Active and passive knee joint artificial limb structure and control and reinforcement learning parameter adjusting method

Abstract

The invention relates to the technical field of medical instruments and robots, and discloses an active and passive knee joint artificial limb structure and a control and reinforcement learning parameter adjusting method.The active and passive knee joint artificial limb has various working mode combinations such as complete active, complete passive and active and passive combination and meanwhile has the extremely high exercise ability upper limit and cruising ability upper limit; the invention further discloses an active and passive knee joint artificial limb control method, an active mode or a passive mode is respectively entered in different stages of human body movement gaits, and more movement modes including walking, going up and down stairs, running and the like are met. The invention further discloses a reinforcement learning parameter adjusting method of the active and passive knee joint prosthesis, gait features of the uninjured side are used as a learning target, the motion of the prosthesis joint is matched with the uninjured side as much as possible by minimizing angle errors and phase deviation, and personalized efficient customization and continuous optimization of prosthesis control parameters are achieved.

Description

Technical Field

[0001] The present invention relates to the fields of medical devices and robotics, and in particular to an active and passive knee joint prosthesis structure, control, and reinforcement learning parameter adjustment method. Background Art

[0002] Knee prostheses can help patients with thigh amputations regain basic mobility and reintegrate into society. Based on the driving method, knee prostheses are mainly divided into two types: passive and active. Passive knee prostheses are lightweight and have long battery life, but because they cannot provide active torque, they have difficulty adapting to more complex movements. Active knee prostheses can replace leg muscles to provide driving torque, and theoretically can fully compensate for the function of the missing joint. However, they have problems such as high battery capacity requirements, large size and weight, and short battery life, making them difficult to adapt to daily use.

[0003] In order to achieve a natural walking effect, different control parameters are usually set for different gait phases of the prosthesis (such as swing flexion, swing extension, etc.). However, existing parameter adjustment methods rely heavily on the experience of professional prosthetists. Various movement forms, speeds, and gaits represent a large number of parameter adjustment requirements. The debugging process is time-consuming and labor-intensive, and it is difficult to accurately adapt to the gait characteristics of individual users and the performance deviations between prosthetic systems. Especially at the factory stage, due to the lack of feedback and evaluation of the end user's gait, traditional methods often can only use a universal default parameter configuration. This non-personalized setting may lead to problems such as unstable gait, uncoordinated steps, and excessive impact on the knee joint, which in turn affects the user's adaptation process to the prosthesis and the rehabilitation effect. Summary of the Invention

[0004] The purpose of the present invention is to overcome the deficiencies of the above-mentioned prior art and to provide an active and passive knee joint prosthesis structure, control and reinforcement learning parameter adjustment method.

[0005] The present invention adopts the following technical solutions: An active and passive knee joint prosthesis structure includes a standard receiving cavity and its fixing parts, a lever arm ball head, active and passive hydraulic actuators, a pressure sensor, a rotary valve drive motor box, a knee joint encoder, a shell bracket, a power supply and sensor control system packaging box, a uniaxial force sensor, a connecting tensioner, a standard leg tube, and a standard carbon fiber foot plate.

[0006] The standard receiving cavity and its fixing parts are connected to the lever arm ball head through a quadrangular pyramid. The lever arm ball head is designed with a knee joint axis hole and an upper hinge axis hole of the actuator. The outer shell bracket is designed with a knee joint axis hole and a lower hinge axis hole of the actuator. The three axis holes form a triangle, which is the geometric model of knee joint prosthesis support and drive. The active and passive hydraulic actuators consist of a passive damping hydraulic cylinder, an active motor pump, a pressure sensor, and a rotary valve drive motor box. The upper and lower hinge points are connected to the lever arm ball head and housing bracket via steel shafts. The knee joint magnetic encoder consists of a rotor and a stator. The encoder rotor is threaded into the knee shaft end, while the stator is threaded into the knee shaft hole end of the housing bracket. The power supply and sensor control system enclosure consists of a battery, a control board with an IMU welded on it, and the enclosure is threadedly secured to the outside of the housing bracket (on the same side as the knee joint magnetic encoder). The uniaxial force sensor is threadedly secured to the bottom of the housing bracket and the top of the connecting tensioner. Circular bosses are designed on the bottom and top of the connecting tensioner to align the force sensor and resist lateral forces. A standard leg tube is inserted into the tensioner and tightened via external threads. The bottom is connected to a standard carbon fiber foot plate via a square bezel.

[0007] A control method for active and passive knee prostheses uses a microprocessor to acquire real-time sensor data from the knee encoder, uniaxial force sensor, and IMU (Inertial Measurement Unit). A motion recognition algorithm is used to determine the user's current motion form (including walking, running, climbing stairs, and ascending and descending slopes). Based on open-source data from normal human knee joints, different gait phases are classified and active and passive operating modes are designed. The transition conditions between each state of the finite state machine control algorithm are designed in conjunction with motion logic. Depending on the active and passive operating mode, open-loop damping control (passive mode) or closed-loop impedance control (active mode) is used. Taking walking gait as an example, the walking gait can be subdivided into five phases: stance flexion, stance extension, stance pre-swing, swing flexion, and swing extension, based on the curves of joint angle and ground reaction force. The active work during stance flexion, stance extension, and stance pre-swing phases is ignored in the present invention. That is, the stance flexion and stance extension phases utilize joint locking control (passive mode), the stance pre-swing phase utilizes a gradually opening rotary valve strategy (passive mode), the swing flexion phase utilizes open-loop damping control (passive mode), and the swing extension phase utilizes closed-loop impedance control (active mode). Taking the stair gait as an example, the gait is subdivided into three phases: stance phase, swing flexion phase, and swing extension phase, and closed-loop impedance control (active mode) is employed throughout the entire cycle. In particular, in the event of failure between components of the active mode of the knee prosthesis or insufficient battery power, the knee prosthesis can switch to a completely passive operating mode, sacrificing a certain degree of gait symmetry in exchange for greater safety and longer battery life.

[0008] A reinforcement learning parameter adjustment method for active and passive knee prostheses uses the angular trajectory and motion rhythm of the unaffected knee as reference targets. An adaptive parameter adjustment mechanism is established, using the difference in the motion state of the residual and unaffected knee joints as feedback. This method uses a reinforcement learning algorithm to optimize prosthetic control parameters online, thereby automatically adapting the prosthesis to the individual user's gait characteristics and compensating for performance deviations between prostheses. This method does not rely on precise modeling of the human-prosthesis system. Instead, it uses actual gait data collected, using angular errors and phase deviations as feedback to adjust control parameters and optimize strategies. This method has a two-stage operation characteristic: on the one hand, during the prosthesis factory stage, that is, when the prosthesis is assembled and cooperated with the tester or subject to conduct standardized gait testing, multi-cycle gait data of the healthy side and the disabled side are collected online through the reinforcement learning algorithm, and a gait difference model is quickly established and the learning and convergence of preliminary control parameters are completed, so as to replace the traditional workflow that relies on the experience of the prosthetist for repeated trial adjustments, and significantly improve the personalized adaptation level of the prosthesis when it first leaves the factory; on the other hand, after the user wears the prosthesis and puts it into actual use, he or she can choose to continue to retain the learning ability of the system, and continue to obtain real gait data during his or her normal walking, going up and down stairs, fast walking and slow walking and other daily activities, and continuously iterate and optimize the control parameters through the reinforcement learning strategy, so that the prosthesis can gradually adapt to the user's own movement habits, changes in physical state and changes in scene environment, thereby realizing the long-term evolution and deep personalization of the control strategy.

[0009] The beneficial effects of the present invention are: 1. Hydraulic drive ensures high power density, high burst, strong compliance and fast response of knee joint prosthesis, and can adapt to more forms and higher intensity movement conditions.

[0010] 2. Active and passive knee prostheses combine the advantages of passive and active knee prostheses, and have a variety of working mode combinations such as fully active, fully passive, and active and passive combination. At the same time, they have extremely high upper limits of movement ability and extremely high upper limits of endurance ability.

[0011] 3. Combined with the gait analysis of the normal human body, a finite state machine controller was designed to rationally plan the active and passive working mode switching mechanism and joint control strategy of the knee joint under different movement forms and gaits, fully ensuring the coordination of gait and the comfort and stability of wearing.

[0012] Fourth, an online automatic adjustment method for impedance and damping parameters based on reinforcement learning is introduced, and an adaptive parameter adjustment mechanism based on the healthy side gait is established. This mechanism can adjust the control parameters in real time according to the knee joint angle error and gait phase deviation, and can complete personalized parameter adjustment without human intervention. On the one hand, this method can quickly and efficiently complete the initial parameter configuration at the factory stage of the prosthesis, accurately adapt to the gait characteristics of individual users, and compensate for performance deviations between prosthetic systems, significantly reducing the labor and time costs of parameter adjustment that rely on the prosthetist's experience; on the other hand, it still has the ability to continue learning and self-optimization during the user's subsequent wearing and use, and can adapt to the user's movement habits, physical condition changes, and scene changes in the long term, realizing dynamic adjustment and deep personalization of the prosthesis control strategy. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Figure 1 This is an axonometric diagram of the active and passive knee joint prosthesis structure of the present invention; Figure 2 This is a front view of the active and passive knee joint prosthesis structure of the present invention; Figure 3 The changes of knee joint angle and ground reaction force in a walking gait cycle; Figure 4 This is a schematic diagram of a normal human walking gait cycle; Figure 5 Distribution of active and passive working modes of the knee joint during the walking gait cycle of the present invention; Figure 6 The finite state machine controller of the knee joint prosthesis under walking gait of the present invention; Figure 7 The finite state machine controller of the knee joint prosthesis for stair climbing gait of the present invention; Figure 8 It is the overall control framework of the present invention. DETAILED DESCRIPTION

[0014] The present invention will be further described below with reference to the accompanying drawings and specific embodiments.

[0015] The present invention relates to the structural design of knee joint prosthesis, which is described in detail as follows: As attached Figure 1 As shown, the knee joint prosthesis proposed in the present invention includes a standard receiving cavity and its fixing parts M1, a lever arm ball head M2, active and passive hydraulic actuators M3, a pressure sensor M4, a rotary valve drive motor box M5, a knee joint encoder M6, a power supply and sensor control system packaging box M7, an outer shell bracket M8, a uniaxial force sensor M9, a connecting tensioner M10, a standard leg tube M11, and a standard carbon fiber foot plate M12.

[0016] The standard receiving cavity and its fixing part M1 are connected to the lever arm ball head M2 through a quadrangular pyramid. The lever arm ball head M2 is designed with a knee joint axis hole and an upper hinge axis hole of the actuator. The shell bracket M8 is designed with a knee joint axis hole and a lower hinge axis hole of the actuator. The three axis holes form a triangle, which is the geometric model of knee joint prosthesis support and drive. The active and passive hydraulic actuator M3 consists of a passive damping hydraulic cylinder, an active motor pump, a pressure sensor, and a rotary valve drive motor box. The upper and lower hinge points are connected to the lever arm ball head M2 and the housing bracket M8, respectively, via steel shafts. The knee joint magnetic encoder consists of a rotor and a stator. The encoder rotor is threaded into the end of the knee joint shaft, while the stator is threaded into the end of the knee joint shaft hole in the housing bracket M8. The power supply and sensor control system package M7 consists of a battery, a control board with an IMU welded on it, and a package box. It is threaded onto the outside of the housing bracket (M8), ensuring it is fixed on the same side as the knee joint magnetic encoder M6. The uniaxial force sensor M9 is threaded onto the bottom of the housing bracket M8 and the top of the connecting tensioner M10, respectively. Round shaft bosses are designed on the bottom of the housing bracket and the top of the connecting tensioner to align the force sensor M9 and resist lateral forces. The standard leg tube M11 is inserted into the tensioner M10 and tightened via external threads. The bottom is connected to the standard carbon fiber foot plate M12 via a square prism.

[0017] The present invention relates to a control strategy for a knee joint prosthesis, which is described in detail as follows: Among the many forms of exercise, including walking, running, and climbing stairs, we select and analyze the two most representative and common forms of exercise: walking and climbing stairs. To make the explanation more intuitive, the curves of the changes in knee joint angle and ground reaction force when a normal human body walks are given ( Figure 3 ) and walking gait cycle diagram ( Figure 4 ).

[0018] For walking gait, the active and passive working modes are assigned as shown in the attached Figure 5 As shown, the analysis and design ideas are as follows: According to the attached Figure 3 The relationship curve between the normal human knee joint angle, ground reaction force and gait percentage is shown, which subdivides the walking gait into the following Figure 4 The five phases shown are: stance flexion, stance extension, stance pre-swing, swing flexion, and swing extension.

[0019] Combined with attachment Figure 3 and attached Figure 4It can be seen that in the early stage of the standing phase of normal walking, the knee joint is slightly flexed and then extended and recovered (positive and negative work alternate during this period), the purpose of which is to enhance the buffering of sudden loads and keep the center of gravity height as unchanged as possible when the human body moves forward. In the early stage of the standing pre-swing, there is an auxiliary positive work process of active flexion, the purpose of which is to make the knee joint fully flexed during the swing phase. Apart from this, the rest of the standing phase is a passive damping process. In order to ensure the safety of the user, the knee joint prosthesis described in the present invention ignores the standing flexion and standing extension of the knee joint, that is, during this period, the knee joint is locked. Since the joint speed in the standing pre-swing phase and the swing flexion phase can be compensated by the thigh drive and low damping control, and the subsequent stage is not a standing phase with high stability requirements, the active flexion auxiliary work in the early stage of the standing pre-swing phase can be abandoned, that is, the knee joint is completely in a damping state in the standing phase, and the passive working mode of the actuator is adopted to provide compression damping force. During the swing flexion phase, the knee joint passively flexes under the influence of the thigh, and the knee muscles perform negative work. Therefore, the knee prosthesis also uses the passive working mode of the actuator during the swing flexion phase. However, the minimum resistance of the actuator (primarily related to friction and the initial pressure of the closed hydraulic system) must be sufficiently small to ensure that the knee joint can fully flex during the passive process. During the swing extension phase, the knee joint of the swinging leg extends forward, and there is a brief active extension work process (with low power) to provide the knee joint with sufficient initial extension angular velocity to ensure full leg movement. The rest of the swing phase is a damping process. Considering that the joint extension speed during this phase should match the walking speed, the control requirements are relatively high. If the actuator is used in a passive mode, it is very likely to cause gait asymmetry and the user's sense of insecurity. Therefore, to ensure control accuracy and stability, the knee prosthesis uses the active working mode of the actuator throughout the swing extension phase.

[0020] Generally speaking, when a normal person walks, the knee joint muscles mainly perform negative work, that is, they are mainly in a damping state. The motion control of the knee joint prosthesis designed with this as a reference is also mainly in a passive mode, and only the active working mode is used during the swing and extension phase, which saves a lot of energy while ensuring the balance and stability of the gait.

[0021] For walking gait, the finite state machine controller is as shown in the attached Figure 6 As shown, the analysis and design ideas are as follows: For example, during the early stance phase of a right knee prosthesis, the pressure sensor reading increases continuously from zero to approximately 1.1 times body weight. As the body's center of gravity shifts forward, the pressure sensor reading decreases from its peak value to approximately 0.7 times body weight, marking the entry of the right prosthesis into the pre-swing phase of stance. During the pre-swing phase, the knee is passively flexed, driven by the residual thigh. With the landing of the left foot, the center of gravity continues to shift forward, and the knee passively flexes, the pressure sensor reading decreases until it approaches zero. The 20N threshold is used to mark the entry of the right prosthesis into the swing flexion phase. During the swing flexion phase, the knee continues to passively flex to its maximum angle, driven by the thigh. The -65-degree threshold is used to mark the entry of the right prosthesis into the swing extension phase. During the swing extension phase, the actuator enters active mode, actively extending the knee to its maximum angle. The entry into the stance flexion phase is marked by a knee angle of -4 degrees and a force sensor reading of 20N. For safety reasons, the force sensor reading takes precedence over the knee angle.

[0022] Next, design the specific control strategy for each state in the finite state machine.

[0023] Combined with the above analysis, the knee joint prosthesis of the present invention ignores the standing flexion and extension of the knee joint, and adopts joint locking control in this stage. In the standing pre-swing stage, a strategy of gradually opening the rotary valve is adopted, so that the knee joint can produce a small amount of passive bending driven by the thigh, and the risk of falling caused by the rapid removal of the locking damping can be avoided. In the swing flexion stage, open-loop damping control is adopted, and the expression is

[0024] in, is the target angle of the valve, is the current actual joint angle, in particular, and are the gain coefficient and compensation coefficient respectively, and are the damping parameters to be determined.

[0025] In the swing-stretching stage, closed-loop impedance control is adopted, and the expression is:

[0026] in, is the target torque, is the actual knee joint angle, is the actual knee joint angular velocity. In particular, and are the stiffness coefficient and damping coefficient respectively, is the equilibrium angle, and is the impedance parameter to be determined.

[0027] For stair climbing gait, the analysis and design ideas of active and passive working modes are as follows: During the stance phase of a normal person climbing stairs, the knee joint continuously extends, performing positive work and shifting the body's center of gravity upward. During the swing-flexion phase, because the stairs are often high, the required leg lift height (ground clearance) is high when climbing one or even two steps at a time. Relying solely on the thigh to drive the knee flexion is usually insufficient to climb the stairs. Therefore, during this phase, the knee muscles continue to perform positive work, driving the knee joint to fully flex. During the swing-extension phase, the knee joint continues to extend under the action of gravity until it touches the ground. This period is primarily negative work, but similar to walking motion analysis, the extension speed of the swing leg knee joint should match the climbing speed and the height of the stairs. At the same time, the work power during this phase is relatively low, so to ensure landing accuracy and stability, the actuator's active working mode is also required.

[0028] Overall, the knee prosthesis adopts the active working mode of the actuator during the entire process of climbing stairs.

[0029] For stair climbing gait, the finite state machine controller is as shown in the attached Figure 7 As shown, the analysis and design ideas are as follows: Taking the right knee prosthesis as an example, during the stance phase, to shift the center of gravity upward, the knee joint continuously extends and performs positive work to the maximum joint angle, causing the pressure sensor value to rise. Subsequently, as the left foot steps onto a step, the center of gravity shifts leftward, and the right leg gradually lifts off the table under the traction of the hip joint. The pressure sensor value continues to decrease, and the right prosthesis is judged to have entered the swing flexion phase using 20N as the dividing line. During the swing flexion phase, the knee joint actively flexes to its maximum angle. To ensure sufficient clearance from the table, the right prosthesis is judged to have entered the swing extension phase using a joint angle of -90 degrees. During the swing extension phase, the knee joint angle actively extends until it steps onto the step, and the right prosthesis is judged to have entered the stance phase using an axial force of 20N as the dividing line.

[0030] Next, design the specific control strategy for each state in the finite state machine.

[0031] Since the active working mode is used in the whole cycle of stair climbing, the whole cycle is consistent with the swing-stretch phase of walking gait, and closed-loop impedance control is used. The expression is:

[0032] For other gaits such as running, going up and down hills, and jumping, the above analysis method is used. Combined with open source joint data of normal human bodies, the gait stages are divided, and the active and passive working mode allocation schemes for each stage are determined. Combined with motion logic, the finite state machine controller and the transition conditions between each state are designed. Closed-loop impedance control is used in active mode, and open-loop damping control is used in passive mode.

[0033] The present invention relates to an online adjustment method for impedance and damping parameters based on reinforcement learning, which is described in detail as follows: Stiffness coefficient in the aforementioned closed-loop impedance control , damping coefficient , balance angle and the gain coefficient in open-loop damping control , compensation coefficient Both require optimization and adjustment to achieve the optimal gait. The method proposed in this invention takes the motion characteristics of the healthy knee joint under natural gait as the learning target, constructs an optimization model with the angular error and phase deviation between the motion state of the stump prosthesis and the healthy side as feedback indicators, and uses a reinforcement learning algorithm to achieve automatic optimization and personalized matching of the active and passive knee joint prosthesis control parameters. Specifically, it includes the following steps: 1) Gait feature acquisition: During prosthetic testing, the angle change curves and gait phase information of the healthy and stumped knee joints are collected in real time, and key motion feature parameters of each complete gait cycle are extracted, including but not limited to angle extremes, gait duration, and phase start and end points. 2) State space definition: The angle difference and phase time difference between the injured and healthy sides in each gait cycle are defined as the state vector of the reinforcement learning system; 3) Action space definition: The changes in impedance and damping control parameters used in each gait phase of the active and passive knee prosthesis are used as learning actions. Each action represents a fine-tuning of the current parameter values. 4) Cost function design: Formulate an immediate cost function in quadratic form, including gait deviation and parameter adjustment amplitude, and combine it with the future discounted cost function to form the final cost function; 5) Policy Optimization Algorithm: Using reinforcement learning methods such as policy iteration, the policy model is iteratively updated based on continuous gait cycle feedback data, thereby guiding the parameter adjustment strategy to optimize towards a smaller error. 6) Parameter update and control execution: After each gait cycle, the system adjusts the control parameters of the current gait stage based on the actions output by the learning strategy, and applies the adjusted parameters to the control execution of the next gait cycle, forming a closed-loop control.

[0034] 7) Convergence and termination conditions: When the angle error and phase deviation are both lower than the set tolerance for several consecutive gait cycles (for example, the angle error is less than 2 degrees and the phase deviation is less than 3% of the gait cycle), the system determines that the parameter adjustment process has converged, saves the current parameter configuration as the user's personalized initial parameters, and allows the user to choose to terminate the parameter adjustment or enter the monitoring and maintenance state.

[0035] In summary, the overall control framework of the present invention is as shown in the attached Figure 8 shown.

Claims

1. An active and passive knee joint prosthesis structure, characterized in that: include: The standard receiving cavity and its fixing part (M1) are connected to the lever arm ball head (M2) through a quadrangular pyramid. The lever arm ball head (M2) is designed with a knee joint axis hole and an actuator upper hinge point axis hole; The housing bracket (M8) is provided with a knee joint axis hole and an actuator lower hinge axis hole, and the upper and lower hinge axis holes of the lever arm ball head (M2) are connected by a steel shaft; A power supply and sensor control system packaging box (M7) is provided on the outside of the housing bracket (M8); The knee joint magnetic encoder (M6) consists of a rotor and a stator. The rotor is embedded in the knee joint shaft end through a threaded connection, and the stator is fixed to the knee joint shaft hole end of the housing bracket (M8) through a threaded connection. A single-axis force sensor (M9) is fixed at both ends to the bottom of the housing bracket (M8) and the top of the connecting tensioner (M10) through threaded connections; The standard leg tube (M11) is inserted into the tensioning piece (M10) and tightened through the external thread. The bottom is connected to the standard carbon fiber foot plate (M12) through a quadrangular pyramid.

2. The active and passive knee joint prosthesis structure according to claim 1, characterized in that: The invention also includes an active and passive hydraulic actuator (M3), which includes a passive damping hydraulic cylinder, an active motor pump, a pressure sensor (M4) and a rotary valve driving motor box (M5).

3. The active and passive knee joint prosthesis structure according to claim 1, characterized in that: The power supply and sensor control system packaging box (M7) comprises a battery, a control board welded with an IMU inertial measurement unit, and a packaging box. The power supply and sensor control system packaging box (M7) is fixed to the outside of the housing bracket (M8) through a threaded connection.

4. The active and passive knee joint prosthesis structure according to claim 1, characterized in that: The power supply and sensor control system packaging box (M7) is arranged on the same side as the knee joint magnetic encoder (M6).

5. The active and passive knee joint prosthesis structure according to claim 1, characterized in that: The bottom of the housing bracket (M8) and the top of the connecting tensioner (M10) are respectively designed with circular shaft bosses for centering the force sensor (M9) and resisting lateral forces.

6. A control method for active and passive knee joint prosthesis, characterized in that: The active and passive knee joint prosthesis structure according to any one of claims 1 to 5 comprises the following steps: Based on open-source data of normal human knee joints, different stages of various gaits are divided and active and passive working modes are designed. In combination with motion logic, the transition conditions between the various states of the finite state machine control algorithm are designed. Depending on the active and passive working modes, open-loop damping control or closed-loop impedance control is adopted. The open-loop damping control mode is passive mode; The closed-loop impedance control mode is active mode.

7. The control method of active and passive knee joint prosthesis according to claim 6, characterized in that: According to the curves of joint angles and ground reaction forces, walking gait can be divided into five phases: stance flexion phase, stance extension phase, stance pre-swing phase, swing flexion phase, and swing extension phase. The passive mode corresponds to the movements of standing flexion, standing extension, standing pre-swing phase and swing flexion phase; The active mode includes the swing-extension phase; Joint locking control was used during standing flexion and standing extension phases; During the pre-swing period of stance, a strategy of gradually opening the rotary valve was adopted; Open-loop damping control is used during the swing flexion phase; Closed-loop impedance control was used during the swing-extension phase.

8. The control method of active and passive knee joint prosthesis according to claim 7, characterized in that: In the event of failure between the active mode components of the knee prosthesis or low battery, the knee prosthesis can switch to a completely passive working mode, sacrificing a certain degree of gait symmetry in exchange for higher safety and longer battery life.

9. A reinforcement learning parameter adjustment method for active and passive knee prostheses, used to implement the active and passive knee prostheses structure according to any one of claims 1 to 5, characterized in that: Taking the angle change trajectory and movement rhythm of the healthy knee joint as reference targets, an adaptive parameter adjustment mechanism is established with the difference between the movement state of the disabled side and the healthy side as feedback. The prosthetic control parameters are optimized online through the reinforcement learning algorithm to achieve automatic adaptation of the prosthesis to the individual user's gait characteristics and compensation for performance deviations between prostheses.

10. The reinforcement learning parameter adjustment method for active and passive knee joint prostheses according to claim 9, characterized in that: It adopts a two-stage operation feature, carrying out reinforcement learning and parameter adjustment in the factory stage and the subsequent use stage respectively, to achieve long-term evolution and deep personalization of the control strategy.

Images