Gait prediction compensation admittance control method and system for lower limb rehabilitation exoskeleton

Abstract

The invention relates to the technical field of rehabilitation robot control, and particularly discloses a gait prediction compensation admittance control method and system for lower limb rehabilitation exoskeleton, and the method comprises the steps: inputting a multi-dimensional motion vector into a time sequence prediction model, and outputting a prediction result of a gait phase at a future moment; if the maximum probability output by the time sequence prediction model is lower than a preset threshold value, automatically switching to a pre-constructed rule criterion to carry out gait phase recognition; an admittance control parameter matched with the gait phase prediction result is called from an admittance parameter mapping table, and a main control torque is calculated based on the called admittance control parameter; calculating a compensation torque based on the deviation between the actual human-computer interaction force and the expected interaction force; and the main control torque and the compensation torque are superposed to serve as a final torque driving instruction of the exoskeleton motor. Real-time gait prediction, admittance parameter self-adaptive adjustment and interaction force error dynamic compensation can be fused, and high-flexibility and high-safety man-machine collaborative rehabilitation training is achieved.

Description

Technical Field

[0001] This invention relates to the field of rehabilitation robot control technology, and more specifically to a gait prediction and compensation admittance control method and system for a lower limb rehabilitation exoskeleton. Background Technology

[0002] Currently, lower limb rehabilitation exoskeleton robots widely employ position control strategies based on preset trajectories, which force patients to follow fixed reference gait trajectories during training. While this method ensures proper movement, it ignores the active movement intentions that patients may develop in the later stages of rehabilitation, leading to human-machine aggression and reducing training initiative and rehabilitation effectiveness. Especially for patients with some residual motor abilities, forcing them to follow fixed trajectories not only limits their active participation but may also inhibit the recovery of neuroplasticity, contradicting the modern rehabilitation concept of "patient-centered care."

[0003] To improve the compliance of human-machine interaction, some studies have introduced admittance control strategies, mapping human-machine interaction forces to positional or torque adjustments. Admittance control establishes a dynamic relationship between interaction forces and motor responses, enabling the exoskeleton to exhibit spring-damped system-like compliance characteristics, thus allowing patients to autonomously guide movement within a certain range. However, existing admittance control typically uses fixed stiffness and damping parameters, failing to consider the significant differences in assistive force requirements across different gait phases. For example, lower stiffness is needed in the swing phase to allow free forward limb swing, while higher stiffness is required in the stance phase to provide antigravity support and stability. Fixed parameters can easily lead to insufficient or excessive assistance, affecting gait naturalness and safety.

[0004] Furthermore, although existing studies have utilized plantar pressure sensors, inertial measurement units (IMUs), or surface electromyography (sEMG) signals for gait phase recognition, most methods rely on data from the current moment or a short time window for state assessment. While these methods perform reasonably well under uniform, regular gait conditions, phase recognition often suffers from delays when patients experience changes in gait speed, uneven stride length, or abnormal gait. This delay causes control commands to lag behind actual movement demands, resulting in the exoskeleton applying incorrect assistive forces at the wrong time. This not only reduces training effectiveness but may also trigger muscle compensation or increase the risk of falls.

[0005] Existing solutions generally lack compensation mechanisms for prediction uncertainties or transient errors in phase switching. When the prediction results deviate from the actual gait, the system cannot correct the auxiliary force output in time, resulting in sudden changes in human-machine interaction force, which affects comfort and safety.

[0006] Therefore, improving the current lower limb rehabilitation exoskeleton's ability to anticipate intent perception, adapt control parameters, and robustly compensate for errors has become a pressing technical problem for those skilled in the art. Thus, a solution is urgently needed. Summary of the Invention

[0007] In view of the above problems, this invention proposes a gait prediction compensation admittance control method and system for lower limb rehabilitation exoskeleton, which can integrate real-time gait prediction, phase-dependent online adjustment of admittance parameters and dynamic compensation of interaction force error to achieve highly compliant, highly safe and highly personalized human-machine collaborative rehabilitation training.

[0008] To achieve the above objectives, the present invention adopts the following technical solution: In a first aspect, the present invention provides a gait prediction and compensation admittance control method for a lower limb rehabilitation exoskeleton, comprising: Multi-source motion information is collected in real time, and after filtering and time alignment processing, a unified multi-dimensional motion vector is obtained; The multidimensional motion vector is input into a pre-trained temporal prediction model, which outputs the prediction result of gait phase at future time. If the maximum probability output by the temporal prediction model is lower than a preset threshold, it automatically switches to a pre-built rule criterion for gait phase recognition. The admittance control parameters that match the gait phase prediction results are retrieved from the pre-built admittance parameter mapping table, and the master control torque is calculated based on the retrieved admittance control parameters. Based on the deviation between the actual human-computer interaction force and the expected interaction force, the compensation torque is calculated. The main control torque and the compensation torque are superimposed to form the final torque drive command for the exoskeleton motor.

[0009] Furthermore, the multidimensional motion vector includes at least: the real-time angles and angular velocities of the joints, the pressure distribution of the forefoot and heel of the left and right feet, and the triaxial acceleration and angular velocity of the thigh and calf.

[0010] Furthermore, the process of gait phase recognition based on rule-based criteria includes: A four-phase finite state machine is constructed based on plantar pressure and joint velocity, and a ground contact threshold is set. T on and ground threshold T 0ff The states of the finite state machine are defined as including the initial support state, the intermediate support state, the final support state, and the swing state. Plantar pressure data and joint angular velocity data were used as input data, with plantar pressure data including heel pressure. P h and forefoot pressure P f Joint angular velocity data should include at least the ankle joint angular velocity. ; The finite state machine applies a state transition rule based on the current state and input data. The state transition rule is as follows: If total plantar pressure And ankle joint angular velocity When a value changes from negative to positive, it transitions from an initial support state to a medium-term support state. If heel pressure P h The decrease exceeded the preset value, and the forefoot pressure Ankle joint angular velocity When the value is greater than 0, the support level shifts from a medium-term support level to a late-term support level. If total plantar pressure And ankle joint angular velocity When the trend is upward, it shifts from a late-stage support state to a swing state. If total plantar pressure At that time, it transitions from the swing state to the initial support state.

[0011] Furthermore, the admittance control parameters include stiffness and damping. Different phase states in the admittance parameter mapping table correspond to different stiffnesses and dampings. The calculation method for the master control torque is as follows:

[0012] in, Indicates the controlling torque. Indicates gait phase prediction results The corresponding stiffness, Indicates gait phase prediction results The corresponding damping, Indicates joint angle, Indicates joint angular velocity, and This represents the reference trajectory and its derivative generated based on a normal gait database.

[0013] Furthermore, the calculation process for the compensating torque includes: Calculate the expected interaction force:

[0014] in, Indicates gait phase prediction results Corresponding stiffness; Indicates joint angle, This represents a reference trajectory generated based on a normal gait database; The actual human-computer interaction force is obtained by reverse calculation of electrode current or measurement by force sensor; Calculate the compensation torque based on the deviation between the actual human-computer interaction force and the expected interaction force:

[0015] in, Indicates the compensating torque. G c Indicates adaptive compensation gain, LPF represents low-pass filtering, ΔF = F int F ref This represents the deviation between the actual human-computer interaction force and the expected interaction force.

[0016] Furthermore, it also includes: The final torque drive command is subject to dual safety limits. The first limit is to restrict the amplitude of the final torque drive command to no more than 80% of the motor's rated value. The second limit is to ensure that the joint angle is always within the physiological safety range. If the amplitude of the final torque drive command or the joint angle exceeds the limit, the system will switch to the high stiffness impedance safety mode within the first preset time period. If the dual safety limit conditions are not met within the second preset time period in the high stiffness impedance safety mode, the system will automatically stop.

[0017] Furthermore, the formula for calculating the control torque under high stiffness impedance safety mode is as follows:

[0018] in, This represents the control torque under high-stiffness impedance safety mode; This represents the stiffness in the high-stiffness impedance safety mode, and its value is the original stiffness. 2 to 4 times; This indicates damping in the high-stiffness impedance safety mode; This indicates the joint safety holding angle in the high stiffness impedance safety mode.

[0019] Secondly, the present invention provides a gait prediction and compensation admittance control system for a lower limb rehabilitation exoskeleton, which employs the method described above, including: The data acquisition module is used to collect multi-source motion information in real time, and after filtering and time alignment processing, a unified multi-dimensional motion vector is obtained. The gait phase prediction module is used to input multi-dimensional motion vectors into a pre-trained temporal prediction model and output the prediction results of gait phase at future moments. If the maximum probability output by the temporal prediction model is lower than a preset threshold, it automatically switches to a pre-built rule criterion for gait phase recognition. The adaptive admittance control module is used to retrieve admittance control parameters that match the gait phase prediction results from a pre-built admittance parameter mapping table, and calculate the master control torque based on the retrieved admittance control parameters. The error compensation module is used to calculate the compensation torque based on the deviation between the actual human-computer interaction force and the expected interaction force. The drive execution module is used to superimpose the main control torque and the compensation torque as the final torque drive command for the exoskeleton motor.

[0020] Thirdly, the present invention provides a computer-readable storage medium having a computer program stored thereon, characterized in that, when a processor executes the computer program, it implements the steps of the gait prediction compensation admittance control method for a lower limb rehabilitation exoskeleton as described above.

[0021] Fourthly, the present invention provides a computer device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor; when the processor executes the computer program, it implements the steps of the gait prediction compensation admittance control method for a lower limb rehabilitation exoskeleton as described above.

[0022] As can be seen from the above technical solution, this invention effectively overcomes the core defects of existing technologies, such as control lag, parameter rigidity, and poor robustness, by integrating three mechanisms: gait phase forward prediction, admittance parameter phase adaptive adjustment, and real-time compensation for interaction force errors. This significantly improves the compliance, adaptability, and safety of human-machine collaboration. Compared with traditional fixed-parameter admittance control or reactive control methods based on the current state, this invention has the following outstanding technical effects: (1) Significantly improves control foresight and intention tracking ability: This invention predicts the gait phase at the next moment through a lightweight temporal prediction model and pre-adjusts the control parameters accordingly, so that the exoskeleton can complete the adjustment of assistive force characteristics before the phase switch occurs. It can effectively reduce the control delay from 200-500 ms in the traditional method to less than 50 ms, greatly reducing human-machine aggression caused by response lag, and making the assistive movements more in line with the patient's true movement intention.

[0023] (2) Achieve dynamic adaptive matching of auxiliary force within the gait cycle: This invention establishes a mapping relationship between gait phase and admittance control parameters (stiffness K, damping B), automatically increases stiffness in the support phase to enhance stability, and actively decreases stiffness in the swing phase to reduce motion resistance. While reducing the peak value of knee joint interaction force, it also improves the gait symmetry index and the patient's subjective comfort.

[0024] (3) Effectively suppress force mutation caused by prediction error and transient disturbance: The present invention introduces a compensation torque based on the interaction force deviation ΔF to form a prediction-execution-correction closed loop. Even in the case of sudden change in gait or abnormal gait, the auxiliary force output can be smoothly corrected within 1–2 control cycles, significantly reducing muscle compensation or discomfort caused by force mutation.

[0025] (4) Ensuring high safety and clinical applicability: This invention integrates torque limiting, joint angle protection and abnormal shutdown mechanism to form a multi-level safety protection, avoid falls or human-machine conflict caused by loss of control, and fully meet the stringent safety requirements of rehabilitation training. Attached Figure Description

[0026] To more clearly illustrate the technical solutions in the embodiments of the present invention 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 only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0027] Figure 1 This is a flowchart of the gait prediction compensation admittance control method for a lower limb rehabilitation exoskeleton provided in an embodiment of the present invention; Figure 2 This is a structural block diagram of the gait prediction and compensation admittance control system for a lower limb rehabilitation exoskeleton provided in an embodiment of the present invention. Detailed Implementation

[0028] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0029] like Figure 1 As shown, this embodiment of the invention discloses a gait prediction and compensation admittance control method for a lower limb rehabilitation exoskeleton, comprising the following steps: S1. Real-time acquisition of multi-source motion information, followed by filtering and time alignment processing, yields a unified multi-dimensional motion vector; S2. Input the multidimensional motion vector into the pre-trained temporal prediction model and output the prediction result of the gait phase at future time. If the maximum probability output by the temporal prediction model is lower than the preset threshold, automatically switch to the pre-built rule criteria for gait phase recognition. S3. Retrieve the admittance control parameters that match the gait phase prediction results from the pre-built admittance parameter mapping table, and calculate the master control torque based on the retrieved admittance control parameters. S4. Calculate the compensation torque based on the deviation between the actual human-computer interaction force and the expected interaction force; S5. The main control torque and the compensation torque are superimposed to form the final torque drive command for the exoskeleton motor.

[0030] The following is a further explanation of each of the above steps.

[0031] S1. By deploying multiple sensors on the joints, soles, and limb segments of the lower limb exoskeleton, user movement status information is acquired synchronously, providing highly reliable input for subsequent gait intention recognition. Specifically, this includes: Angle encoders are installed in the hip, knee, and ankle joints to collect real-time joint angles. and angular velocity; An array of pressure sensors is embedded in the insole to acquire the pressure distribution on the forefoot and heel of the left and right feet, and to calculate the pressure ratio. As a key criterion for support / oscillation status; among them; t The sampling time (in seconds) is collected at the same frequency as the foot / joint sensor; Instantaneous pressure on the forefoot (metatarsophalangeal / forefoot area); Instantaneous pressure in the heel area; To prevent the denominator from being zero during the empty period; Forefoot bearing ratio, dimensionless, theoretical value. .

[0032] Inertial measurement units (IMUs) were strapped to the thighs and calves to acquire triaxial acceleration and angular velocity data. All signals were synchronously sampled at a frequency of at least 100 Hz and then low-pass filtered and time-aligned to form a unified multidimensional feature vector x(t) for subsequent gait prediction.

[0033] S2. Gait Phase Prediction: The core of this step is to anticipate the user's next gait intention, rather than passively responding to the current state. Specifically, it includes: The multidimensional feature vector x(t) within a sliding time window (e.g., 200 ms) is input into a lightweight temporal prediction model (e.g., a single-layer LSTM network), and the output is the prediction result of the gait phase at a future time (e.g., 100 ms later). This phase can be represented as a discrete category (such as "initial support phase", "oscillation phase", etc.) or a normalized continuous variable. .

[0034] The time of a gait cycle is normalized using plantar contact events, specifically by recording the timestamps of two consecutive heel strikes on the same side. , For any time definition

[0035] When the foot sensors are unstable, the lower leg IMU (tibia angular velocity peak) or ankle angular velocity zero crossover can be used as a backup time scale, and smoothed with a first-order phase-locked loop (PLL) or Kalman filter.

[0036] To improve robustness, a prediction confidence threshold is set for the time-series prediction model. When the maximum probability output by the model falls below 0.8, the model automatically switches to phase recognition based on rule-based criteria of plantar pressure and joint velocity, avoiding misjudgments caused by abnormal gait. This mechanism significantly overcomes the control delay problem caused by the lag in phase recognition in traditional methods.

[0037] Specifically, the process of gait phase recognition based on rule-based criteria includes: 1) Construct a four-phase finite state machine based on plantar pressure and joint velocity, and set a ground contact threshold. T on and ground threshold T 0ff The finite state machine is defined to include an initial support state, a mid-term support state, a final support state, and a swing state; in this embodiment, the bottoming threshold is... Ground threshold .

[0038] 2) Use plantar pressure data and joint angular velocity data as input data, where plantar pressure data includes heel pressure. P h and forefoot pressure P f Joint angular velocity data should include at least the ankle joint angular velocity. In practical implementation, the angular velocity of the knee joint can also be collected simultaneously. It is used for subsequent gait analysis or as redundant monitoring information, but the state transition rules of the finite state machine take the ankle joint angular velocity as the main criterion.

[0039] 3) The finite state machine applies state transition rules based on the current state and input data. The state transition rules are as follows: If total plantar pressure And ankle joint angular velocity When the value changes from negative to positive (i.e., a zero cross occurs from the direction of back extension to the direction of expansion), it transitions from the initial support state to the intermediate support state. If heel pressure P h The decrease exceeded the preset value, and the forefoot pressure Ankle joint angular velocity When the value is greater than 0, the support level shifts from a medium-term support level to a late-term support level. If total plantar pressure And ankle joint angular velocity When the trend is upward, it shifts from a late-stage support state to a swing state. If total plantar pressure At that time, it transitions from the swing state to the initial support state.

[0040] A minimum dwell time is set for each state, ranging from 50 to 100 milliseconds. Within this minimum dwell time, the state does not respond to changes in transition conditions to prevent misjudgments caused by signal fluctuations. Simultaneously, a first-order IIR filter is used... P h and P f Perform a 10–15 Hz low-pass filter to effectively suppress impulse noise.

[0041] When the confidence of the time series prediction model is insufficient, this finite state machine (FSM) can be used to directly provide the phase; at the same time, the gait phase is reset / corrected, and the corrected gait phase is linearly interpolated to the corresponding phase segment.

[0042] S3: Calculation of master control torque. This step is based on the predicted future gait phase. It dynamically retrieves the matching admittance control parameters to achieve on-demand adjustment of the auxiliary force characteristics.

[0043] Admittance control parameters include stiffness and damping. By pre-establishing an admittance parameter mapping table, different phase states are mapped to different stiffnesses. With damping For example, the support phase uses high stiffness to provide anti-gravity support, while the swing phase uses low stiffness to reduce motion drag. The parameter mapping table is shown in Table 1. Only a portion of the parameters is shown in the table: Table 1 Parameter Mapping Table

[0044] The admittance control law (i.e., the master control torque) is calculated as follows:

[0045] in, Indicates the controlling torque. Indicates gait phase prediction results The corresponding stiffness, Indicates gait phase prediction results t +1 corresponds to damping, , This term refers to the angles and angular velocities of a single driven joint, which in practice correspond to the sagittal flexion-extension angles of the hip and knee joints, respectively. and its angular velocity; this admittance control law is calculated independently for each controlled joint. These are then combined to form a joint torque vector, which serves as part of the motor drive command.

[0046] and This represents a reference trajectory and its derivative generated based on a normal gait database, which can be scaled and adjusted according to the patient's rehabilitation stage. The normal gait database consists of hip, knee, and ankle joint angle sequences collected from the unaffected side of healthy subjects or patients under natural walking conditions. These sequences are divided into single-step cycles after heel strike events and normalized to a 0–100% time scale. The average of multiple cycles at the same phase point is then calculated to form standard angle curves for each joint as a function of gait phase. In practical applications, the predicted gait phase is used... The value is obtained by interpolation through table lookup. And scale the amplitude according to the patient's recovery stage. Its time derivative.

[0047] The above formula allows the control parameters to change adaptively with the predicted phase rather than being fixed, thereby significantly improving the naturalness of human-machine collaboration.

[0048] To further illustrate the mapping relationship between admittance control parameters and gait phase, two routes are presented: offline and online. 1) Offline data-driven fitting: Collect a segment of auxiliary training data and minimize...

[0049] Boundary constraints , Below, smoothing is obtained using spline / piecewise constants. , This represents the actual human-computer interaction force calculated at time t using a torque sensor or motor current during offline assisted training. Indicates gait phase A pre-defined template for the expected interaction force of the independent variable, such as a "comfortable interaction force-phase" curve constructed based on healthy individuals or experience thresholds; This represents the regularization weight coefficient, used to balance the requirements of "fitting the desired interaction force" and "keeping parameter changes smooth and the magnitude not too large". and This represents the adjustment vector of the admittance parameter relative to the initial setting value.

[0050] This objective function is used to construct gait phase → admittance parameters offline. , The initial mapping table. The specific implementation method is: under the supervision of a therapist, a segment of auxiliary training data is collected, and the actual human-computer interaction force at each sampling moment is mapped... Interaction template with expectations The goal is to minimize this difference across all sampling times, while also applying a regularization term. Limiting the variation range of stiffness and damping, so that the obtained It can approximate the desired interaction force without abruptly changing phases. Solving under boundary constraints yields a smooth admittance parameter curve that can be used for initialization.

[0051] 2) Online small-step adaptive: Projected gradient updates are performed based on the correlation between interaction force deviation and pose error, and amplitude is limited.

[0052] in, For a small learning rate, sat() is the interval projection; , These represent the angular error of the current joint angle relative to the reference trajectory and the error of the current joint angular velocity relative to the reference velocity, respectively. Clinically, "high support / low swing" is commonly used as the initialization table, and then progressively individualized.

[0053] The above update formula is essentially a gradient descent process with boundary projection: when the interaction force deviation is at a certain phase With pose error When the correlation is high, through items , Correspondingly, the stiffness and damping are finely adjusted so that the same type of error gradually decreases in subsequent training; while the small learning rate and interval projection ensure that the parameters change slowly and stably, without causing abrupt changes in control.

[0054] Online small-step adaptive adjustment is based on offline parameter tables and takes into account the patient's real-time performance, making slow adjustments. , The process of gradually adjusting parameters to match individual abilities can be divided into the following steps: 1) At each sampling time The gait phase at the current and next time steps is obtained based on time series prediction or FSM. ; 2) Read the stiffness and damping of the corresponding phase from the current parameter table. ; 3) Obtain the reference joint angle and angular velocity for this phase by looking up a table in the normal gait database. And compared with the actual joint angles and angular velocities Calculate pose error:

[0055] 4) The expected interaction force obtained from the previous step and the measured actual interaction force Calculate the interaction force deviation ; 5) and Substituting into the update formula, we make a small step adjustment to the stiffness and damping at the current phase:

[0056] in, This is an interval projection operator used to restrict the updated parameters to a certain range. or Within the specified range, avoid admittance parameters that are too large or have abnormal signs.

[0057] 6) After multiple training steps, the values ​​under each phase , It will gradually converge to a personalized value suitable for the current patient.

[0058] S4. Calculate the interaction force error and generate a compensation torque. Although phase prediction improves control foresight, prediction bias or model mismatch may still exist, leading to actual human-machine interaction force errors. F int Compared with expected value F ref A deviation occurred. Therefore, this step introduces an error compensation mechanism to correct the interaction force deviation in real time. Specifically, this includes: Calculate the expected interaction force F ref :

[0059] in, Indicates gait phase prediction results Corresponding stiffness; Indicates joint angle, This represents a reference trajectory generated based on a normal gait database; The actual human-computer interaction force is obtained by reverse electrode current calculation or force sensor measurement. F int ; Calculate the compensation torque based on the deviation between the actual human-computer interaction force and the expected interaction force:

[0060] in, Indicates the compensating torque; G c This indicates adaptive compensation gain, which can be dynamically adjusted according to the patient's muscle strength level (the weaker the muscle strength, the better). G c The larger the value, the more this compensation term is added to the main control torque, effectively suppressing sudden force changes caused by prediction errors and enhancing system robustness and comfort; LPF represents low-pass filtering, ΔF= F int Fref This represents the deviation between the actual human-computer interaction force and the expected interaction force.

[0061] S5. Synthesize and execute control commands: Indicate the main control torque... With compensation torque The superposition of these forces forms the final torque drive command for the exoskeleton motor:

[0062] More advantageously, embodiments of the present invention further include: applying dual safety limits to the final torque drive command; the first limit is: restricting the amplitude of the final torque drive command to no more than 80% of the motor's rated value; the second limit is: ensuring that the joint angle is always within the physiological safety range, for example, the knee joint angle must meet the following requirements. ; If the amplitude of the final torque drive command or the joint angle exceeds the limit, the system will switch to the high stiffness impedance safety mode within the first preset time (100ms). If the dual safety limit conditions are not met within the second preset time (5s) in the high stiffness impedance safety mode, the system will automatically stop.

[0063] The formula for calculating the control torque under high stiffness impedance safety mode is:

[0064] in, This represents the control torque under high-stiffness impedance safety mode; This represents the stiffness in the high-stiffness impedance safety mode, and its value is the original stiffness. 2 to 4 times; This indicates the damping in the high-stiffness impedance safety mode, slightly higher than... To suppress oscillations, its value is taken as the original stiffness. 1.2 to 2 times; This indicates the joint safety holding angle in the high stiffness impedance safety mode.

[0065] In addition, speed / acceleration limiting and power limiting can be enabled, along with continuous health monitoring; if recovery is not achieved within 5 seconds, an alarm will sound to shut down. This dual limiting mechanism ensures the high reliability and clinical applicability of this invention in real-world rehabilitation scenarios.

[0066] In one embodiment, such as Figure 2 As shown, the present invention also provides a gait prediction and compensation admittance control system for a lower limb rehabilitation exoskeleton, which employs the above-described method, including: The data acquisition module is used to collect multi-source motion information in real time, and after filtering and time alignment processing, a unified multi-dimensional motion vector is obtained. The gait phase prediction module is used to input multi-dimensional motion vectors into a pre-trained temporal prediction model and output the prediction results of gait phase at future moments. If the maximum probability output by the temporal prediction model is lower than a preset threshold, it automatically switches to a pre-built rule criterion for gait phase recognition. The adaptive admittance control module is used to retrieve admittance control parameters that match the gait phase prediction results from a pre-built admittance parameter mapping table, and calculate the master control torque based on the retrieved admittance control parameters. The error compensation module is used to calculate the compensation torque based on the deviation between the actual human-computer interaction force and the expected interaction force. The drive execution module is used to superimpose the main control torque and the compensation torque as the final torque drive command for the exoskeleton motor.

[0067] Based on the same inventive concept, embodiments of the present invention also provide a computer-readable storage medium having a computer program stored thereon, characterized in that, when a processor executes the computer program, it implements the steps of the gait prediction compensation admittance control method for a lower limb rehabilitation exoskeleton as described above.

[0068] Based on the same inventive concept, embodiments of the present invention also provide a computer device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor; characterized in that, when the processor executes the computer program, it implements the steps of the gait prediction compensation admittance control method for lower limb rehabilitation exoskeleton as described above.

[0069] Next, two examples will be used to further illustrate the invention.

[0070] Example 1: Taking a unilateral lower limb rehabilitation exoskeleton system as an example, it is used to assist stroke hemiplegic patients in ground walking training. The system mainly includes high-precision angle encoders installed in the hip and knee joints, four-point plantar pressure sensors embedded in the insoles, an inertial measurement unit (IMU) strapped to the thigh, a main control computing unit, and a torque motor to drive joint movement.

[0071] When the system is working, it first collects signals such as the patient's joint angles, plantar pressure distribution, and limb acceleration in real time using sensors at a sampling frequency of 100 Hz. This data is then fed into a lightweight temporal prediction model (based on an LSTM structure with fewer than 10,000 parameters) to predict the gait phase the patient will enter 100 milliseconds in advance.

[0072] Based on the prediction results, the system automatically retrieves the corresponding stiffness and damping values ​​from the preset parameter table: higher stiffness is used during the support phase to provide stable support, while lower stiffness is switched during the oscillation phase to reduce motion resistance. Simultaneously, the system estimates the interaction force between the human and machine in real time using the motor current and compares it with the expected interaction force for the current gait phase. If a deviation exists, a smooth compensating torque is generated and superimposed on the main control command to suppress sudden force changes caused by prediction errors or gait abnormalities.

[0073] The final control commands undergo safety limiting before output: the torque does not exceed 80% of the motor's rated value, and the knee joint angle is limited to a safe range of 0 to 90 degrees. If a persistent anomaly is detected, the system will switch to a high-stiffness safety mode within 100 milliseconds and automatically shut down if the anomaly does not resolve within 5 seconds.

[0074] This example involved a 5-day training test in 10 stroke patients (Brunnstrom stages III to V). Results showed significant improvement in gait symmetry, more complete knee flexion during the swing phase, significantly reduced fluctuations in human-machine interaction forces, no falls or discomfort events throughout the training, and increased patient willingness to participate. These results validate the practical effectiveness of this invention in improving flexibility, safety, and rehabilitation efficacy.

[0075] Example 2: Taking a bilateral lower limb rehabilitation exoskeleton system as an example, this system is used to assist patients with incomplete spinal cord injury in treadmill-assisted walking training. Unlike Example 1, this system deploys complete sensing and actuation units in both legs, including hip and knee joint encoders, bilateral plantar pressure sensors, thigh and calf segment IMUs, and independent bilateral torque motor drive modules.

[0076] The system synchronously acquires motion and physiological signals from both limbs at a frequency of 100 Hz and predicts the gait phase of each leg separately. To accommodate the asymmetrical gait characteristics of spinal cord injury patients, the system allows independent phase prediction for each leg and calls upon corresponding admittance parameters to achieve asymmetric control. For example, when the right leg is in the early swing phase while the left leg is still in the middle stance phase, the system can simultaneously provide low stiffness for the right leg and high stiffness for the left leg, thus matching the patient's actual movement needs.

[0077] Regarding interactive force compensation, the system automatically sets a higher compensation gain based on the patient's muscle strength level (in this case, the lower limb muscle strength is grade 2–3) to enhance the response to weak active intentions. The reference trajectory is dynamically matched to a normal gait database based on the patient's height and gait speed, and scaled to 60%–80% of the range according to the rehabilitation stage.

[0078] Safety mechanisms have been further enhanced: In addition to torque and angle limits, the system also monitors the consistency between treadmill speed and stride frequency. If gait disjointness occurs (such as when the patient stops walking but the treadmill is still running), emergency braking is triggered immediately.

[0079] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since it corresponds to the method disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to in the method section.

[0080] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. 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 the invention. Therefore, the invention 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.

Claims

1. A gait prediction and compensation admittance control method for a lower limb rehabilitation exoskeleton, characterized in that, include: Multi-source motion information is collected in real time, and after filtering and time alignment processing, a unified multi-dimensional motion vector is obtained; The multidimensional motion vector is input into a pre-trained temporal prediction model, which outputs the prediction result of gait phase at future time. If the maximum probability output by the temporal prediction model is lower than a preset threshold, it automatically switches to a pre-built rule criterion for gait phase recognition. The admittance control parameters that match the gait phase prediction results are retrieved from the pre-built admittance parameter mapping table, and the master control torque is calculated based on the retrieved admittance control parameters. Based on the deviation between the actual human-computer interaction force and the expected interaction force, the compensation torque is calculated. The main control torque and the compensation torque are superimposed to form the final torque drive command for the exoskeleton motor.

2. The gait prediction and compensation admittance control method for lower limb rehabilitation exoskeleton as described in claim 1, characterized in that, The multidimensional motion vector includes at least: the real-time angle and angular velocity of the joints, the pressure distribution of the forefoot and heel of the left and right feet, and the triaxial acceleration and angular velocity of the thigh and calf.

3. The gait prediction and compensation admittance control method for lower limb rehabilitation exoskeleton as described in claim 1, characterized in that, The process of gait phase recognition based on rule-based criteria includes: A four-phase finite state machine is constructed based on plantar pressure and joint velocity, and a ground contact threshold is set. T on and ground threshold T 0ff The states of the finite state machine are defined as including the initial support state, the intermediate support state, the final support state, and the swing state. Plantar pressure data and joint angular velocity data were used as input data, with plantar pressure data including heel pressure. P h and forefoot pressure P f Joint angular velocity data should include at least the ankle joint angular velocity. ; The finite state machine applies a state transition rule based on the current state and input data. The state transition rule is as follows: If total plantar pressure And ankle joint angular velocity When a value changes from negative to positive, it transitions from an initial support state to a medium-term support state. If heel pressure P h The decrease exceeded the preset value, and the forefoot pressure Ankle joint angular velocity When the value is greater than 0, the support level shifts from a medium-term support level to a late-term support level. If total plantar pressure And ankle joint angular velocity When the trend is upward, it shifts from a late-stage support state to a swing state. If total plantar pressure At that time, it transitions from the swing state to the initial support state.

4. The gait prediction and compensation admittance control method for lower limb rehabilitation exoskeleton as described in claim 1, characterized in that, Admittance control parameters include stiffness and damping. Different phase states in the admittance parameter mapping table correspond to different stiffness and damping values. The calculation method for the master control torque is as follows: in, Indicates the controlling torque. Indicates gait phase prediction results The corresponding stiffness, Indicates gait phase prediction results The corresponding damping, Indicates joint angle, Indicates joint angular velocity, and This represents the reference trajectory and its derivative generated based on a normal gait database.

5. The gait prediction and compensation admittance control method for a lower limb rehabilitation exoskeleton as described in claim 1, characterized in that, The calculation process for the compensating torque includes: Calculate the expected interaction force: in, Indicates gait phase prediction results Corresponding stiffness; Indicates joint angle, This represents a reference trajectory generated based on a normal gait database; The actual human-computer interaction force is obtained by reverse calculation of electrode current or measurement by force sensor; Calculate the compensation torque based on the deviation between the actual human-computer interaction force and the expected interaction force: in, Indicates the compensating torque. G c Indicates adaptive compensation gain, LPF represents low-pass filtering, ΔF = F int F ref This represents the deviation between the actual human-computer interaction force and the expected interaction force.

6. The gait prediction and compensation admittance control method for a lower limb rehabilitation exoskeleton as described in claim 4, characterized in that, Also includes: The final torque drive command is subject to dual safety limits. The first limit is to restrict the amplitude of the final torque drive command from not exceeding 80% of the motor's rated value. The second level of limitation is: the joint angle is always within the physiological safety range; If the amplitude of the final torque drive command or the joint angle exceeds the limit, the system will switch to the high stiffness impedance safety mode within the first preset time period. If the dual safety limiting conditions are not met within the second preset time period under the high stiffness impedance safety mode, the machine will automatically shut down.

7. The gait prediction and compensation admittance control method for a lower limb rehabilitation exoskeleton as described in claim 6, characterized in that, The formula for calculating the control torque under high stiffness impedance safety mode is: in, This represents the control torque under high-stiffness impedance safety mode; This represents the stiffness in the high-stiffness impedance safety mode, and its value is the original stiffness. 2 to 4 times; This indicates damping in the high-stiffness impedance safety mode; This indicates the joint safety holding angle in the high stiffness impedance safety mode.

8. A gait prediction and compensation admittance control system for a lower limb rehabilitation exoskeleton, characterized in that, It employs the method described in any one of claims 1-7, comprising: The data acquisition module is used to collect multi-source motion information in real time, and after filtering and time alignment processing, a unified multi-dimensional motion vector is obtained. The gait phase prediction module is used to input multi-dimensional motion vectors into a pre-trained temporal prediction model and output the prediction results of gait phase at future moments. If the maximum probability output by the temporal prediction model is lower than a preset threshold, it automatically switches to a pre-built rule criterion for gait phase recognition. The adaptive admittance control module is used to retrieve admittance control parameters that match the gait phase prediction results from a pre-built admittance parameter mapping table, and calculate the master control torque based on the retrieved admittance control parameters. The error compensation module is used to calculate the compensation torque based on the deviation between the actual human-computer interaction force and the expected interaction force. The drive execution module is used to superimpose the main control torque and the compensation torque as the final torque drive command for the exoskeleton motor.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the processor executes the computer program, it implements the steps of the gait prediction compensation admittance control method for a lower limb rehabilitation exoskeleton as described in any one of claims 1-7.

10. A computer device, comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor; characterized in that, when the processor executes the computer program, it implements the steps of the gait prediction compensation admittance control method for a lower limb rehabilitation exoskeleton as described in any one of claims 1-7.

Images