Lower limb exoskeleton control system and method based on phase recognition and model predictive control

Abstract

The invention discloses a lower limb exoskeleton control system and method based on phase recognition and model prediction control. The system comprises a kinematics modeling module, a gait database module, a data processing and feature extraction module, a gait phase recognition module, a dynamics modeling module, a model prediction controller and an actuator control unit. By recognizing the gait phase and dynamically adjusting and controlling the target, precise assistance of the natural gait of the wearer is achieved. The system calculates the trajectory of the ankle joint relative to the hip joint in real time based on a kinematic model, and constructs a database in combination with gait data collected under multiple working conditions to provide an expected trajectory and reference parameters for control. In the dynamic modeling process, motor driving torque and man-machine interaction torque are introduced, a system dynamic model is established in a unified mode, and control characteristics under different gait phases are reflected.

Description

Technical Field

[0001] The present invention relates to the field of exoskeleton robot technology, and specifically to a lower limb exoskeleton control system and method based on phase recognition and model predictive control, which is suitable for scenarios such as rehabilitation training, gait assistance and enhanced mobility support systems. Background Art

[0002] Lower-limb exoskeletons, as wearable assistive devices, have been widely used in recent years for rehabilitation training, gait assistance, and enhanced mobility support. Existing exoskeleton control systems typically rely on preset gait trajectories or control strategies based on single joint angle feedback, making them difficult to fully adapt to individual differences and the dynamic changes in the wearer's gait. Some systems use fixed control models to control the entire gait cycle, lacking dynamic differentiation and personalized response between the stance and swing phases. This can easily lead to uncoordinated movements, delayed responses, and even compromise the wearer's gait stability and comfort.

[0003] Furthermore, existing technologies for identifying gait phases often rely on a single joint angle or foot contact state, failing to fully integrate multi-source information for accurate judgment. This results in a lack of continuity and stability in the control strategy during phase switching. Furthermore, traditional controllers often employ rule-based control or linear feedback, making it difficult to predict and optimize complex gait behaviors. They lack model adaptation capabilities and are unable to achieve personalized control with high precision and real-time response. Therefore, there is an urgent need for a lower limb exoskeleton control system and method that can combine multi-source gait recognition information and perform dynamic control adjustments based on a model prediction algorithm to improve the system's ability to follow, assist, and respond naturally to the wearer's movements. Summary of the Invention

[0004] The present invention aims to provide a lower limb exoskeleton control system and method based on phase recognition and model predictive control, to solve the problems existing in the prior art such as inaccurate phase recognition, rigid control strategy, and lack of dynamic adjustment capability, and to enhance the lower limb exoskeleton's ability to assist and respond to the wearer's natural gait.

[0005] A lower limb exoskeleton control system based on phase recognition and model predictive control, comprising: a kinematic modeling module, a gait database module, a data processing and feature extraction module, a gait phase recognition module, a dynamics modeling module, a model predictive controller, and an actuator control unit;

[0006] The kinematic modeling module calculates the trajectory of the ankle joint relative to the hip joint through a forward kinematic model based on the real-time acquired hip and knee joint angles, which is used to identify the wearer's gait phase;

[0007] The gait database module uses a motion capture laboratory to collect gait data in various scenarios: different speeds, different loads, uphill and downhill. It then builds a structured database, extracts and stores the expected gait trajectory and corresponding key parameters as reference input for model prediction.

[0008] The data processing and feature extraction module is used to pre-process and structure the raw signals collected by the sensor, providing high-quality input for gait phase recognition and controller decision-making;

[0009] The dynamic modeling module includes the dynamic equations of the stance phase and the swing phase, which are designed to describe the motion characteristics and dynamic behaviors of the lower limb exoskeleton at different stages of the gait cycle.

[0010] The model predictive controller (MPC) adjusts the dynamic model based on the identified current phase and predicts the gait behavior in the future to generate the optimal control sequence.

[0011] A lower limb exoskeleton control method based on phase recognition and model predictive control. It uses the pressure signal collected by the plantar sensor and the movement trajectory of the ankle joint relative to the hip joint calculated from the joint angle to jointly identify the phase in the wearer's gait cycle, thereby achieving precise assistance and natural following of the gait.

[0012] The gait data under different working conditions are collected through the motion capture system, and a database containing various gait patterns is constructed to generate the desired trajectory and reference control parameters.

[0013] The gait cycle includes the swing phase and the stance phase. A dynamic model is established for each phase to reflect the movement of the human body in different phases.

[0014] The identification result of the gait phase is input to the model predictive controller MPC, which dynamically switches the control model and the corresponding cost function according to the identified current phase, driving the exoskeleton to perform angle tracking control in the support phase and dynamic parameter adjustment in the swing phase.

[0015] During the stance phase, high-precision tracking control is achieved by optimizing the error between the exoskeleton joint angles and the wearer's actual joint angles in the model predictive controller (MPC) cost function.

[0016] During the swing phase, the model predictive controller (MPC) model parameters are dynamically adjusted to conform to the natural swinging behavior of the human body, thereby improving wearing comfort and naturalness.

[0017] In order to ensure smooth transition of control strategies between different gait phases, the Sigmoid function is introduced into the cost function of the model predictive controller (MPC) to achieve smooth weight switching.

[0018] The control method is applicable to a motor-driven lower limb exoskeleton device, in which the hip joint and knee joint are driven by motors, and the controller outputs the motor torque required for joint driving according to the recognition results.

[0019] The controller determines the wearer's gait phase based on the pressure signal collected by the plantar sensor and the movement trajectory of the ankle joint relative to the hip joint, adjusts the corresponding dynamic model and controller cost function according to the identified gait phase, and generates and outputs motor control instructions using a model predictive control algorithm; the drive unit is driven by a motor, and the controller outputs a motor torque signal for driving the hip joint and knee joint;

[0020] The model predictive controller includes an angular error optimization mechanism for the support phase and a dynamic model adjustment mechanism for the swing phase;

[0021] The cost function in the controller uses the Sigmoid function to smoothly switch the weights of different gait phases to ensure control continuity and stability;

[0022] The gait database module provides individualized reference trajectories and parameters to enhance the adaptability of the controller to different users;

[0023] The system is suitable for various application scenarios such as rehabilitation training, gait assistance and enhanced mobility support;

[0024] The kinematic model module is used to calculate the motion trajectory of the ankle joint relative to the hip joint based on the joint angle data collected in real time, and use the trajectory as auxiliary information, combined with the pressure data collected by the plantar sensor, to identify the wearer's gait phase.

[0025] The control system of the present invention mainly includes: a kinematic modeling module, a gait database module, a data processing and feature extraction module, a gait phase recognition module, a dynamic modeling module, a model predictive controller, and an actuator control unit.

[0026] The kinematic modeling module uses a forward kinematics model to calculate the trajectory of the ankle relative to the hip joint based on the real-time acquisition of hip and knee joint angles. This is used to identify the wearer's gait phase. By establishing a coordinate system {0}{1}{2}{3} for the centers of the hip, knee, and ankle joints, the forward kinematic modeling model is obtained:

[0027]

[0028] Where, is the homogeneous transformation matrix from the base coordinate system {0} to the coordinate system {1}, T1 2 is the homogeneous transformation matrix from coordinate system {1} to coordinate system {2}, is the homogeneous transformation matrix from coordinate system {2} to coordinate system {3}. The above forward kinematic model is obtained through three homogeneous transformations.

[0029] The gait database module collects gait data in various scenarios through the motion capture laboratory: gait data of different speeds, different loads, uphill and downhill, establishes a structured database, extracts and stores the expected gait trajectory and corresponding key parameters as reference input for model prediction.

[0030] The data processing and feature extraction module is used to preprocess and structure the raw signals collected by the sensor, providing high-quality input for gait phase recognition and controller decision-making.

[0031] The plantar pressure signal is first filtered through a low-pass filter to suppress high-frequency noise, and then combined with a sliding window filter algorithm to smooth fluctuations and enhance data stability. The sliding window filter algorithm processes the plantar pressure signal as follows:

[0032]

[0033] Where M is the window length, Δt is the sampling period, and F raw is the plantar signal before filtering, and F(t) is the plantar signal after filtering.

[0034] The ankle joint trajectory is calculated by first-order difference to obtain instantaneous velocity, and then normalized and standardized to adapt to different individual physiological parameters. The specific method of first-order difference calculation is as follows:

[0035]

[0036] Where p ankle (t) represents the position of the ankle joint at time t, v ankle (t) represents the velocity of the ankle joint at time t.

[0037] Key features extracted include the rate of change of plantar pressure threshold, peak ankle velocity, and the time difference between adjacent gait cycles, which are used to construct a state recognition vector. This feature data is labeled by combining threshold judgment rules with empirical models to assist in the precise delineation of gait phases and provide the controller with a basis for adjustable phase weights. The module supports online updating of feature weights and adaptive adjustment driven by historical data, enhancing the system's robustness in recognizing complex working conditions and individual differences.

[0038] The gait phase recognition module combines the pressure signal and ankle joint trajectory data collected by the plantar sensor, uses threshold judgment, and identifies the wearer's current gait phase in real time, including the support phase, swing phase, and transition phase.

[0039] The threshold judgment conditions for the support phase are as follows:

[0040]

[0041] Where, F threshould is the plantar pressure threshold, These are all empirical thresholds.

[0042] The threshold judgment conditions of the swing phase are as follows:

[0043]

[0044] Where, F threshould is the plantar pressure threshold, and These are all empirical thresholds.

[0045] In particular, the threshold judgment results are not swing phase / support phase, and are all identified as transition phase.

[0046] The dynamic modeling module contains the dynamic equations of the support phase and the swing phase, which are designed to describe the motion characteristics and dynamic behaviors of the lower limb exoskeleton in different stages of the gait cycle.

[0047] The kinetic equation of the support phase is:

[0048]

[0049] Where q is the joint angle, is the joint angular velocity, is the joint angular acceleration, M sup is a symmetric positive definite inertia matrix, C sup is the centrifugal force and the Coriolis matrix, G sup is the gravitational torque, τ act is the motor driving torque, τ int It is the human-computer interaction torque.

[0050] The dynamic equation of the swing phase is:

[0051]

[0052] Where q is the joint angle, is the joint angular velocity, is the joint angular acceleration, M swg is a symmetric positive definite inertia matrix, C swg is the centrifugal force and the Coriolis matrix, G swg is the gravitational torque, τ act is the motor driving torque, τ int It is the human-computer interaction torque.

[0053] The model predictive controller (MPC) adjusts the dynamic model based on the identified current phase and predicts the gait behavior in the future to generate the optimal control sequence.

[0054] The steps of the model predictive controller are as follows:

[0055] Step 1: Controller input state variables

[0056] Step 2: Determine the prediction time domain N and the dynamic equation.

[0057] To prepare for building a predictive model, rewrite the kinetic equation as follows:

[0058]

[0059] Step 3: Build a prediction model. According to the input state variables and the rewritten dynamic equation, the state differential equation is obtained, which is:

[0060]

[0061] In the formula, u=τ act , x1=q, f(x,u) is the state transition function.

[0062] The forward Euler method is used to discretize the state differential equation to obtain the prediction model, specifically: x(k+1)=x(k)+Δt·f(x(k),u(k));

[0063] Where f(x(k),u(k)) is the state transfer function at the current moment, which is determined by the dynamic equation and adjusted in real time according to the recognition results of the phase recognition module. Specifically:

[0064]

[0065] Where, f sup (x(k),u(k)) is the state transfer function of the support phase; f swg (x(k),u(k)) is the state transfer function of the swing phase; γ(k) is the weight coefficient, γ(k)∈[0,1], which is used to describe the degree of fusion of the support / swing model at the transition phase.

[0066] Step 4: Construct the cost function.

[0067] In the stance phase, the wearer's limbs bear the wearer's weight, and the control goal is to track the joint trajectory with high precision, minimizing the deviation between the exoskeleton joint angle and the actual joint angle of the human body. The cost function of the stance phase is:

[0068]

[0069] Where q d is the desired joint angle, q(k) is the current joint angle, Q s is the error weighting matrix of the support phase, α1 and α2 are the weight coefficients for adjusting the motor torque and the human-machine interaction torque, τ act is the motor driving torque, τ int It is the human-computer interaction torque.

[0070] In the swing phase, the legs are in an air-swinging state, and the control goal is to guide and coordinate the motion trajectory, which needs to conform to the natural inertial movement of the human body and reduce unnecessary interference. The cost function of the swing phase is:

[0071]

[0072] Where x d is the expected state, x(k) is the current state, Q w is the error weighting matrix of the support phase, β1 and β2 are the weight coefficients for adjusting the motor torque and the human-machine interaction torque, τ act is the motor driving torque, τ int It is the human-computer interaction torque.

[0073] During the phase transition phase, the system needs to smoothly switch the control strategy to avoid sudden changes. Therefore, a dynamic weight adjustment term is introduced into the cost function. The weight change of the control term is controlled by the Sigmoid function to ensure control stability and continuity. The cost function of the transition phase is specifically:

[0074] J t =(1-σ(t))·J sup +σ(t)·J swg ;

[0075] Where, J sup is the cost function of the support phase; J swg is the cost function of the swing phase; σ(t) is the smoothing factor, which is a Sigmoid function used to control the weights of the support phase and the swing phase in the transition phase, specifically:

[0076]

[0077] Where μ is the transition speed control coefficient, and t0 is the phase transition center moment.

[0078] Step 5: Solve the optimal control sequence, specifically:

[0079]

[0080] Step 6: Generate control instructions and take the optimal sequence u0 as the control instruction.

[0081] Step 7: Output control instructions, that is, output τ act =u0.

[0082] The beneficial effects of the present invention are as follows:

[0083] The present invention introduces a dynamic adjustment mechanism of the control target based on the gait phase into the model predictive control, thereby enhancing the intelligent adaptability of the control strategy under different gait phases.

[0084] In the stance phase, the control objectives are precise trajectory tracking and step length control to ensure that the foot safely crosses the ground; in the swing phase, the control objectives prioritize system stability and force balance, and the controller tends to minimize the mismatch of human-machine interaction forces; in the transition phase, the support phase and swing phase control objectives are weightedly fused through the smoothing factor σ(t) to ensure the continuity and robustness of the control strategy during the phase transition process.

[0085] The actuator control unit executes the corresponding drive instructions according to the instructions output by the model predictive control to achieve stable control and precise movement of the lower limb exoskeleton in different gait stages.

[0086] The optimal torque command output by the controller is transmitted through the actuator control unit to drive the motor in the exoskeleton device to complete the tracking movements of the hip and knee joints.

[0087] The control system adopts a modular design, which makes it easy to deploy on various types of electric exoskeleton platforms and can flexibly adjust control parameters according to mission requirements.

[0088] Compared with traditional rule-based or PID control methods, the present invention adopts a model prediction mechanism to significantly improve control performance and has predictive ability and robustness.

[0089] The gait database supports online updates. Combining historical records with machine learning algorithms, it can learn the wearer's habits over a long period of time and implement personalized control strategies.

[0090] This system is suitable for rehabilitation robots, and can improve the training effect and autonomous movement ability of patients with hemiplegia and paraplegia through precise auxiliary movements.

[0091] In industrial assistance scenarios, it can provide energy compensation and stable support for people who carry and handle heavy objects for a long time, reducing fatigue and injury risks.

[0092] In military scenarios, positioning and command modules can be embedded to perform complex tasks such as long-distance navigation and load-bearing marching, thereby improving soldiers' mobility.

[0093] The control method is applicable to lower limb motor-driven exoskeleton systems of any structure and does not rely on specific mechanical design.

[0094] This method has good versatility and platform independence, and is compatible with knee-hip dual-joint or multi-degree-of-freedom expansion solutions.

[0095] Compared with existing exoskeleton control methods, the present invention significantly improves control response speed, following accuracy and comfort.

[0096] In summary, the present invention provides a lower limb exoskeleton control system and method that integrates gait phase recognition and model predictive control, achieving precise assistance and natural following of the wearer's gait. It has good practicality, stability and scalability, and is suitable for multiple fields such as rehabilitation medicine, industrial assistance and military equipment. BRIEF DESCRIPTION OF THE DRAWINGS

[0097] Figure 1 It is the overall flow chart of the present invention.

[0098] Figure 2 Predict the control flow graph for the model.

[0099] Figure 3 Schematic diagram of the human lower limb gait cycle.

[0100] Figure 4 Schematic diagram of a lower limb exoskeleton worn by a human body.

[0101] Figure 5 Schematic diagram of the lower limb exoskeleton system structure.

[0102] Figure 6 Schematic diagram of the leg structure of the lower limb exoskeleton system. DETAILED DESCRIPTION

[0103] The present invention will be described in further detail below with reference to the embodiments and drawings, but the embodiments of the present invention are not limited thereto.

[0104] The control strategy of the present invention uses model predictive control (MPC) and combines plantar sensors with ankle-hip trajectory information for gait phase identification. During the stance phase, the control objective emphasizes accurate tracking of joint angles; during the swing phase, the model parameters are adjusted to accommodate natural swinging behavior. To achieve smooth transitions between phases, a Sigmoid function is introduced into the cost function for weight switching, ensuring control continuity and stability. This method is applicable to motor-driven lower limb exoskeleton systems and is widely used in scenarios such as rehabilitation training, gait assistance, and enhanced mobility support.

[0105] Example 1

[0106] This example, based on a rehabilitation training scenario, selected a patient with mild hemiplegia as the experimental subject and used the lower limb exoskeleton system of the present invention to perform assisted gait training. The system uses encoders installed at the hip and knee joints to collect joint angle information and uses plantar pressure sensors to obtain ground contact signals.

[0107] The system first calculates the ankle joint trajectory in real time using the forward kinematics modeling method based on the cubic homogeneous transformation of the coordinate system {0} to {3}, specifically:

[0108]

[0109] Where, is the homogeneous transformation matrix from the base coordinate system {0} to the coordinate system {1}, T1 2 is the homogeneous transformation matrix from coordinate system {1} to coordinate system {2}, is the homogeneous transformation matrix from coordinate system {2} to coordinate system {3}.

[0110] Then the instantaneous velocity of the ankle joint is calculated by the first-order difference method, specifically:

[0111]

[0112] Where p ankle (t) represents the position of the ankle joint at time t, v ankle (t) represents the velocity of the ankle joint at time t.

[0113] The gait phase recognition module identifies the current "stance phase" based on the plantar pressure threshold conditions. The threshold conditions met are:

[0114]

[0115] Where, F threshould is the plantar pressure threshold, These are all empirical thresholds.

[0116] The control system enters the support phase control strategy and adopts the corresponding dynamic model, specifically:

[0117]

[0118] Where q is the joint angle, is the joint angular velocity, is the joint angular acceleration, M sup is a symmetric positive definite inertia matrix, C sup is the centrifugal force and the Coriolis matrix, G sup is the gravitational torque, τ act is the motor driving torque, τ int It is the human-computer interaction torque.

[0119] The controller uses a model predictive control strategy, with a prediction horizon of 1 second and a control interval of 0.05 seconds. In the stance phase cost function, the goal is to minimize the error between the desired joint angle and the current angle. The optimization is combined with the stance phase cost function, specifically:

[0120]

[0121] Where q d is the desired joint angle, q(k) is the current joint angle, Q s is the error weighting matrix of the support phase, α1 and α2 are the weight coefficients for adjusting the motor torque and the human-machine interaction torque, τ act is the motor driving torque, τ int It is the human-computer interaction torque.

[0122] The gait enters the swing phase through the threshold judgment condition, specifically:

[0123]

[0124] Where, F threshould is the plantar pressure threshold, and These are all empirical thresholds.

[0125] The control strategy automatically switches to the swing-corresponding control target and adjusts the dynamic model, specifically:

[0126]

[0127] Where q is the joint angle, is the joint angular velocity, is the joint angular acceleration, M swg is a symmetric positive definite inertia matrix, C swg is the centrifugal force and the Coriolis matrix, G swg is the gravitational torque, τ act is the motor driving torque, τ int It is the human-computer interaction torque.

[0128] Synchronously, the cost function corresponding to the switching swing is:

[0129]

[0130] Where x d is the expected state, x(k) is the current state, Q w is the error weighting matrix of the support phase, β1 and β2 are the weight coefficients for adjusting the motor torque and the human-machine interaction torque, τ act is the motor driving torque, τ int It is the human-computer interaction torque.

[0131] In order to ensure a smooth transition from the support phase to the swing phase, a phase transition hybrid cost function is introduced, specifically:

[0132] J t=(1-σ(t))·J sup +σ(t)·J swg ;

[0133] Where, J sup is the cost function of the support phase; J swg is the cost function of the swing phase; σ(t) is the smoothing factor, which is a Sigmoid function used to control the weights of the support phase and the swing phase in the transition phase, specifically:

[0134]

[0135] Where μ is the transition speed control coefficient, and t0 is the phase transition center moment.

[0136] The final controller uses the above cost function to solve the optimal control sequence in the prediction time domain, specifically:

[0137]

[0138] Only the torque value at the first moment is applied and sent to the actuator, specifically:

[0139] τ act =u0;

[0140] In this embodiment, the system realizes continuous control of the wearer's smooth transition from the support phase to the swing phase. During the entire training cycle, the system exhibits good gait tracking performance and human-computer coordination performance.

[0141] Example 2

[0142] This embodiment is applied to an industrial assistance scenario, targeting a worker in a warehouse handling operation who must carry heavy objects at medium to low speeds for extended periods of time. The lower limb exoskeleton system is deployed at the hip and knee joints of both legs, collecting real-time joint status and plantar pressure to assist in maintaining posture stability and gait continuity during load-bearing handling.

[0143] In this application, the system uses the "heavy load" label data subset through the gait database module to obtain the expected gait trajectory and key reference features under the target working condition. The plantar pressure signal is low-pass filtered and smoothed, and the plantar pressure change rate is extracted as one of the key features, specifically:

[0144]

[0145] Where M is the window length, Δt is the sampling period, and F raw is the plantar signal before filtering, and F(t) is the plantar signal after filtering.

[0146] The gait phase recognition module identifies the current phase transition (from support to swing) based on the plantar pressure threshold and ankle joint velocity. To adapt to high-frequency gait fluctuations, the control system uses a transition cost function, specifically:

[0147] J t =(1-σ(t))·J sup +σ(t)·J swg ;

[0148] Where, J sup is the cost function of the support phase; J swg is the cost function of the swing phase; σ(t) is the smoothing factor, which is a Sigmoid function used to control the weights of the support phase and the swing phase in the transition phase, specifically:

[0149]

[0150] Where μ is the transition speed control coefficient, and t0 is the phase transition center moment.

[0151] The controller predicts the control quantity for the next 10 steps through Euler discrete prediction (0.5s prediction range), specifically:

[0152] x(k+1)=x(k)+Δt·f(x(k),u(k));

[0153] Where f(x(k),u(k)) is the state transfer function at the current moment, which is determined by the dynamic equation and adjusted in real time according to the recognition results of the phase recognition module.

[0154] Solving the optimal control torque τ based on the optimization objective function act =u0, sent to the driver, specifically:

[0155]

[0156] The control system demonstrates good stability and load adaptability in different handling tasks. Even under a load of 15kg, it can still maintain the human-machine interaction torque within a low range, effectively reducing the operator's energy consumption and fatigue.

[0157] Example 3

[0158] This embodiment is applied to military long-distance load-bearing scenarios, simulating a soldier performing a 5km long-distance hiking mission, wearing an electrically driven lower limb exoskeleton (with a carrying system and positioning module), carrying a load of about 20km, and walking at a speed of about 1.1m / s.

[0159] Gait data is collected on-site during the training period to establish an individualized reference model, and the gait cycle feature vector and recognition threshold are adjusted online.

[0160] To resist the dynamic disturbance caused by terrain changes, the system modifies the transfer function f(x(k),u(k)) in the prediction model according to environmental calibration parameters (such as slope and load), that is, adding a disturbance factor δ(t):

[0161]

[0162] Where, f sup (x(k),u(k)) is the state transfer function of the support phase; f swg (x(k),u(k)) is the state transition function for the swing phase; γ(k) is the weight coefficient, γ(k)∈[0,1], which describes the degree of integration between the support and swing models during the transition phase. δ(t) represents the prediction correction term introduced by non-ideal terrain, derived from the ground contact instability estimate.

[0163] The controller solves the optimization problem in real time and outputs torque commands to drive the hip and knee motors. Because soldiers carry heavy loads, the actuator control unit incorporates an anti-saturation strategy into the drive control to prevent output overload. This strategy incorporates dynamic torque limiting based on the current motor model.

[0164] The control system maintained good tracking accuracy (±3.2°) under high-load and complex terrain conditions, significantly outperforming traditional PID control (±6.4°). Wearers exhibited delayed muscle fatigue and more consistent gait during extended, loaded marches, demonstrating the control strategy's adaptability to real-world situations.

Claims

1. A lower limb exoskeleton control system based on phase recognition and model predictive control, characterized in that include: Kinematic modeling module, gait database module, data processing and feature extraction module, gait phase recognition module, dynamic modeling module, model predictive controller, and actuator control unit; The kinematic modeling module calculates the trajectory of the ankle joint relative to the hip joint through a forward kinematic model based on the real-time acquired hip and knee joint angles, which is used to identify the wearer's gait phase; The gait database module uses a motion capture laboratory to collect gait data in various scenarios: different speeds, different loads, uphill and downhill. It then builds a structured database, extracts and stores the expected gait trajectory and corresponding key parameters as reference input for model prediction. The data processing and feature extraction module is used to pre-process and structure the raw signals collected by the sensor, providing high-quality input for gait phase recognition and controller decision-making; The dynamic modeling module includes the dynamic equations of the stance phase and the swing phase, which are designed to describe the motion characteristics and dynamic behaviors of the lower limb exoskeleton at different stages of the gait cycle. The model predictive controller (MPC) adjusts the dynamic model based on the identified current phase and predicts the gait behavior in the future to generate the optimal control sequence.

2. A lower limb exoskeleton control method based on phase recognition and model predictive control, characterized by: The pressure signal collected by the plantar sensor and the movement trajectory of the ankle joint relative to the hip joint calculated from the joint angle are used to jointly identify the phase in the wearer's gait cycle, thereby achieving precise assistance and natural following of the gait.

3. The method according to claim 1, wherein: The gait data under different working conditions are collected through the motion capture system, and a database containing various gait patterns is constructed to generate the desired trajectory and reference control parameters.

4. The method according to claim 1, wherein: The gait cycle includes the swing phase and the stance phase. A dynamic model is established for each phase to reflect the movement of the human body in different phases.

5. The method according to claim 1, wherein: The identification result of the gait phase is input to the model predictive controller MPC, which dynamically switches the control model and the corresponding cost function according to the identified current phase, driving the exoskeleton to perform angle tracking control in the support phase and dynamic parameter adjustment in the swing phase.

6. The method according to claim 5, characterized in that: During the stance phase, high-precision tracking control is achieved by optimizing the error between the exoskeleton joint angles and the wearer's actual joint angles in the model predictive controller (MPC) cost function.

7. The method according to claim 5, characterized in that: During the swing phase, the model predictive controller (MPC) model parameters are dynamically adjusted to conform to the natural swinging behavior of the human body, thereby improving wearing comfort and naturalness.

8. The method according to claim 5, characterized in that: In order to ensure smooth transition of control strategies between different gait phases, the Sigmoid function is introduced into the cost function of the model predictive controller (MPC) to achieve smooth weight switching.

9. The method according to claim 1, wherein: The control method is applicable to a motor-driven lower limb exoskeleton device, in which the hip joint and knee joint are driven by motors, and the controller outputs the motor torque required for joint driving according to the recognition results.

10. The method according to claim 1, wherein: The controller determines the wearer's gait phase based on the pressure signal collected by the plantar sensor and the movement trajectory of the ankle joint relative to the hip joint, adjusts the corresponding dynamic model and controller cost function according to the identified gait phase, and generates and outputs motor control instructions using a model predictive control algorithm; the drive unit is driven by a motor, and the controller outputs a motor torque signal for driving the hip joint and knee joint; The model predictive controller includes an angular error optimization mechanism for the support phase and a dynamic model adjustment mechanism for the swing phase; The cost function in the controller uses the Sigmoid function to smoothly switch the weights of different gait phases to ensure control continuity and stability; The gait database module provides individualized reference trajectories and parameters to enhance the adaptability of the controller to different users; The system is suitable for various application scenarios such as rehabilitation training, gait assistance and enhanced mobility support; The kinematic model module is used to calculate the motion trajectory of the ankle joint relative to the hip joint based on the joint angle data collected in real time, and use the trajectory as auxiliary information, combined with the pressure data collected by the plantar sensor, to identify the wearer's gait phase.

Images