A flexible drive exoskeleton control method and device based on personalized trajectory planning

Abstract

A method and device for controlling a flexible driven exoskeleton based on personalized trajectory planning is disclosed. The method includes: 1) dividing the gait trajectory into a single-leg support phase and a double-leg support phase based on pressure sensing, and storing the corresponding gait trajectory data; 2) planning the double-leg support phase trajectory using the DMP algorithm during the single-leg support phase, and planning the gait trajectory of the single-leg swing phase using the DMP algorithm and adaptive planning during the double-leg support phase; 3) inputting the generated reference trajectory into an impedance controller as the reference trajectory for the impedance controller, and generating the output torque; 4) completing the accurate modeling of the torque based on hysteresis force modeling, and using a kinematic controller and flexible actuator to complete the exoskeleton-assisted control. This invention utilizes pressure sensors to divide the gait trajectory into single-leg support and double-leg support states, avoiding deviations in initial and end positions, simplifying the gait trajectory, reducing trajectory learning time, making the trajectory deterministic, eliminating the need to predict unknown trajectories, and providing sufficient time for trajectory learning.

Description

Technical Field

[0001] This invention relates to a flexible drive exoskeleton control method and device based on personalized trajectory planning. It utilizes the wearer's real-time gait data to generate a reference trajectory through generalization of Dynamic Movement Primitives (DMPs). Based on the reference trajectory, impedance control is used to achieve human-computer interaction control, and hysteresis force modeling and flexible actuators are used to achieve precise torque output. Background Technology

[0002] With the increasing aging population, the disability rates caused by diseases such as stroke, traumatic brain injury, and spinal cord injury are rising year by year, leading to a continuous increase in rehabilitation needs. However, my country faces a severe shortage of rehabilitation physicians, nurses, and rehabilitation therapists, placing enormous pressure on the social medical rehabilitation system. To address this issue, many scholars and rehabilitation medical professionals have begun to focus on rehabilitation robots. Exoskeleton robots integrate sensing and control technologies, featuring interdisciplinary characteristics including bionics, robotics, information and control science, and medicine. They can provide assistance to wearers or help patients with rehabilitation training, improving patients' spinal cord injuries and walking abilities.

[0003] Current exoskeleton robots primarily achieve assisted control through physical human-machine interaction or based on fixed gait planning trajectories. The goal of gait planning is to generate gait trajectories consistent with the patient's normal movement patterns. A good control strategy can not only meet the needs of patients at different stages of rehabilitation but also allow for timely adjustments to training strategies, improving patient comfort and achieving ideal rehabilitation outcomes. However, existing exoskeleton robots lack personalized adaptation and online learning capabilities. Most control methods require predefined reference trajectories, making it difficult to cover the gait information of different patients. This may lead to a poor experience for the wearer, and rigid control can even cause secondary injuries.

[0004] For wearers, imitation learning of gait trajectories can assist walking and rehabilitation training, resulting in more personalized gait trajectories. DMP (Device Modeling) is a typical example of imitation learning, but besides distortion and mirroring issues when the initial and final positions are close, it typically divides the trajectory learning based on a single-sided trajectory, dividing it into a cycle or support and swing phases, resulting in a relatively lagging motion trajectory. Furthermore, exoskeleton-assisted control has strong human-computer interaction, requiring strict safety measures. For safety, flexible actuators and flexible control algorithms are suitable methods, but flexible actuators themselves exhibit significant hysteresis, causing a lag in output torque relative to the actuator's rotation angle, severely hindering the exoskeleton's control performance. This problem needs to be overcome through appropriate hysteresis force modeling. Therefore, providing a flexible-driven exoskeleton control method based on personalized trajectory planning is of great significance. Summary of the Invention

[0005] This invention aims to overcome the aforementioned shortcomings of existing technologies by providing a method and apparatus based on DMP imitation learning and flexible drive control. This invention divides the gait trajectory into a two-legged support phase and a single-leg support phase based on bilateral gait trajectories, and combines this with DMP adaptive trajectory planning to generalize and generate a gait trajectory that more closely resembles the human body's motion reference. Furthermore, through impedance control and a flexible actuator modeled with hysteresis force, it provides patients with more personalized and compliant assistance.

[0006] The technical solution adopted in this invention is: a method for trajectory planning and control of a lower limb exoskeleton, comprising:

[0007] S1. Divide the gait trajectory into single-leg support phase and double-leg support phase based on pressure sensing, and store the corresponding gait trajectory data;

[0008] S2. During the single-leg support phase, the DMP algorithm is used to plan the trajectory of the double-leg support phase. During the double-leg support phase, the DMP algorithm and adaptive planning are used to plan the gait trajectory of the single-leg swing phase.

[0009] S3. Input the generated reference trajectory into the impedance controller as the reference trajectory for the impedance controller to generate the output torque;

[0010] S4. Based on the hysteresis force modeling, complete the accurate modeling of torque, and use the kinematic controller and flexible actuator to complete the exoskeleton assisted control.

[0011] Preferably, step S1 specifically includes:

[0012] Upon entering walking mode, based on the values ​​from multiple different locations on the plantar pressure sensors of both legs, if all values ​​on the plantar pressure sensors of one leg are below a threshold, the current gait is determined to be a single-leg support state; otherwise, it is a two-leg support state. In the single-leg support state, the pressure sensor readings determine whether the support is on the left or right leg. After determination, the current gait state is recorded, and the sensor data from the wearer at this time is stored in the current gait state data. The gait state switching and data flow are as follows: Figure 1 As shown.

[0013] When the gait state changes, check if the switching time is less than a time threshold to avoid incorrect switching due to anomalies such as sensor data loss. If the time is greater than the set time threshold, switch the current motion state and proceed to step 2. The gait state determination is as follows: Figure 2 As shown.

[0014] This simplified segmentation of the trajectory avoids situations where the initial and final positions of the trajectory are close together, while also simplifying the segmented trajectory and reducing learning time. Furthermore, the division of the trajectory into left and right legs makes trajectory planning more coordinated, providing the necessary time for trajectory learning without the need to predict missing trajectories.

[0015] Preferably, step S2 specifically includes:

[0016] When the wearer needs to adjust their walking direction, they can combine the contralateral trajectory with the Dynamic Motion Element (DMP) for adaptive planning to adjust the swing phase trajectory and thus plan a more suitable gait trajectory.

[0017] Furthermore, step S1 includes the following steps:

[0018] S2.1: When the direction of movement deviates from the original direction by Δθ, the healthy stride length is L, and the human stride width is w. s When this happens, the deviation radius R can be obtained:

[0019]

[0020] The planned stride length L is obtained based on the principle that both legs have the same walking speed. g :

[0021] L g =L+w s sin(Δθ) (2)

[0022] When the leg length is L leg The reference hip joint angle is θ. h At that time, the reference step size L is approximately:

[0023] L = L leg sinθ h(3)

[0024] When Δθ s The orientation angle during the current planning phase, Δθ l Given the orientation angle during the current planning phase, Δθ is the walking offset angle:

[0025] Δθ=Δθ s -Δθ l (4)

[0026] Planning the lateral hip joint θ gh for:

[0027]

[0028] Planning the lateral knee joint θ gh for:

[0029]

[0030] Among them, w s For step width, θ h For reference hip joint angle, L leg For leg length, l h Leg length at the hip joint.

[0031] S2.2: The joint trajectory of the data recorded in S1 This is used as the teaching trajectory for the DMP. The model of the DMP motion primitives is as follows:

[0032]

[0033] The role of τ is the time scaling factor of the control system, and θ g y0 is the target position, y0 is the starting position, and α is the target position. y and β y This is the gain coefficient.

[0034] The nonlinearity forcing f is:

[0035]

[0036] Among them, Ψ i Let ω be the Gaussian function. i The weight is x, which comes from a first-order system.

[0037] Based on the teaching trajectory, we can obtain the nonlinear target value f. target Defined as:

[0038]

[0039] In this invention, a loss function is constructed, and the locally weighted regression (LWR) optimization method is used to learn and solve the model parameters of the basis functions. The constructor J is as follows:

[0040]

[0041] Where P represents the time step of the trajectory, and ξ(t) is x(t)(θ) g -y0), its solution is:

[0042]

[0043] in:

[0044]

[0045]

[0046] S2.3: When the trajectory is in the single-leg support phase, a double-leg support phase trajectory needs to be planned. The current angle is used as the initial angle for DMP trajectory generalization. Since the double-leg support phase trajectory does not change much, the data recorded at the end of S1 can be retained as the end position. When in the double-leg support phase, a single-leg support phase is planned. The current trajectory is used as the initial position for DMP generalization to generate the gait trajectory. The angle planned in S2.1 is used as the end angle of the generalized gait trajectory to achieve adaptive step length adjustment. The planned swing phase trajectory is used as the gait reference trajectory for the S3 swing phase, and the planned support phase trajectory is used as the gait reference trajectory for the S3 support phase.

[0047] In existing exoskeleton control systems, the Data Manipulation Platform (DMP) is typically used to reproduce the gait trajectory of a reference trajectory or unilateral leg movement. However, this method suffers from lag when encountering curvilinear motion changes such as turning, resulting in poor human-computer interaction. In this invention, instead of using the DMP to reproduce the previous movement trajectory, the gait trajectory is planned using the contralateral trajectory to improve the responsiveness of trajectory planning.

[0048] Step S3 preferably includes:

[0049] By controlling impedance, the exoskeleton exhibits the characteristics of spring damping, thereby enhancing its controllability and safety.

[0050] The output calculation formula for the impedance controller is:

[0051]

[0052] Where M, B, and K represent the system's inertia coefficient, damping coefficient, stiffness coefficient, and τ, respectively. ext For controller output.

[0053] q can be a reference trajectory generated by S2 adaptive planning trajectory, q r This indicates the current position of the joint. In practice, various resistances τ are taken into account. dis The formula for calculating the human-computer interaction force τ can be simplified to:

[0054]

[0055] By substituting the torque obtained from impedance control calculations into S4, safe and compliant coordinated control can be achieved.

[0056] Preferably, step S4 specifically includes:

[0057] This invention achieves precise control of the output torque by designing an asymmetric hysteresis force modeling method to fit the hysteresis output curve of a flexible actuator.

[0058] The hysteresis force model designed in this invention divides the hysteresis force curve into three sub-curves: the rising segment, the falling segment, and the transition segment. The rising segment and the falling segment sub-curves represent the hysteresis force curves of the flexible actuator output torque as the angle increases and decreases, respectively, while the transition segment sub-curve represents a series of complex curves located between the two.

[0059] S4.1: Consider modeling each segmented hysteresis force curve to obtain equation (17), τ al (s), τ dl (s), τ tl (s) represent the fitting functions for the rising segment, falling segment, and transition segment sub-curves, respectively, and are defined by the segmentation point s. 01 and s 02 As the basis for judging the sub-curve, the specific data of the segment points are updated in real time by equations (18) and (19). After obtaining the data of one segment point, the other point can also be determined at the same time.

[0060]

[0061]

[0062] S4.2: The rising and falling sub-curves are fitted by selecting calibration points. Figure 3 This is a schematic diagram of selecting calibration points on the hysteresis curve. Connecting each calibration point in sequence is used as a fitting curve to obtain equations (20) and (21), where i, j and p, q represent calibration points at adjacent positions on the rising and falling segments, respectively.

[0063] τ al (s)=k ij (ss i )+τ i if s∈[s i ,s j(20)

[0064] τ dl (s)=k pq (ss p )+τ p if s∈[s p ,s q ] (twenty one)

[0065] S4.3: Since the transition segment curve has various shapes, directly fitting with the calibration point has obvious errors. Consider using a power function and Maxwell-slip combined model with more similar shapes to fit the transition segment curve. The torque fitting function is expressed as equation (22)-(24).

[0066]

[0067] Denotes the power function term, b(s) 0i ) is the exponential term, used to control the specific shape of the hysteresis fitting curve; Δτ m (s,s 0i The expression ) represents the difference function between the actual torque and the power function term established by the Maxwell-slip model, used to fit the hysteresis characteristics, and consists of multiple basic sliding units F with different parameters. i Obtained by superposition. Basic sliding element and Δτ m (s,s 0i The output result is as follows Figure 4 As shown.

[0068] S4.4: By applying the motion trajectory obtained from the hysteresis model to a dynamic motion control algorithm, precise position control of the motor can be achieved, thereby enabling precise torque output from the flexible actuator. The kinematic control block diagram is shown below. Figure 5 As shown.

[0069] Upon system startup, the system first initializes all modules, enabling parallel processing. Initially, unaware of the wearer's movement, the system enters a follow-up state, with the exoskeleton moving with the user. When a reference trajectory is available, the exoskeleton performs human-machine collaborative control based on that trajectory. The overall process is as follows: Figure 6 As shown. Its trajectory generation flowchart is as follows. Figure 7 As shown, when the state machine switches, and the time and data volume requirements are met, a gait trajectory is generated and the generated motion reference trajectory is given to the controller to achieve human-machine interaction control. Based on the state machine, it can be known that the system will enter collaborative control after the wearer takes their second step.

[0070] A second aspect of the present invention relates to a flexible driven exoskeleton control method and apparatus based on personalized trajectory planning, comprising a memory and one or more processors, wherein the memory stores executable code, and the one or more processors execute the executable code to implement the lower limb exoskeleton auxiliary control method of the present invention.

[0071] The third invention relates to a computer-readable storage medium having a program stored thereon, which, when executed by a processor, implements the lower limb exoskeleton-assisted control method of the present invention.

[0072] In this invention, the control strategy first uses a pressure sensor to determine whether the gait is in a single-leg or double-leg support phase and stores the corresponding inertial sensor data. Then, after adaptively adjusting the step length, a gait reference trajectory is generated using a dummy power management system (DMP). Impedance control generates the output torque based on the reference trajectory. Finally, after modeling the flexible actuator using hysteresis force modeling, personalized, safe, and compliant auxiliary control can be achieved through the motor and the flexible actuator.

[0073] This invention aims to address the shortcomings of existing exoskeleton control systems in terms of personalization and safety compliance. It utilizes pressure sensors to divide gait trajectories into single-leg and double-leg support states, avoiding situations where initial and end-effector positions are similar, simplifying the gait trajectory, reducing trajectory learning time, ensuring a predictable trajectory without the need for unknown trajectory prediction, and providing ample time for trajectory learning. The DMP algorithm, combined with the current position and adaptively planned target position, plans asynchronous motion trajectories for both joints. Furthermore, impedance control is used to plan auxiliary torques in real-time based on the planned trajectory and the current joint trajectory. The hysteresis model divides the complete hysteresis force curve into rising, falling, and transition segments. By using a model combining power functions and Maxwell-slip parameters instead of the traditional transition segment sub-curve fitting method, the accuracy of hysteresis force modeling is improved, enhancing the exoskeleton's control performance. Finally, kinematic control achieves personalized and compliant control of the system, improving wearer comfort and safety.

[0074] Compared with the prior art, the beneficial effects of the present invention are reflected in:

[0075] 1. By using bipedal pressure sensors, the gait trajectory is divided into single-leg support and double-leg support states, avoiding situations where the initial and final positions of the trajectory are close. This simple segmentation method simplifies the gait trajectory and reduces trajectory learning time.

[0076] 2. This division method, alternating between single-leg and double-leg support phases, allows the trajectory to be determined without predicting unknown trajectories, and provides ample time for trajectory learning. Furthermore, this method can coordinate the planning of both-leg movement trajectories.

[0077] 3. This invention fully considers the needs of patients when changing direction or turning while walking. It adaptively adjusts stride length and joint angles to cope with the offset angle during changes of direction or turns, and uses Dynamic Motion Planning (DMP) to generate corresponding personalized motion trajectories, thereby avoiding the problem of overly stiff movements during changes of direction or turns. This not only improves the assistive effect of the exoskeleton, but also significantly improves the user experience.

[0078] 4. By modeling hysteresis forces, the inherent hysteresis problem of flexible actuators can be solved, thereby improving the control performance of exoskeletons.

[0079] 5. The system of this invention starts up quickly; it can enter normal working mode as soon as the wearer takes the second step.

[0080] 6. By providing assistance to the wearer through upper impedance control and flexible actuators, the flexibility, safety, and human-computer interaction performance of the exoskeleton are improved. Attached Figure Description

[0081] Figure 1 This is a schematic diagram of gait state switching and data flow of the present invention.

[0082] Figure 2 This is a flowchart of the gait state switching process of the present invention.

[0083] Figure 3 This is a schematic diagram of selecting calibration points on the hysteresis force curve of the present invention.

[0084] Figure 4 This is a schematic diagram of fitting the hysteresis force curve using the Maxwell-slip method of the present invention.

[0085] Figure 5 This is a kinematic control block diagram of the present invention.

[0086] Figure 6 This is the overall process control diagram of the present invention.

[0087] Figure 7 This is a flowchart of the trajectory generation process of the present invention.

[0088] Figure 8 This is a schematic diagram illustrating the working principle of the present invention. Detailed Implementation

[0089] The technical solution of the present invention will be further described below with reference to the accompanying drawings.

[0090] Example 1

[0091] Reference Figures 1-8 This embodiment relates to an exoskeleton control method, including the following steps:

[0092] S1. Divide the gait trajectory into single-leg support phase and double-leg support phase based on pressure sensing, and store the corresponding gait trajectory data;

[0093] S2. During the single-leg support phase, the DMP algorithm is used to plan the trajectory of the double-leg support phase. During the double-leg support phase, the DMP algorithm and adaptive planning are used to plan the gait trajectory of the single-leg swing phase.

[0094] S3. Input the generated reference trajectory into the impedance controller as the reference trajectory for the impedance controller to generate the output torque;

[0095] S4. Based on the hysteresis force modeling, complete the accurate modeling of torque, and use the kinematic controller and flexible actuator to complete the exoskeleton assisted control.

[0096] Step S1 specifically includes:

[0097] Upon entering walking mode, based on the values ​​from multiple different locations on the plantar pressure sensors of both legs, if all values ​​on the plantar pressure sensors of one leg are below a threshold, the current gait is determined to be a single-leg support state; otherwise, it is a two-leg support state. In the single-leg support state, the pressure sensor readings determine whether the support is on the left or right leg. After determination, the current gait state is recorded, and the sensor data from the wearer at this time is stored in the current gait state data. The gait state switching and data flow are as follows: Figure 1 As shown.

[0098] When gait state changes, check if the switching time is less than a time threshold to avoid abnormal switching due to sensor data loss or other anomalies. If the time is greater than the set time threshold, switch the current motion state and proceed to step 2. The gait state switching flowchart is shown below. Figure 2 As shown.

[0099] Step S2 specifically includes:

[0100] When the wearer needs to adjust their walking direction, they can combine the contralateral trajectory with the Dynamic Motion Element (DMP) for adaptive planning to adjust the swing phase trajectory and thus plan a more suitable gait trajectory.

[0101] S2.1: When the direction of movement deviates from the original direction by Δθ, the reference step length is L, and the human stride width is w. s When this happens, the deviation radius R can be obtained:

[0102]

[0103] The planned stride length L is obtained based on the principle that both legs have the same walking speed. g :

[0104] L g =L+ws sin(Δθ) (2)

[0105] When the leg length is L leg The reference hip joint angle is θ. h At that time, the reference step size L is approximately:

[0106] L = L leg sinθ h (3)

[0107] When Δθ s The orientation angle during the current planning phase, Δθ l Given the orientation angle during the current planning phase, Δθ is the walking offset angle:

[0108] Δθ=Δθ s -Δθ l (4)

[0109] Planning the lateral hip joint θ gh for:

[0110]

[0111] Planning the lateral knee joint θ gh for:

[0112]

[0113] Among them, w s For step width, θ h To plan the lateral hip joint angle, L leg For leg length, l h Leg length at the hip joint.

[0114] S2.2: The joint trajectory of the data recorded in S1 This is used as the teaching trajectory for the DMP. The model of the DMP motion primitives is as follows:

[0115]

[0116] The role of τ is the time scaling factor of the control system, and θ g Let y0 be the target position and α be the starting position. y and β y This is the gain coefficient.

[0117] The nonlinearity forcing f is:

[0118]

[0119] Among them, Ψ i Let ω be the Gaussian function. i The weight is x, which comes from a first-order system.

[0120] Based on the teaching trajectory, we can obtain the nonlinear target value f. target Defined as:

[0121]

[0122] In this scheme, a loss function is constructed, and the locally weighted regression (LWR) optimization method is used to learn and solve the model parameters of the basis functions. The constructor J is as follows:

[0123]

[0124] Where P represents the time step of the trajectory, and ξ(t) is x(t)(θ) g -y0), its solution is:

[0125]

[0126] in:

[0127]

[0128] S2.3: When the trajectory is in the single-leg support phase, a double-leg support phase trajectory needs to be planned. The current angle is used as the initial angle for DMP trajectory generalization. Since the double-leg support phase trajectory does not change much, the data recorded at the end of S1 can be retained as the end position. When in the double-leg support phase, a single-leg support phase is planned. The current trajectory is used as the initial position for DMP generalization to generate the gait trajectory. The angle planned in S2.1 is used as the end angle of the generalized gait trajectory to achieve adaptive step length adjustment. The planned swing phase trajectory is used as the gait reference trajectory for the S3 swing phase, and the planned support phase trajectory is used as the gait reference trajectory for the S3 support phase.

[0129] Step S3 specifically includes:

[0130] By using impedance control as the upper-level control of the system, the exoskeleton exhibits the characteristics of spring damping, thereby enhancing its control compliance and safety.

[0131] The output calculation formula for the impedance controller is:

[0132]

[0133] Where M, B, and K represent the system's inertia coefficient, damping coefficient, stiffness coefficient, and τ, respectively. ext For controller output.

[0134] q can be a reference trajectory generated by S2 adaptive planning trajectory, q r This indicates the current position of the joint. In practice, various resistances τ are taken into account. dis The formula for calculating the human-computer interaction force τ can be simplified to:

[0135]

[0136] By substituting the torque obtained from impedance control calculations into S4, safe and compliant coordinated control can be achieved.

[0137] Step S4 specifically includes:

[0138] This invention achieves precise control of output torque by designing a hysteresis force modeling method to fit the hysteresis output curve of a flexible actuator.

[0139] The hysteresis force model designed in this invention divides the hysteresis force curve into an ascending segment, a descending segment, and a transition segment sub-curve. The ascending segment and descending segment sub-curves represent the hysteresis force changes during the rising and falling processes of the flexible actuator's output torque, respectively, while the transition segment sub-curve represents a series of complex curves during the transition between the two states.

[0140] S4.1: Consider modeling the segmented hysteresis curve to obtain equation (17), τ al (s), τ dl (s), τ tl (s) represent the fitting functions for the rising segment, falling segment, and transition segment sub-curves, respectively, and are defined by the segmentation point s. 01 and s 02 As the basis for judging the sub-curve, the specific data of the segment points are updated in real time by equations (18) and (19). After determining the data of one segment point, the corresponding other segment point is calculated by the fitting function that describes the relationship between the segment points.

[0141]

[0142] S4.2: The sub-curves of the rising and falling segments are fitted by selecting calibration points. Figure 3 This is a schematic diagram of selecting calibration points on the hysteresis force curve. Connecting each calibration point in sequence yields the polynomial fitting torque formulas of equations (20) and (21), where i, j and p, q represent adjacent calibration points in the rising and falling segments, respectively.

[0143] τ al (s)=k ij (ss i )+τ i if s∈[s i ,s j (20)

[0144] τ dl (s)=k pq (ss p )+τ p if s∈[s p ,sq ] (twenty one)

[0145] S4.3: Since there is a significant error in directly fitting the transition segment curve with the calibration point, we consider using a combination of the power function and the Maxwell-slip method with more similar shape to achieve model fitting. The torque fitting function can be expressed as equation (22)-(24).

[0146]

[0147] The term represents the power function, used to control the specific shape of the hysteresis fitting curve; Δτ m (s,s 0i The expression ) represents the difference function between the actual torque and the power function term established by the Maxwell-slip model, used to fit the hysteresis characteristics, and consists of multiple basic sliding units F with different parameters. i Obtained by superposition. Basic sliding element and Δτ m (s,s 0i The output result is as follows Figure 4 As shown.

[0148] S4.4: By applying the motion trajectory obtained from the hysteresis model to a dynamic motion control algorithm, precise position control of the motor can be achieved, thereby enabling precise torque output from the flexible actuator. The kinematic control block diagram is shown below. Figure 5 As shown.

[0149] Upon system startup, the system first initializes all modules, enabling parallel processing. Initially, unaware of the wearer's movement, the system enters a follow-up state, with the exoskeleton moving with the user. When a reference trajectory is available, the exoskeleton performs human-machine collaborative control based on that trajectory. The overall process is as follows: Figure 6 As shown. Its trajectory generation flowchart is as follows. Figure 7 As shown, when the state machine switches, and the time and data volume requirements are met, a gait trajectory is generated and the generated motion reference trajectory is given to the controller to achieve human-machine interaction control. Based on the state machine, it can be known that the system will enter collaborative control after the wearer takes their second step.

[0150] Example 2

[0151] This embodiment relates to a lower limb exoskeleton assistive control device, including a memory and one or more processors. The memory stores executable code, and when the one or more processors execute the executable code, they are used to implement the lower limb exoskeleton assistive control method of Embodiment 1.

[0152] Example 3

[0153] This embodiment relates to a computer-readable storage medium storing a program that, when executed by a processor, implements the lower limb exoskeleton-assisted control method of Embodiment 1.

[0154] The embodiments described in this specification are merely examples of implementations of the inventive concept. The scope of protection of this invention should not be considered as limited to the specific forms stated in the embodiments. The scope of protection of this invention also extends to equivalent technical means that can be conceived by those skilled in the art based on the inventive concept.

Claims

1. A flexible actuation exoskeleton control method based on personalized trajectory planning, characterized in that: Includes the following steps: S1. Divide the gait trajectory into single-leg support phase and double-leg support phase based on pressure sensing, and store the corresponding gait trajectory data; S2. During the single-leg support phase, the DMP algorithm is used to plan the trajectory of the double-leg support phase. During the double-leg support phase, the DMP algorithm and adaptive planning of the gait trajectory of the single-leg trajectory swing phase are used. When the trajectory is in the single-leg support phase, the double-leg support phase trajectory needs to be planned. The current angle is used as the initial angle for DMP trajectory generalization, and the data recorded at the end of S1 is used as the end position. When in the two-legged support phase, plan the single-leg support phase, use the current trajectory as the initial position for the DMP generalized gait trajectory, and use the planned angle as the end angle of the generalized gait trajectory to achieve adaptive adjustment of stride length; S3. Input the generated reference trajectory into the impedance controller as the reference trajectory for the impedance controller to generate the output torque; S4. Based on the hysteresis force modeling, complete the accurate torque modeling, and use a kinematic controller and flexible actuator to complete the exoskeleton-assisted control; specifically including: S4.1: Consider modeling each segmented hysteresis force curve to obtain equation (17). , , These represent the fitting functions for the rising, falling, and transition sub-curves, respectively, and are expressed through the segmentation points. and As the basis for judging the sub-curve, the specific data of the segmentation points are updated in real time by equations (18) and (19); S4.2: The rising and falling sub-curves are fitted by selecting calibration points. The calibration points are connected in sequence to form the fitting curve, resulting in equations (20) and (21), where i, j and p, q represent the calibration points at adjacent positions on the rising and falling segments, respectively. S4.3: Since the transition segment curve has various shapes, directly fitting with the calibration point has obvious errors. Consider using a power function and Maxwell-slip combined model with more similar shapes to fit the transition segment curve. The torque fitting function is expressed as equation (22)-(24). Represents the power function term. It is an exponential term used to control the specific shape of the hysteresis fitting curve; This represents the difference function between the actual torque and the power function term established by the Maxwell-slip model, used to fit the hysteresis characteristics, and consists of multiple basic sliding units with different parameters. Obtained by superposition; S4.4: By using the motion trajectory obtained from the above hysteresis modeling and a dynamic motion control algorithm, precise position control of the motor can be achieved, thereby enabling precise torque output of the flexible actuator.

2. The flexible drive exoskeleton control method based on personalized trajectory planning as described in claim 1, characterized in that: Step S1 specifically includes: When entering walking mode, based on the values ​​from multiple different locations of the plantar pressure sensors on both legs, if all values ​​on the plantar pressure sensors of one leg are below a threshold, it can be determined that the current gait is in a single-leg support state; otherwise, it is in a two-leg support state. When in a single-leg support state, the pressure sensor can be used to determine whether the left or right leg is supporting the gait. When the gait state changes, it is checked whether the switching time is less than a time threshold to avoid abnormal switching due to sensor data loss or other anomalies. If the time is greater than the set time threshold, the current movement state is switched.

3. The flexible drive exoskeleton control method based on personalized trajectory planning as described in claim 1, characterized in that: Step S2 further includes: S2.1: Based on the orientation angle during the current planning phase. and the orientation angle in the current plan The offset angle is calculated. for: The planned hip joint position is obtained based on the adaptive planning of the end position. : Planning the lateral knee joint for: in, For step width, For reference side hip joint angle, For leg length, Leg length at the hip joint; S2.2: The joint trajectory of the data recorded in S1 This is used as the teaching trajectory for the DMP; after modeling the gait trajectory using the DMP, it is solved using the Local Weighted Optimization (LWR) method; the model of the DMP motion primitives is as follows: The role of τ is the time scaling factor of the control system. For target location, For the starting position, and This is the gain coefficient; The nonlinearity forcing f is: in, For Gaussian functions, The weight is x, which comes from a first-order system.

4. The flexible drive exoskeleton control method based on personalized trajectory planning as described in claim 1, characterized in that: Step S3 specifically includes: Based on the reference trajectory of the generating side and Joint angular velocity and q and compensation items As the input for impedance control, the controller is as follows: The planned torque can be obtained by calculating the difference between the actual motion state and the planned motion state, thereby achieving compliant control.

5. The flexible drive exoskeleton control method based on personalized trajectory planning as described in claim 1, characterized in that: The bilateral gait trajectory is divided into a two-leg support phase and a single-leg support phase. Combined with dynamic motion primitive (DMP) adaptive trajectory planning, a more generalized gait trajectory closer to human motion reference is generated. Through impedance control and flexible actuators modeled with hysteresis force, patients can obtain more personalized compliant assistance.

6. A lower limb exoskeleton auxiliary control device, characterized in that, The device includes a memory and one or more processors, wherein the memory stores executable code, and the one or more processors execute the executable code to implement a flexible drive exoskeleton control method based on personalized trajectory planning as described in any one of claims 1-5.

7. A computer-readable storage medium, characterized in that, It stores a program that, when executed by a processor, implements a flexible drive exoskeleton control method based on personalized trajectory planning, as described in any one of claims 1-5.

Images