanical modeling are performed on the complete posture point cloud data contained in the root node. The goal of the modeling is to calculate how the theoretical motion trajectories of each joint should adaptively adjust to maintain an efficient and low-impact ideal gait under the newly added complexity parameters. Kinematic complexity is a set of multidimensional parameters that quantify the difficulty of a motion task, including motion speed, percentage of body weight support, complexity of the motion plane, and asymmetry requirements. The higher the complexity, the higher the requirements for balance, strength, and coordination. The second kinematic complexity is more complex than the first. The support and swing phases are developed according to the first kinematic complexity, resulting in multiple new posture patterns from a single root node, which together constitute the first leaf node layer. Each newly generated posture pattern fully contains detailed posture point cloud data for the support and swing phases under new speeds and loads. The first leaf node layer is a set of end-effector posture patterns initially developed from the root node. A posture pattern is a data structure that defines the ideal motion trajectories of all relevant joints and the plantar pressure distribution within a complete gait cycle. Each pattern corresponds to a specific combination of motion complexities.

[0029] Starting with all nodes in the first leaf node layer, a second motion complexity rule is applied—that is, adding a difficulty level standard to the first motion complexity—and the simulation and deduction process is repeated to generate the second layer of leaf nodes. This hierarchical developmental construction process, like tree branching, can be recursively repeated multiple times, ultimately generating a developmental pose tree. The developmental pose tree is a structured representation method based on hierarchical development. The root node of the tree represents the basic standing balance pattern, and the child nodes represent refined motion patterns under different motion complexities. Each layer of nodes represents a different pose pattern, which is defined and divided based on the pose point cloud data under gait cycles. Hierarchical developmental construction is an algorithmic process that uses existing nodes (such as the root node or a certain leaf node layer) as parents, and generates new pose patterns (child nodes) with higher difficulty by adding one or more dimensions of motion complexity parameters, thus expanding the tree structure layer by layer.

[0030] Based on a developmental posture tree, a posture correction component is constructed that can navigate and plan correction paths within the tree according to real-time assessments of the user's abilities. This hierarchical developmental approach ensures that each more challenging posture correction pattern is scientifically derived from lower-difficulty patterns based on biomechanical principles, avoiding unreasonable jumps between training phases and significantly improving training safety and feasibility.

[0031] Furthermore, this application also includes the following steps: acquiring the motion posture signal of the target user based on the multi-source sensor array assembled on the anti-gravity treadmill; receiving the motion posture signal, performing preprocessing and feature extraction, and determining posture features, wherein the posture features include real-time posture point cloud, gait phase, symmetry and stability indices; triggering the posture correction component to perform similarity matching and pruning of the posture features in the developmental posture tree, and determining the posture correction sequence.

[0032] Furthermore, this application also includes the following steps: determining a target pose node by performing similarity matching on the pose features in the developmental pose tree; using the target pose node, performing hierarchical forward and backward propagation-based pruning processing in the developmental pose tree as a pose correction sequence; wherein, in the pose correction sequence, the sequence node weights are reduced in order with the target pose node as the center.

[0033] Specifically, the anti-gravity treadmill is equipped with a multi-source sensor array to synchronously collect the user's raw motion signals during actual training at high frequency, and send them to a data platform in real time for data cleaning and alignment. The multi-source sensor array is a collection of various sensors integrated into the anti-gravity treadmill, typically including: a force measuring platform embedded in the running belt to measure the three-dimensional force and pressure center of the foot; a high-speed infrared optical motion capture camera to track reflective markers attached to key parts of the user's body; an inertial measurement unit worn on the user's torso and limbs to provide acceleration and angular velocity; and treadmill body sensors to record speed, incline, and pressure of the weight-reducing airbags.

[0034] In the data platform, the collected motion posture signals first undergo preprocessing, including time synchronization, filtering and noise reduction, and coordinate system unification. Then, feature extraction is performed, analyzing the signals to extract key posture features such as real-time posture point clouds, gait phase, symmetry, and stability indices. Real-time posture point clouds represent the spatial positions of various body parts; gait phase represents different phases of the user's gait cycle, such as the support phase and swing phase; symmetry represents the degree of coordination between the left and right limbs, especially consistency within the gait cycle; stability indices are used to assess the user's ability to maintain balance during movement, typically measured by indicators such as the smoothness of movement and the stability of the center of gravity.

[0035] The posture correction component is triggered to perform similarity matching of posture features in the developmental posture tree, identifying the target posture node that best matches the current user's posture, representing the user's current posture pattern. It determines the pattern unconsciously exhibited by the user, finding the closest match to a preset ideal pattern in the developmental posture tree. Specifically, it traverses each node in the developmental posture tree, reading the ideal posture point cloud and associated phase information stored in each node under specific training parameters. The distance between the user's real-time data and the template data of each node is calculated. Since gait speeds may have slight differences, direct comparison at specific time points is inaccurate. Therefore, algorithms such as dynamic time warping are typically used to first perform optimal non-linear alignment of two time series, such as the real-time knee joint angle curve, with the target curve stored in the node on the time axis to eliminate the influence of speed differences. Then, the root mean square error or correlation between the aligned sequences is calculated. Symmetry and stability indices are used as scalar values ​​and compared with the expected range stored in the node. A weighted fusion function is used to calculate a total matching score for each tree node, integrating the overall similarity of the joint angle curves, phase alignment, and the degree of conformity of the symmetry / stability indices. The node with the highest score, that is, the predefined ideal pattern that is most similar to the user's current posture features and has the smallest difference, is identified as the target posture node. This means that the system intelligently diagnoses that the gait pattern that the user is currently unconsciously exhibiting is closest to the gait with Y features under X parameters in the knowledge base.

[0036] For example, the current anti-gravity treadmill is set to reduce body weight by 50% at a speed of 2.5 km / h. Collected posture features: Real-time posture point cloud shows that the average dorsiflexion angle of the right ankle joint during mid-stance phase is only -5 degrees (plantar flexion), while the left is 10 degrees (dorsiflexion). The peak flexion angle of the right knee joint at the end of the swing phase is only 35 degrees, lower than the healthy side's 55 degrees. Gait phase detection shows a right gait cycle duration of 1.3 seconds and a left cycle of 1.1 seconds, indicating asymmetry. The stride asymmetry index (right stride length / left stride length) is 0.7. The trunk swing amplitude in the coronal plane is ±8 degrees. DTW alignment and error calculation are performed between the curves of each angle and the curves stored in multiple nodes of the posture tree. Simultaneously, the preset symmetry (e.g., asymmetry index > 0.85) and stability (trunk swing < ±5 degrees) thresholds for each node are checked. Weighted calculations revealed the highest match with a node characterized by a 50% weight loss, a speed of 2.8 km / h, and moderate ankle dorsiflexion deficiency and limited knee flexion. Template data for this node showed an expected ankle dorsiflexion of 0 degrees, a peak knee flexion of 40 degrees, an asymmetry index tolerance range of 0.65-0.8, and a trunk sway tolerance of ±10 degrees. The root mean square error of the joint angle curves between the user's current data and this node's template was the smallest, and all scalar indices fell within its tolerance range. Therefore, this node was identified as the target posture node.

[0037] Starting from the target pose node, pruning is performed on the developmental pose tree using hierarchical forward and backward propagation. From this node, forward propagation searches several layers towards the parent node, while backward propagation explores possible directions towards child nodes. Pruning is performed according to biomechanical coherence rules; for example, nodes requiring abrupt changes in joint range of motion are skipped, or nodes exhibiting regression in symmetry indices are excluded. The remaining nodes are arranged in order of difficulty, forming a personalized pose correction sequence. This sequence defines a step-by-step training roadmap that starts from the user's current state and gradually reverts to a basic pattern or progresses to a higher-order goal, guiding the various actuators in the anti-gravity treadmill to adjust their posture.

[0038] In the posture correction sequence, a dynamic weight is assigned to each node, implementing a weight reduction strategy. This means that nodes immediately before and after the target node receive higher weights, representing the immediate focus for improvement and consolidation, while nodes at the ends of the sequence have lower weights, representing long-term goals or regressive training. Sequence node weight reduction refers to assigning different importance or training priorities to each node in the generated correction sequence. Typically, the target posture node is the center, receiving the highest weight; the weights of nodes moving forward (more difficult) and backward (easier) decrease with increasing distance from the center node, forming a progressive training strategy of focusing on key breakthroughs while consolidating peripheral areas.

[0039] By accurately extracting features and matching similarity between developmental pose trees, personalized pose correction is performed based on the user's movement characteristics, ensuring the accuracy of the correction effect. Pruning simplifies the pose tree structure, and weight reduction is applied to nodes in the correction sequence, effectively improving the computational efficiency of the correction process and reducing unnecessary computation.

[0040] The first corrected posture based on the posture correction sequence is determined, the control decision component is triggered, the ternary single-thread decision and physical consensus protocol coordination under the decomposition of the posture error components are executed, and the first posture correction parameters are determined. The posture error components and the ternary single-thread are defined based on the air pressure weight reduction device, the balance auxiliary arm and the treadmill drive mechanism.

[0041] Furthermore, this application also includes the following steps: for the pre-control components of the anti-gravity treadmill, defining the control dimensions and scope of each pre-control component, building a control decision module through parallel training, wherein the control decision module includes a support force field modulator, a spatial torque guide, and a ground dynamic guide corresponding to the pre-control components; deploying a first decomposition layer based on attitude error component decomposition, deploying a second decision layer based on independent component decisions of the control decision module, and deploying a third consensus layer by introducing a physical consensus protocol to build a control decision component.

[0042] Furthermore, this application also includes the following steps: For the air pressure weight reduction device, a first control dimension and a first range of action are defined, and a support force field modulator is constructed, wherein the first control dimension includes vertical support force distribution, pressure center offset and weight reduction gradient, and the first range of action is a gravity scalar field; For the balance auxiliary arm, a second control dimension and a second range of action are defined, and a spatial torque guide is constructed, wherein the second control dimension includes six-dimensional torque, contact point position and compliance stiffness, and the second range of action is a spatial pose vector field; For the treadmill drive mechanism, a third control dimension and a third range of action are defined, and a ground dynamic guide is constructed, wherein the third control dimension includes surface velocity field, local deformation and friction characteristics, and the third range of action is a tensor field of the foot-ground interaction interface.

[0043] Specifically, the pre-control components refer to the various sub-components in the anti-gravity treadmill, such as the pneumatic weight reduction device, the balance auxiliary arm, and the treadmill drive mechanism. These components work together to ensure that all physical devices can precisely coordinate and generate appropriate forces and torques during posture correction, ensuring that the user's movement posture is effectively corrected. For each component, its control dimensions and scope of action are clearly defined by combining its mechanical design principles and biomechanical knowledge. The control dimension refers to the key physical parameters or variables that affect the working performance of the anti-gravity treadmill components, such as force, pressure, and displacement. The scope of action refers to the range or area that the control dimension can effectively influence, representing the spatial or physical range within which the component can exert its influence.

[0044] For pneumatic weight reduction devices, their regulation is abstracted as managing a gravitational scalar field. A support force field modulator is constructed to fine-tune the position of the body's center of pressure projected onto the sole of the foot (force center offset) by coordinating the pressure of different airbag zones (i.e., the vertical support force distribution), and to control the rate of change in overall body load (i.e., the weight reduction gradient). By defining the control dimensions and range of action, a support force field modulator is built to control the working state of the pneumatic weight reduction device. By adjusting the distribution of the air pressure field, the weight reduction effect is controlled, thereby providing weight reduction support for the user. The support force field refers to the force field formed by changes in air pressure, and its function is to regulate the gravitational interaction between the user and the treadmill.

[0045] For a balance assist arm, its control is abstracted as applying a spatial pose vector field. A spatial torque guide is constructed, calculating the required six-dimensional force / torque combination, i.e., forces in three directions and torques in three directions. Optimal contact point locations, such as the pelvis or torso, are selected, and contact compliance, i.e., compliant stiffness, is set to produce the desired posture guidance effect. By defining the dimensions and range of action, a spatial torque guide is generated to adjust the movement of the balance assist arm, ensuring that the arm provides precise torque. Spatial torque refers to the torque applied in three-dimensional space, determining the direction and force of the balance assist arm's support for the user's body. The purpose of the spatial torque guide is to precisely control the mechanical action of the arm to maintain the user's balance.

[0046] For the treadmill drive mechanism, its regulation is abstracted as shaping a tensor field that modulates the foot-ground interaction interface. A ground dynamics guide is constructed to plan the velocity distribution of the running belt, i.e., the surface velocity field, such as a faster initial velocity followed by a slower one to promote movement. It simulates the characteristics of different ground surfaces, such as generating local deformation through localized active deformation, and even adjusting the frictional properties of the surface material within a safe range. The ground dynamics guide is used to control the treadmill drive mechanism to ensure that the treadmill's motion is consistent with the user's gait, regulate the interaction between the treadmill and the ground, control the moving surface and frictional characteristics, and thus influence the thrust and resistance during the user's movement.

[0047] Defining the control dimension means precisely identifying all programmable parameters that drive the hardware and directly affect its output physical effects. Defining the scope of action is key to theoretically elevating physical intervention. For example, a weight-reduction device, by altering the vertical support force, essentially locally modifies the intensity of gravity experienced by the user; therefore, its scope of action is abstracted as a gravitational scalar field. A balance assist arm applies spatial forces and torques, aiming to change the position and orientation of the user's body segments; its scope of action is abstracted as a spatial pose vector field. A treadmill drive mechanism directly shapes all the mechanical conditions of the foot contact surface; its scope of action is abstracted as the most complex foot-ground interaction tensor field. The magnitude of gravity at various points in space is a scalar; a weight-reduction device, by changing the airbag pressure, essentially locally modifies the intensity of the equivalent gravitational field acting on the human body. Pose includes position and orientation (direction), both of which are vectors. A balance assist arm applies contact forces and torques to the human body through an end effector; its effect is to attempt to apply a force and torque vector at the human's spatial location to correct posture, thereby forming a virtual guiding field. Tensors can simultaneously describe properties in multiple directions. The contact area between the treadmill surface and the foot, with its dynamic characteristics such as friction coefficients in different directions, surface velocity, and local stiffness, collectively constitutes a complex mechanical impedance tensor field, directly affecting the shear force, propulsive force, and stability experienced by the foot. After defining the control dimensions and range of action of each component, the parallel training phase begins. In this phase, the working state and control dimensions of each pre-controlled component are simultaneously trained and optimized. Through parallel training, it is ensured that each component can adjust in real time according to changes in the user's posture and gait.

[0048] The first decomposition layer is an analyzer that receives the attitude error determined by the system in real time. It has built-in physical mapping rules to analyze the physical nature of the error, precisely dividing it into three independent attitude error components, each labeled and ready to be dispatched to the corresponding expert for processing. These components include, for example, vertical error, equilibrium error, and dynamic error.

[0049] The second decision layer, serving as the task distribution and parallel computing center, distributes the received error components, much like job assignment, to the support force field modulator, spatial torque guide, and ground dynamic guide. The second decision layer receives the three error components output from the first decomposition layer and assigns them to the corresponding three control decision modules. These three agents work independently and in parallel, rapidly calculating a set of preliminary control parameters for their assigned subtasks based on their respective expertise and internal models.

[0050] The deployment of the third consensus layer is for conflict resolution and global coordination. The three solutions generated by the second layer are sent here, and the physical consensus protocol algorithm is then activated. Constraints include: the user's real-time dynamic state, the physical limits of each actuator, and energy and safety constraints. Its optimization goal is to make the final composite intervention effect as close to the target as possible while minimizing internal losses. The third consensus layer iteratively adjusts the three sets of parameters. For example, it may find that the torque of the balance arm plan partially cancels out the inertial force generated by the treadmill speed adjustment, so it will fine-tune the parameters of both simultaneously until a consensus point is found. At this point, the combined force effect of the outputs of the three actuators on the user's body just meets the correction requirements without internal waste. The final set of globally optimized parameters is the first attitude correction parameter.

[0051] The control decision component is the core module responsible for integrating and executing the control strategies of various pre-control components. Based on the decomposed attitude error components, the output of the control decision module, and the physical consensus protocol, it makes the final attitude correction decision and guides each actuator to adjust the user's attitude. By defining the control dimensions and scope of action and training and optimizing the control decision module in parallel, it precisely adjusts each pre-control component to ensure that every detail in the attitude correction process is properly controlled.

[0052] Furthermore, this application also includes the following steps: determining a first corrected posture based on a first neighborhood of the target posture node according to the posture correction sequence; inputting the first corrected posture into the first decomposition layer to determine a first posture error component; wherein, the decomposition of the posture error component includes: performing a cross-mapping between the posture features and the first corrected posture, and measuring the vertical component, lateral balance component, and propulsion dynamics component based on the cross-mapping error; associating the vertical component with the air pressure weight reduction device, associating the lateral balance component with the balance auxiliary arm, and associating the propulsion dynamics component with the treadmill drive mechanism.

[0053] Specifically, the posture correction sequence contains a predetermined target posture node, representing the user's primary training objective in the near future. The first neighborhood of the target posture node is retrieved within the posture correction sequence. This first neighborhood specifically refers to the node in the sequence that is immediately adjacent to the target posture node in terms of difficulty and serves as the next logical training step. This could be a node that is one step easier or one step more difficult than the current target. If the user performs consistently and well, a more difficult neighborhood node is selected as the challenge; if the user struggles or exhibits compensatory behavior, a simpler neighborhood node is selected for regression consolidation. The ideal posture pattern defined by this selected neighborhood node is then determined as the first correction posture that needs to be tracked in real-time during the current control cycle.

[0054] The first corrected posture is input into the first decomposition layer, which decomposes the error between the current posture and the target posture. The complex posture error is broken down into several independently adjustable components: a vertical component, a lateral balance component, and a propulsive dynamics component. Simultaneously, real-time user posture features are also delivered. Within the first decomposition layer, the algorithm performs a core cross-mapping operation: precisely matching and comparing the real-time collected user joint angles, center of gravity position, and other data streams with the target values ​​at corresponding time points in the first corrected posture. By calculating the difference between the two, not only is an overall posture error value obtained, but also, based on predefined biomechanical rules, this overall error is analyzed into three physical action dimensions, thereby calculating the first posture error component. The algorithm determines the proportion of the overall error caused by improper vertical swaying of the center of gravity (the vertical component), the proportion caused by torso imbalance in the horizontal plane (the lateral balance component), and the proportion caused by the spatiotemporal dynamic mismatch of foot landing and extension (the propulsive dynamics component). The vertical, lateral balance, and propulsion dynamics components are assigned to their respective actuators: the vertical component is sent to the decision module of the pneumatic weight reduction device, the lateral balance component to the decision module of the balance auxiliary arm, and the propulsion dynamics component to the decision module of the treadmill drive mechanism. The vertical component refers to the deviation of the attitude error in the vertical direction, typically related to the direction of gravity. The lateral balance component refers to the deviation of the attitude error in the horizontal direction, typically affecting the user's balance posture. The propulsion dynamics component refers to the deviation of the attitude error in the direction of motion, typically related to the treadmill's motion control.

[0055] By decomposing posture errors into different components, precise adjustments are made for errors in each direction, avoiding the accumulation and mutual interference of errors. Precise coordination and allocation between various control components make the correction process more efficient and smoother, reducing unnecessary error corrections. Through refined and personalized posture correction, users can recover to their ideal movement posture in the shortest possible time, improving exercise performance and comfort.

[0056] Furthermore, this application also includes the following steps: transferring the first attitude error component to the control decision module in the second decision layer, performing directional input decision-making, and determining the ternary control parameters; and, according to the third consensus layer, performing conflict resolution and spatiotemporal phase coordination on the ternary control parameters to determine the first attitude correction parameters.

[0057] Specifically, the second decision layer is activated, transferring the first attitude error component to the control decision module within it. The vertical component enters the support force field modulator, the lateral component enters the spatial torque guide, and the propulsion component enters the ground dynamic guide. Each module, after being directed to its specific task, immediately activates its internal algorithm. Based on its complex internal models, such as the mapping model between airbag pressure and surface support force, the inverse dynamics model of the robotic arm, the treadmill servo control model, and the optimized strategies developed through training, it quickly calculates a set of control commands that most effectively eliminate input errors within its assigned area of ​​responsibility. These three modules work in parallel, resulting in extremely high computational efficiency. The set of control commands output by each module constitutes the initial ternary control parameters. The ternary control parameters refer to the adjustment parameters involved in the combined action of three control components during attitude correction, such as the control parameters of the air pressure weight reduction device, the balance auxiliary arm, and the treadmill drive mechanism. These parameters coordinate the work of each component to ensure the correction effect.

[0058] Through the third consensus layer, conflict resolution and spatiotemporal phase coordination are performed on the ternary control parameters. Conflict resolution identifies any combinations of parameters that would cause force vectors to cancel each other out or generate harmful coupling torques, and automatically adjusts the parameters to eliminate them. Spatiotemporal phase coordination ensures that the adjusted commands are not only numerically coordinated, but also time-stamped and precisely tied to the gait phase, and spatially aligned with the specific mechanical action point of the body. Since the three control decision modules operate independently, the ternary control parameters they propose may be contradictory or redundant in physical effects; conflict resolution detects and eliminates such physical contradictions between commands. Spatiotemporal phase coordination ensures that the intervention actions of the three actuators are precisely synchronized in time and act on the correct position and direction of the body in space, so that all intervention forces can form the desired resultant force at the same moment and the same point of action. After multiple rounds of iterative optimization, when the protocol finds a consensus where all physical and spatiotemporal constraints are satisfied and the objective function is optimal, this solution is output as the final first attitude correction parameters.

[0059] By coordinating the decision-making module and the consensus layer, the actions of each component are precisely controlled, correcting attitude errors. Optimization of the ternary control parameters and conflict resolution ensure that the correction process is efficient and stable, reducing the time required for error correction.

[0060] In summary, from the attitude correction sequence, the most critical first corrective posture to be achieved is dynamically selected and determined, and sent to the control decision component. The first decomposition layer of the control decision component compares the current user posture with the first corrective posture to calculate the overall posture error. This overall error is decomposed into three distinct posture error components, corresponding to the problems that need to be addressed by air pressure weight reduction, balance assistance, and treadmill drive, respectively. The urgency and importance of the three components are evaluated in real time for a ternary single-thread decision. Generally, stability and fall prevention-related lateral balance issues have the highest priority. One dimension is dynamically selected as the primary correction target, while the other two dimensions are adjusted to support this primary target. The initial control scheme is submitted to the third consensus layer, where a physical consensus protocol is run. While respecting the single-thread priority, the three sets of parameters undergo final conflict resolution and spatiotemporal phase coordination to ensure that the corrective torque of the balance arm, the support force adjustment of the weight reduction device, and the stability control of the treadmill are perfectly matched in terms of action time, point of action, and force magnitude, forming a composite intervention to obtain the first posture correction parameters. The first attitude correction parameters directly drive the three physical actuators to work together.

[0061] Based on the first attitude correction parameters, the anti-gravity treadmill is controlled and driven. By executing multi-stage attitude correction regulation under attitude correction sequence polling, the attitude correction control process is terminated. The attitude correction component and the control decision component are embedded plug-ins of the anti-gravity treadmill central control.

[0062] Specifically, the first attitude correction parameters output by the control decision component are sent to the control drive management module, where they are converted into multi-threaded control commands that can be directly recognized and executed by the three actuator controllers. For example, shifting the vertical support force to the left by 10% translates to increasing the opening of the left airbag solenoid valve by X milliseconds; applying a 10N force to the left translates to setting the torque of the motors at the 2nd and 3rd joints of the balance arm to Y and Z N·m; and reducing the local deceleration of the running belt by 0.05 m / s translates to correcting the speed loop setpoint of servo drive A to V rpm. Attitude correction sequence polling is a program control mechanism that continuously executes the following process at a fixed or variable frequency, such as 100 times per second: acquiring sensor data, comparing it with the current stage target, triggering the control decision component to calculate new parameters, and executing the drive. This process, like a polling check, is continuous, ensuring the real-time performance of the control.

[0063] The entire training process is not a single objective, but rather a series of consecutive stages divided according to a posture correction sequence. Each stage has a specific target node. Within a stage, after the user's posture is stabilized and reaches the target through polling control, the system automatically switches to the next target node in the sequence, starting a new stage, until the entire sequence is completed. When a preset termination condition is met, the active physical intervention loop automatically stops. Termination conditions may include: successfully completing the entire correction sequence, reaching the preset maximum training time, detecting excessive user fatigue or discomfort, and the user voluntarily stopping. In other words, the system continuously monitors the user's posture, compares it with the target of the current stage, and dynamically adjusts the driving commands through the control decision component. When it is determined that the user's posture has stabilized and met the target in the current stage, the system automatically updates the first corrected posture to the next node in the sequence, thereby advancing to the next higher or more refined correction stage.

[0064] The attitude correction component and control decision component are integrated into the central control unit of the anti-gravity treadmill as software plug-ins. They work closely together throughout the process, one responsible for macroscopic route navigation and the other for microscopic real-time driving. When the final stage is completed, or any other termination condition is triggered, such as the user's heart rate exceeding a safety threshold, the central control unit will orderly stop the control loop, smoothly reduce the output of each mechanism, and save the training data, thereby terminating the attitude correction control process.

[0065] The initial corrected posture is displayed on the anti-gravity treadmill's visual interface, typically presented as a standardized, idealized virtual human body animation or static key posture diagram, providing users with intuitive graphical information. Based on the initial posture correction parameters, the control and drive management module converts and determines multi-threaded control commands to regulate and drive the pneumatic weight reduction device, balance auxiliary arm, and treadmill drive mechanism. This involves sending commands via a specific communication protocol to the local controllers of each actuator, such as the airbag pressure servo valve, robotic arm joint servo driver, and treadmill belt servo motor driver, driving them to produce precise physical movements. Upon receiving the commands, each controller immediately drives the motors, valves, and other actuators to generate precise physical effects, thereby applying a coordinated force field intervention to the user's body.

[0066] Based on the posture correction sequence, progressive, multi-stage posture correction adjustments are executed until the posture correction is completed. The system continuously monitors the user's performance in each stage. When preset stage achievement conditions are met, such as keeping key posture indicator errors below a threshold for 30 consecutive seconds, the first corrected posture is automatically updated to the next node in the sequence, and the decision-making and driving loop restarts based on the new goal, thus advancing the training to the next stage. This process repeats continuously until the conditions for ending posture correction are met. Through precise multi-stage posture correction adjustments, every posture error is effectively corrected, ultimately achieving the ideal posture. Multi-threaded control instructions ensure the synchronous operation of each execution component, making the posture correction process more efficient and shortening the correction time. Through real-time posture display and gradual posture adjustments, users can clearly see their correction progress and receive a more accurate and comfortable correction experience.

[0067] In summary, the adaptive control method for attitude correction provided in this application has the following technical effects: By collecting motion posture data from the multi-source sensor array mounted on the anti-gravity treadmill and making correction decisions based on the posture correction component, a posture correction sequence is determined. The developmental posture tree embedded in the posture correction component serves as the decision-making basis. A first corrected posture based on the posture correction sequence is determined, triggering the control decision component to execute a ternary single-thread decision-making process based on the decomposition of posture error components and coordination with a physical consensus protocol. This determines the first posture correction parameter, where the posture error components and the ternary single-thread decision-making process are defined based on the air pressure weight reduction device, the balance auxiliary arm, and the treadmill drive mechanism. Based on the first posture correction parameter, the anti-gravity treadmill is controlled and driven. Through multi-stage posture correction regulation under posture correction sequence polling, the posture correction control process is terminated. The posture correction component and the control decision component are embedded plug-ins in the central control unit of the anti-gravity treadmill. In other words, the attitude of the target user is captured by a multi-source sensor array, the attitude correction sequence is determined by the developmental attitude tree embedded in the attitude correction component, the first correction attitude is determined, and the control decision component is triggered to perform the ternary single-thread decision and physical consensus protocol coordination under the decomposition of attitude error components to determine the first attitude correction parameters. The anti-gravity treadmill is controlled and driven, and multi-stage attitude correction regulation is performed until the attitude correction control process is terminated. The collaborative work between multiple drive mechanisms is optimized to improve the accuracy and efficiency of attitude correction.

[0068] Example 2: Based on the same inventive concept as the adaptive control method for attitude correction in Example 1, this application also provides an adaptive control system for attitude correction. Please refer to the appendix. Figure 2 The adaptive control system for attitude correction includes: The correction decision module 11 is used to collect the motion posture of the target user based on the multi-source sensor array assembled on the anti-gravity treadmill and make correction decisions based on the posture correction component to determine the posture correction sequence, wherein the developmental posture tree embedded in the posture correction component is used as the decision basis; the parameter determination module 12 is used to determine the first correction posture based on the posture correction sequence, trigger the control decision component, execute the ternary single-thread decision under the decomposition of posture error components and coordinate the physical consensus protocol to determine the first posture correction parameters, wherein the posture error components and the ternary single thread are defined based on the air pressure weight reduction device, the balance auxiliary arm and the treadmill drive mechanism; the multi-stage posture correction control module 13 is used to control and drive the anti-gravity treadmill according to the first posture correction parameters, and terminate the posture correction control process by executing multi-stage posture correction control under the posture correction sequence polling, wherein the posture correction component and the control decision component are embedded plug-ins of the anti-gravity treadmill central control.

[0069] Furthermore, the correction decision module 11 in the adaptive control system for posture correction is also used to: determine a basic standing balance pattern, wherein the basic standing balance pattern is characterized as a first support phase and a first swing phase based on the posture point cloud under the gait cycle; using the basic standing balance pattern as the root node, develop the support phase and swing phase according to a first motion complexity to determine a first leaf node layer, wherein the first leaf node layer contains at least one posture pattern that satisfies the first motion complexity, and each posture pattern is identified by a support phase and a swing phase based on the posture point cloud under the gait cycle; using the first leaf node layer, develop the support phase and swing phase according to a second motion complexity, and generate the developmental posture tree through hierarchical development; and construct a posture correction component based on the developmental posture tree.

[0070] Furthermore, the correction decision module 11 in the adaptive control system for posture correction is also used to: collect the motion posture signal of the target user based on the multi-source sensor array assembled on the anti-gravity treadmill; the data platform receives the motion posture signal, performs preprocessing and feature extraction, and determines the posture features, wherein the posture features include real-time posture point cloud, gait phase, symmetry and stability indices; and triggers the posture correction component to perform similarity matching and pruning of the posture features in the developmental posture tree to determine the posture correction sequence.

[0071] Furthermore, the correction decision module 11 in the adaptive control system for attitude correction is also used to: determine a target attitude node by performing similarity matching on the attitude features in the developmental attitude tree; and perform hierarchical forward and backward propagation-based pruning on the developmental attitude tree using the target attitude node as the attitude correction sequence; wherein, in the attitude correction sequence, the sequence node weights are reduced in order with the target attitude node as the center.

[0072] Furthermore, the parameter determination module 12 in the adaptive control system for attitude correction is also used for: defining the adjustment dimension and range of each pre-control component for the anti-gravity treadmill, building an adjustment decision module through parallel training, wherein the adjustment decision module includes a support force field modulator, a spatial torque guide, and a ground dynamic guide corresponding to the pre-control component; deploying a first decomposition layer based on attitude error component decomposition, deploying a second decision layer based on independent component decision of the adjustment decision module, and deploying a third consensus layer by introducing a physical consensus protocol to build a control decision component.

[0073] Furthermore, the parameter determination module 12 in the adaptive control system for attitude correction is also used for: defining a first control dimension and a first range of action for the air pressure weight reduction device, and building a support force field modulator, wherein the first control dimension includes vertical support force distribution, pressure center offset and weight reduction gradient, and the first range of action is a gravity scalar field; defining a second control dimension and a second range of action for the balance auxiliary arm, and building a spatial torque guide, wherein the second control dimension includes six-dimensional torque, contact point position and compliance stiffness, and the second range of action is a spatial pose vector field; defining a third control dimension and a third range of action for the treadmill drive mechanism, and building a ground dynamic guide, wherein the third control dimension includes surface velocity field, local deformation and friction characteristics, and the third range of action is a tensor field of the foot-ground interaction interface.

[0074] Furthermore, the parameter determination module 12 in the adaptive control system for attitude correction is also used to: determine a first corrected attitude based on the attitude correction sequence and a first neighborhood of the target attitude node; input the first corrected attitude into the first decomposition layer to determine a first attitude error component; wherein the decomposition of the attitude error component includes: cross-mapping the attitude features with the first corrected attitude, measuring the vertical component, lateral balance component, and propulsion dynamics component based on the cross-mapping error; associating the vertical component with the air pressure weight reduction device, associating the lateral balance component with the balance auxiliary arm, and associating the propulsion dynamics component with the treadmill drive mechanism.

[0075] Furthermore, the parameter determination module 12 in the adaptive control system for attitude correction is also used to: transfer the first attitude error component to the control decision module in the second decision layer, perform directional input decision, and determine the three-element control parameters; and, according to the third consensus layer, perform conflict resolution and spatiotemporal phase coordination on the three-element control parameters to determine the first attitude correction parameters.

[0076] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The adaptive control method and specific examples for attitude correction in the foregoing Embodiment 1 are also applicable to the adaptive control system for attitude correction in this embodiment. Through the foregoing detailed description of the adaptive control method for attitude correction, those skilled in the art can clearly understand the adaptive control system for attitude correction in this embodiment. Therefore, for the sake of brevity, it will not be described in detail here.

[0077] In embodiment three, based on the same inventive concept as the adaptive control method for attitude correction in embodiment one, this application also provides a computer-readable storage medium storing a computer program that, when executed, implements the steps of the adaptive control method for attitude correction described in any one of embodiments one above.

[0078] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

[0079] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of this application and its equivalents, this application also intends to include such modifications and variations.

Claims

1. An adaptive control method for attitude correction, characterized in that, include: Based on the multi-source sensor array assembled on the anti-gravity treadmill, the motion posture of the target user is collected and the correction decision is made based on the posture correction component to determine the posture correction sequence, wherein the developmental posture tree embedded in the posture correction component is used as the decision basis. The first corrected posture based on the posture correction sequence is determined, the control decision component is triggered, the ternary single-thread decision and physical consensus protocol coordination under the decomposition of the posture error components are executed, and the first posture correction parameters are determined. The posture error components and the ternary single thread are defined based on the air pressure weight reduction device, the balance auxiliary arm and the treadmill drive mechanism. Based on the first attitude correction parameters, the anti-gravity treadmill is controlled and driven. By executing multi-stage attitude correction regulation under attitude correction sequence polling, the attitude correction control process is terminated. The attitude correction component and the control decision component are embedded plug-ins of the anti-gravity treadmill central control.

2. The adaptive control method for attitude correction as described in claim 1, characterized in that, The construction of the posture correction component includes: Determine the basic standing balance pattern, wherein the basic standing balance pattern is characterized as the first support phase and the first swing phase based on the posture point cloud under the gait cycle; Using the basic standing balance pattern as the root node, the support phase and swing phase are developed according to the first motion complexity to determine the first leaf node layer. The first leaf node layer contains at least one posture pattern that satisfies the first motion complexity. Each posture pattern is identified by the support phase and swing phase based on the posture point cloud under the gait cycle. Using the first leaf node layer, the support phase and the oscillation phase are developed according to the second motion complexity. Through hierarchical development, the developmental attitude tree is generated. Based on the developmental attitude tree, an attitude correction component is constructed.

3. The adaptive control method for attitude correction as described in claim 2, characterized in that, The system collects motion posture data from the target user and makes correction decisions based on the posture correction component to determine the posture correction sequence, including: The motion posture signals of the target user are collected using the multi-source sensor array assembled on the anti-gravity treadmill. The data platform receives the motion posture signal, performs preprocessing and feature extraction, and determines the posture features, wherein the posture features include real-time posture point cloud, gait phase, symmetry and stability indicators. The posture correction component is triggered to perform similarity matching and pruning on the posture features in the developmental posture tree to determine the posture correction sequence.

4. The adaptive control method for attitude correction as described in claim 3, characterized in that, In the developmental pose tree, similarity matching and pruning of the pose features are performed, including: The target pose node is determined by performing similarity matching on the pose features in the developmental pose tree; Using the target pose node, perform hierarchical forward and backward propagation-based pruning in the developmental pose tree as the pose correction sequence. In the attitude correction sequence, the weights of the sequence nodes are reduced in order, with the target attitude node as the center.

5. The adaptive control method for attitude correction as described in claim 3, characterized in that, Before triggering the control decision component, the construction of the control decision component includes: For the pre-control components of the anti-gravity treadmill, the adjustment dimensions and range of action of each pre-control component are defined, and an adjustment decision module is built through parallel training. The adjustment decision module includes a support force field modulator, a spatial torque guide, and a ground dynamic guide corresponding to the pre-control components. The first decomposition layer is deployed by decomposing attitude error components, the second decision layer is deployed by making independent component decisions based on the control decision module, and the third consensus layer is deployed by introducing a physical consensus protocol to build a control decision component.

6. The adaptive control method for attitude correction as described in claim 5, characterized in that, The regulation dimensions and scope of each pre-control component are defined, and a regulation decision module is built through parallel training, including: For the air pressure weight reduction device, a first control dimension and a first effective range are defined, and a support force field modulator is built. The first control dimension includes the vertical support force distribution, pressure center offset and weight reduction gradient, and the first effective range is the gravity scalar field. For the balancing auxiliary arm, a second control dimension and a second range of action are defined, and a spatial torque guide is built. The second control dimension includes six-dimensional torque, contact point position and compliance stiffness, and the second range of action is the spatial pose vector field. For the treadmill drive mechanism, a third control dimension and a third range of action are defined, and a ground dynamic guide is built. The third control dimension includes the surface velocity field, local deformation, and friction characteristics, and the third range of action is the tensor field of the foot-ground interaction interface.

7. The adaptive control method for attitude correction as described in claim 6, characterized in that, Decomposition of attitude error components includes: Based on the attitude correction sequence, a first corrected attitude is determined using a first neighborhood of the target attitude node; The first corrected posture is input into the first decomposition layer to determine the first posture error component; The decomposition of the attitude error components includes: The attitude features are cross-mapped with the first corrected attitude, and the vertical component, lateral balance component, and propulsion dynamics component are measured based on the cross-mapped error. The vertical component is associated with the air pressure weight reduction device, the lateral balance component is associated with the balance auxiliary arm, and the propulsion dynamics component is associated with the treadmill drive mechanism.

8. The adaptive control method for attitude correction as described in claim 7, characterized in that, The first attitude error component is transferred to the control decision module in the second decision layer to perform directional input decision and determine the three-dimensional control parameters. Based on the third consensus layer, conflict resolution and spatiotemporal phase coordination are performed on the ternary control parameters to determine the first attitude correction parameters.

9. An adaptive control system for attitude correction, characterized in that, The step of implementing the adaptive control method for attitude correction according to any one of claims 1 to 8, wherein the adaptive control system for attitude correction comprises: The correction decision module is used to collect the motion posture of the target user based on the multi-source sensor array assembled on the anti-gravity treadmill and make correction decisions based on the posture correction component to determine the posture correction sequence, wherein the developmental posture tree embedded in the posture correction component is used as the decision basis. The parameter determination module is used to determine the first corrected posture based on the posture correction sequence, trigger the control decision component, execute the ternary single-thread decision and physical consensus protocol coordination under the decomposition of posture error components, and determine the first posture correction parameters. The posture error components and the ternary single thread are defined based on the air pressure weight reduction device, the balance auxiliary arm and the treadmill drive mechanism. The multi-stage attitude correction and control module is used to control and drive the anti-gravity treadmill according to the first attitude correction parameters. By executing multi-stage attitude correction and control under attitude correction sequence polling, the attitude correction control process is terminated. The attitude correction component and the control decision component are embedded plug-ins of the anti-gravity treadmill central control.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed, implements the steps of the adaptive control method for attitude correction as described in any one of claims 1 to 8.

Images