Joint angle vector generation method and gait control method

Abstract

The invention relates to the technical field of rehabilitation medical robots, and particularly provides a joint angle vector generation method and a gait control method.The generation method comprises the steps that a first gait phase and multi-sensor data at the current moment are obtained; the first gait phase is the gait phase of the lower limb exoskeleton robot at the previous moment, and the multi-sensor data comprises multi-modal motion data and visual data; predicting a second gait phase according to the first gait phase and the multi-modal motion data; the second gait phase is the gait phase of the lower limb exoskeleton robot at the current moment; adjusting a reference joint angle vector and a time scaling factor of a self-adaptive joint angle vector generator of which the weight is optimized in advance according to the visual data, and then generating an expected joint angle vector according to the second gait phase by utilizing the adjusted self-adaptive joint angle vector generator; according to the method, the joint angle vector which adapts to the current environment and is high in gait bionic performance can be generated.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of rehabilitation medical robots, in particular to a joint angle vector generation method and a gait control method. BACKGROUND

[0002] With the acceleration of the global population aging process and the continuous increase in the incidence of nervous system diseases such as stroke and spinal cord injury, the number of patients with lower limb motor dysfunction caused by nervous system damage has increased significantly. It is the core demand for such patients to recover the ability to walk independently to realize self-care and return to society. Traditional rehabilitation training highly depends on manual assistance by therapists, and has inherent bottlenecks such as difficulty in standardizing treatment intensity, high labor cost, limited training efficiency, and difficulty in quantitatively evaluating rehabilitation effect.

[0003] As a product of the deep integration of modern rehabilitation medicine and robot technology, lower limb exoskeleton robots provide support and assistance to patients through mechanical structures to assist in completing standardized walking training, thereby providing an effective technical solution to the above problems. However, the current technology still faces many challenges in actual application: first, existing systems rely on single sensor data for simple threshold judgment and cannot achieve accurate phase estimation; second, existing systems use fixed parameter control models and lack the ability to learn and adapt to the gait characteristics of healthy people; finally, existing systems cannot dynamically adjust the joint angle vector according to the surrounding environment.

[0004] If the above problems are not solved, the lower limb exoskeleton robot will be limited to structured indoor environments and cannot meet the rehabilitation needs of patients in real-life scenarios; the lack of environmental perception makes the device respond slowly to terrain changes, increasing the risk of abnormal gait, and the lack of gait bionics may induce non-physiological movement patterns, which is not conducive to the reestablishment of damaged neural pathways, ultimately limiting the effectiveness of rehabilitation training and narrowing the scope of application.

[0005] At present, there is no effective technical solution to the above problems. SUMMARY

[0006] The purpose of the present application is to provide a joint angle vector generation method and a gait control method, which can generate a joint angle vector that is adaptive to the current environment and has high gait bionics.

[0007] In a first aspect, the present application provides a joint angle vector generation method applied in a lower limb exoskeleton robot, which includes the following steps: A1, obtaining a first gait phase and multi-sensor data at a current time; the first gait phase is the gait phase of the lower limb exoskeleton robot at the previous time, and the multi-sensor data includes multi-modal motion data and visual data; A2, predicting a second gait phase according to the first gait phase and the multi-modal motion data; the second gait phase is a gait phase of the lower extremity exoskeleton robot at the current time; A3, adjusting a reference joint angle vector and a time scaling factor of the adaptive joint angle vector generator with pre-optimized weights according to the visual data, and then generating an expected joint angle vector according to the second gait phase by using the adjusted adaptive joint angle vector generator; The adaptive joint angle vector generator adopts a dynamic motion primitive architecture, and the pre-optimization weight process of the adaptive joint angle vector generator includes: S1, collecting demonstration gait data of healthy people in different environments; S2, generating an optimal weight set of the adaptive joint angle vector generator according to all the demonstration gait data, the optimal weight set including optimal weights corresponding to different lower extremity joints.

[0008] The joint angle vector generation method provided by the application has the advantages that the adaptive joint angle vector generator has the deep learning and adaptive adjustment capabilities for the gait characteristics of healthy people by pre-generating the optimal weight set of the adaptive joint angle vector generator by using the demonstration gait data of healthy people in different environments, the phase prediction based on multi-sensor data is realized by predicting the second gait phase according to the first gait data and the multi-modal motion data, so as to effectively improve the accuracy and reliability of the phase estimation, and the expected joint angle vector is dynamically adjusted according to the surrounding environment by first adjusting the reference joint speed vector and the time scaling factor of the adaptive joint angle vector generator according to the visual data, and then generating the expected joint angle according to the second gait phase by using the adjusted adaptive joint angle vector generator, so that the application provides a joint angle vector that can adapt to the current environment and has high gait bionics, thereby effectively meeting the rehabilitation needs of patients in real-life scenarios, avoiding the increase in abnormal gait due to the lack of environmental perception capability, and inducing the problems of non-physiological motion mode and the re-establishment of damaged neural pathways, and thus effectively improving the rehabilitation training effect and expanding the application range of rehabilitation training.

[0009] Optionally, the step A2 includes: A21, generating a first initial gait phase according to the first gait phase by using a state transition equation of extended Kalman filtering; the first initial gait phase is a predicted value of the gait phase at the current time generated according to the first gait phase; A22, generating a second initial gait phase according to the multi-modal motion data by using an observation equation of extended Kalman filtering; the second initial gait phase is an observed value of the gait phase at the current time generated according to the multi-modal motion data. A23, correcting the first initial gait phase according to the second initial gait phase by using an extended Kalman filter to obtain the second gait phase.

[0010] Optionally, the multi-modal motion data comprises a plurality of motion parameters, and the step A22 comprises: A221, for each motion parameter, generating a third initial gait phase corresponding to the motion parameter according to the motion parameter by using an observation equation of an extended Kalman filter; A222, obtaining the second initial gait phase according to all the third initial gait phases.

[0011] Optionally, the motion parameter is a heel pressure or a knee joint angle, and a calculation formula of the third initial gait phase corresponding to the heel pressure is as follows: ; wherein, represents the heel pressure, represents a preset coefficient corresponding to the heel pressure, represents the third initial gait phase corresponding to the heel pressure; a calculation formula of the third initial gait phase corresponding to the knee joint angle is as follows: ; wherein, represents the knee joint angle, represents a preset coefficient corresponding to the knee joint angle, represents the third initial gait phase corresponding to the knee joint angle, represents a preset phase offset corresponding to the knee joint angle.

[0012] Optionally, a calculation formula of the desired joint angle vector is as follows: ; wherein, t represents time, represents a time scaling factor, and z(t) represents an auxiliary vector, represents a derivative of the auxiliary vector, and both represent gain constants, represents the desired joint angle vector, represents a derivative of the desired joint angle vector, and g represents a reference joint angle vector, represents the second gait phase, represents a nonlinear forcing term corresponding to the second gait phase. A calculation formula of the nonlinear forcing term is as follows: ; wherein, denotes a Gaussian radial basis function, N denotes the number of Gaussian radial basis functions, denotes the width of the i-th Gaussian radial basis function, denotes the center of the i-th Gaussian radial basis function, denotes the weight corresponding to the i-th Gaussian radial basis function.

[0013] Optionally, step S2 comprises: S21, generating an optimal weight set of the adaptive joint angle vector generator according to all the teaching gait data by using a locally weighted regression algorithm.

[0014] In a second aspect, the present application also provides a gait control method applied in a lower extremity exoskeleton robot, comprising the following steps: B1, acquiring a first gait phase and multi-sensor data at a current time; the first gait phase is a gait phase of the lower extremity exoskeleton robot at a previous time, and the multi-sensor data includes multi-modal motion data and visual data; B2, predicting a second gait phase according to the first gait phase and the multi-modal motion data; the second gait phase is a gait phase of the lower extremity exoskeleton robot at the current time; B3, adjusting a reference joint angle vector and a time scaling factor of the adaptive joint angle vector generator with pre-optimized weights according to the visual data, and then generating an expected joint angle vector according to the second gait phase by using the adjusted adaptive joint angle vector generator; B4, determining a current gait phase according to the second gait phase and the multi-modal motion data; B5, acquiring an expected joint torque according to the expected joint angle vector; B6, adjusting the stiffness and damping of the lower extremity exoskeleton robot according to the current gait phase while keeping the expected joint torque unchanged; The adaptive joint angle vector generator adopts a dynamic motion primitive architecture, and the pre-optimization weight process of the adaptive joint angle vector generator comprises: S1, collecting teaching gait data of healthy people in different environments; S2, generating an optimal weight set of the adaptive joint angle vector generator according to all the teaching gait data, the optimal weight set including optimal weights corresponding to different lower extremity joints.

[0015] The gait control method provided by the application has the adaptive joint angle vector generator pre-generated by using the teaching gait data of healthy people in different environments, so that the adaptive joint angle vector generator has the deep learning and adaptive adjustment capability for the gait characteristics of healthy people, the phase prediction based on multi-sensor data is realized by predicting the second gait phase according to the first gait data and the multi-modal motion data, the accuracy and reliability of phase estimation are effectively improved, the reference joint speed vector and the time scaling factor of the adaptive joint angle vector generator are adjusted according to the visual data, the parameters of the adaptive joint angle vector generator are adjusted according to the surrounding environment, the adjusted adaptive joint angle vector generator is used to generate the expected joint angle according to the second gait phase, and the expected joint angle vector is dynamically adjusted according to the surrounding environment. Therefore, the application provides a joint angle vector that can adapt to the current environment and has high gait bionics, thereby effectively meeting the rehabilitation needs of patients in real-life scenarios and avoiding the problems of increased abnormal gait due to lack of environmental perception ability, induced non-physiological motion mode, and affected re-establishment of damaged neural pathways. Therefore, the rehabilitation training effect is effectively improved and the application range of rehabilitation training is expanded.

[0016] Optionally, step B6 comprises: B61, acquiring a human-computer interaction torque; B62, adjusting the stiffness and damping of the lower limb exoskeleton robot according to the current gait phase and the human-computer interaction torque while keeping the expected joint torque unchanged.

[0017] Optionally, the calculation formula of the expected joint torque is as follows: ; Wherein, represents the expected joint torque, J represents the Jacobian matrix of the lower limb exoskeleton robot, represents the transpose of the Jacobian matrix of the lower limb exoskeleton robot, represents a preset virtual stiffness, represents a preset virtual damping, represents an expected joint angle vector, q(t) represents an actual joint angle vector, represents the derivative of the expected joint angle vector, represents the derivative of the actual joint angle vector, represents a preset gravity compensation term.

[0018] Optionally, step B4 comprises: B41, taking the gait phase corresponding to the preset phase mapping interval to which the second gait phase belongs and the preset condition satisfied by the multi-modal motion data as the current gait phase.

[0019] As can be seen from the above, the joint angle vector generation method and the gait control method provided by the application make the adaptive joint angle vector generator have the deep learning and adaptive adjustment capability for the gait characteristics of healthy people by using the teaching gait data of healthy people in different environments to pre-generate the optimal weight combination of the adaptive joint angle vector generator, realize phase prediction based on multi-sensor data by predicting the second gait phase according to the first gait data and the multi-modal motion data, effectively improve the accuracy and reliability of phase estimation, and adjust the parameters of the adaptive joint angle vector generator according to the surrounding environment by first adjusting the reference joint velocity vector and the time scaling factor of the adaptive joint angle vector generator according to the visual data, and then generating the expected joint angle according to the second gait phase by using the adjusted adaptive joint angle vector generator to dynamically adjust the expected joint angle vector according to the surrounding environment. Therefore, the application provides a joint angle vector that can adapt to the current environment and has high gait bionics, thereby effectively meeting the rehabilitation needs of patients in real-life scenarios and avoiding the problems of increased abnormal gait due to lack of environmental perception ability, induced non-physiological motion mode, and affected reestablishment of damaged neural pathways, thereby effectively improving the rehabilitation training effect and expanding the application range of rehabilitation training. BRIEF DESCRIPTION OF DRAWINGS

[0020] Figure 1 A flowchart of a joint angle vector generation method provided by an embodiment of the application.

[0021] Figure 2 A flowchart of a gait control method provided by an embodiment of the application. DETAILED DESCRIPTION

[0022] The technical solutions in the embodiments of the application will be described clearly and completely below with reference to the accompanying drawings in the embodiments of the application. Obviously, the described embodiments are only part of the embodiments of the application, rather than all the embodiments of the application. The components of the embodiments of the application described and shown in the accompanying drawings herein can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but only represents selected embodiments of the application. Based on the embodiments of the application, all other embodiments obtained by those skilled in the art without creative work fall within the scope of the application.

[0023] It should be noted that similar reference numerals and letters refer to like items throughout the accompanying drawings, and once an item is defined in one drawing, it is not necessary to further define and explain it in subsequent drawings. Meanwhile, in the description of the present application, the terms "first", "second" and the like are only used to distinguish descriptions, and cannot be understood as indicating or implying relative importance.

[0024] In a conventional existing lower limb rehabilitation exoskeleton robot system, the lack of environmental perception capability causes the device to be unable to adapt to complex environmental terrain, which is specifically manifested in the lack of a multi-sensor fusion perception framework, the reliance on joint encoders for joint angle vector tracking only, and the inability to accurately estimate gait phase and identify environmental features in real time. At the same time, the lack of gait bionics is manifested in that the control strategy is based on a predefined fixed joint angle curve, lacks dynamic adaptability modeling, and causes the generated gait joint angle vector to be rigid and not in line with the principles of biomechanics, affecting the continuity and naturalness of the training process. Among them, the lack of environmental perception makes it difficult for the device to distinguish between flat ground and obstacles, and the lack of gait bionics causes the joint movement to lack the flexibility of healthy human gait, thereby causing the training scene to be disconnected from the actual life scene and limiting the effectiveness of rehabilitation training.

[0025] For example, in a home rehabilitation scene, when a patient uses a lower limb exoskeleton robot for walking training and encounters a carpet edge or a slight slope, the device cannot recognize the change in terrain through plantar pressure sensors or visual data because it only works based on joint encoder data, resulting in gait phase prediction deviation. Further, the predefined joint angle vector is not dynamically adjusted according to environmental information, and the exoskeleton continues to execute the fixed motion mode of flat ground, causing the gait to be interrupted or the center of gravity to be unstable when the patient crosses the obstacle, and the training process is forced to stop. As a result, the device cannot maintain continuous walking training in unstructured environments, the patient cannot obtain a rehabilitation experience that matches daily activities, and the achievement of training goals is directly affected.

[0026] If the above problems are not solved, the lower limb exoskeleton robot will be limited to structured indoor environments, and cannot meet the rehabilitation needs of patients in real-life scenarios. The lack of environmental perception causes the device to respond slowly to terrain changes, increasing the risk of gait abnormalities, while the lack of gait bionics may induce non-physiological movement patterns, which is not conducive to the reestablishment of damaged neural pathways, ultimately resulting in limited rehabilitation training effectiveness and narrow application range.

[0027] To this end, in a first aspect, the present application provides a joint angle vector generation method applied in a lower limb exoskeleton robot, which includes the following steps: A1, obtaining a first gait phase and multi-sensor data at a current time; the first gait phase is the gait phase of the lower limb exoskeleton robot at the previous time, and the multi-sensor data includes multi-modal motion data and visual data; A2, predicting a second gait phase according to the first gait phase and the multi-modal motion data; the second gait phase is a gait phase of the lower extremity exoskeleton robot at the current time; A3, adjusting the reference joint angle vector and the time scaling factor of the adaptive joint angle vector generator with pre-optimized weights according to the visual data, and then generating the expected joint angle vector according to the second gait phase by using the adjusted adaptive joint angle vector generator; The adaptive joint angle vector generator adopts a dynamic motion primitive architecture, and the pre-optimization weight process of the adaptive joint angle vector generator includes: S1, collecting demonstration gait data of healthy people in different environments; S2, generating an optimal weight set of the adaptive joint angle vector generator according to all the demonstration gait data, and the optimal weight set includes optimal weights corresponding to different lower extremity joints.

[0028] For ease of understanding, some key terms designed in the present application are explained as follows. The first gait phase of the embodiment refers to the position of the lower extremity exoskeleton robot in the gait cycle at the previous time, which can be expressed in percentage from 0 to 100%, where 0% represents the beginning of a gait cycle (e.g. heel strike), and 100% represents the end of a gait cycle, which is equivalent to using the first gait phase to provide historical reference for predicting the gait at the current time. The multi-sensor data of the embodiment refers to the comprehensive information obtained from multiple sensors at the current time, which includes multi-modal motion data and visual data. The multi-modal motion data is usually derived from inertial measurement units (IMU) or plantar pressure sensors, etc. to capture the motion state and force condition of the lower extremity. The visual data is derived from visual sensors such as cameras to perceive environmental features and terrain information. The embodiment provides more comprehensive and accurate environmental perception capability by obtaining the multi-sensor data at the current time. The second gait phase of the embodiment refers to the position of the lower extremity exoskeleton robot in the gait cycle at the current time, which is predicted according to the gait phase at the previous time and the current multi-modal motion data. The second gait phase is a key input for generating the desired joint angle vector. The adaptive joint angle vector generator of the embodiment refers to a controller based on the dynamic motion primitive (DMP) architecture, whose core function is to generate a desired joint angle vector that meets the biomechanical characteristics according to the input gait phase. The generator has adaptive ability to adjust its output according to environmental and task requirements. The reference joint angle vector of the embodiment refers to the benchmark joint angle vector that the adaptive joint angle vector generator relies on when generating the desired joint angle vector. The vector usually represents an ideal and healthy gait pattern, and can be adjusted according to the visual data to adapt to the current environment of the lower extremity exoskeleton robot. The time scaling factor of the embodiment refers to a parameter used to adjust the duration of the gait cycle. The embodiment can change the speed of the joint angle vector by adjusting the time scaling factor, so as to adapt to different walking speeds or environmental requirements. The optimal weight set of the embodiment refers to a set of parameters in the adaptive joint angle vector generator for adjusting the behavior of dynamic motion primitive. These weights are obtained by optimizing and learning the demonstration gait data of healthy people in different environments, to ensure that the generated joint angle vector has good bionics and adaptability. The optimal weight set includes optimal weights corresponding to different lower extremity joints (such as hip joint, knee joint), so that the motion of each joint can be finely controlled.

[0029] The present application provides a joint angle vector generation method applied to a lower extremity exoskeleton robot, aiming to solve the problems of lack of environmental perception ability and insufficient gait bionics of traditional exoskeleton robots when generating gait in complex environments.

[0030] Firstly, in terms of acquiring the first gait phase and multi-sensor data at the current time, various methods can be adopted. For example, the method of acquiring the first gait phase is the same as that of acquiring the second gait phase, that is, this embodiment takes the second gait phase as the first gait phase at the next time. The multi-sensor data can include information collected by a plantar pressure sensor, an inertial measurement unit (IMU), and a visual sensor (such as a depth camera or an RGB camera). Specifically, the plantar pressure sensor can provide pressure distribution data of the foot in contact with the ground, the IMU can provide attitude, angular velocity, and acceleration data of the exoskeleton joint, and the visual sensor can capture images or depth information of the environment. These data can be independently acquired or collected through a data bus or wireless communication to a central processor for unified processing.

[0031] Secondly, in terms of predicting the second gait phase according to the first gait phase and multi-modal motion data, different prediction models can be used. For example, a simple linear regression model can be used to predict the gait phase at the next time according to the historical gait phase and the change trend of the current multi-modal motion data. Another way is to use a rule-based finite state machine to switch the gait phase according to the preset gait events (such as heel strike and toe-off) and multi-modal motion data, and thereby infer the second gait phase. A machine learning-based method can also be used to learn the complex nonlinear relationship between the gait phase and the multi-modal motion data through model training, thereby making predictions.

[0032] Subsequently, the reference joint angle vector and the time scaling factor of the adaptive joint angle vector generator with pre-optimized weights are adjusted according to the visual data, and then the desired joint angle vector is generated according to the second gait phase by using the adjusted adaptive joint angle vector generator. There can be various implementations in terms of how the visual data is used to adjust the reference joint angle vector and the time scaling factor, and how the desired joint angle vector is generated according to the second gait phase. For example, the visual data can be used to identify the type of the current environment, such as flat ground, uphill, downhill, or stairs, and according to the identified type of the environment, a suitable reference joint angle vector can be selected from a pre-set reference joint angle vector library, and at the same time, the time scaling factor can also be adjusted according to the type of the environment, for example, when going uphill, the gait speed can need to be slowed down, thereby reducing the time scaling factor. The adaptive joint angle vector generator of this embodiment adopts a dynamic motion primitive (DMP) architecture, and the pre-optimization of the weights includes: first, collecting demonstration gait data of healthy people in different environments, these demonstration data can include walking data of healthy people in various terrains such as flat ground, uphill, downhill, and stairs; then, generating an optimal weight set of the adaptive joint angle vector generator according to all the demonstration gait data, the optimal weight set includes optimal weights corresponding to different lower limb joints, for example, the weights corresponding to the hip joint and the knee joint can be optimized respectively. When generating the desired joint angle vector, the adaptive joint angle vector generator will combine the adjusted reference joint angle vector, the time scaling factor, and the currently predicted second gait phase, and output a smooth and environment-adaptive desired joint angle vector through the dynamic motion primitive architecture.

[0033] The above technical solutions are further described in more detail through a more specific example: assume that user A is a patient with lower extremity motor dysfunction and is using a lower extremity exoskeleton robot for rehabilitation training. The exoskeleton robot needs to assist user A to transition from flat ground to a slightly uphill road section in an indoor environment. First, when user A walks on flat ground, the lower extremity exoskeleton robot continuously acquires the first gait phase and multi-sensor data at the current time, for example, at a certain time, the exoskeleton robot records that the first gait phase at the last time is 50% (i.e. the midpoint of the gait cycle), and at the same time, acquires multi-modal motion data and visual data at the current time through the foot pressure sensor, inertial measurement unit (IMU) and visual sensor. The multi-modal motion data may include heel strike pressure, knee joint angle, hip joint angular velocity, etc., while the visual data may capture image information of the gradually rising ground in front. Then, according to the acquired first gait phase and multi-modal motion data, the lower extremity exoskeleton robot predicts the second gait phase, for example, using a pre-trained gait phase prediction model, combining the 50% gait phase at the last time and the multi-modal motion data such as foot pressure, knee joint angle, etc. at the current time, to predict the second gait phase at the current time as 52%, this prediction process considers the continuity of gait and the dynamics of motion. Subsequently, the exoskeleton robot adjusts the parameters of the adaptive joint angle vector generator using visual data. When the visual sensor detects that there is an uphill road section in front, the system will adjust the reference joint angle vector of the adaptive joint angle vector generator to 0.9 times of the original reference joint angle vector and increase the time scaling factor of the adaptive joint angle vector generator to 1.1 times of the original time scaling factor according to this visual information. The adaptive joint angle vector generator has generated an optimal weight set containing different lower limb joints such as hip and knee joints by collecting demonstration gait data of healthy people in different environments such as flat ground and uphill during pre-optimization of weights. Using these adjusted reference joint angle vectors and time scaling factors, combined with the currently predicted second gait phase (52%), the adaptive joint angle vector generator will generate a series of expected joint angle vectors that can guide the hip and knee joints of the exoskeleton robot to move in a smooth, natural and adaptive way to the uphill terrain. Thus, the exoskeleton robot can drive its joints to move accordingly according to the generated expected joint angle vectors, thereby assisting user A to smoothly and naturally transition from flat ground to uphill road section. The whole process realizes dynamic adaptation from environment perception to joint angle vector generation, solving the problem of gait stiffness and unnaturalness of traditional lower extremity exoskeleton robots in complex environments.

[0034] The present application significantly improves the environmental perception ability and gait bionics of lower limb exoskeleton robots in complex environments by fusing multi-sensor data and dynamic motion primitive architecture. Traditional exoskeleton robots usually rely on pre-programmed fixed gait patterns. When facing complex terrains such as uphill and downhill, they cannot adjust the gait in real time, resulting in stiff and unnatural gait, and even causing the patient to fall. For example, in the scenario where user A transitions from flat ground to uphill, the existing fixed gait pattern may not be able to adapt to the changes in joint angles and speeds required for uphill, making it difficult for user A to walk. The present application acquires multi-sensor data, especially introduces visual data, so that the exoskeleton robot can perceive environmental changes in real time, such as recognizing the uphill road ahead. This is in sharp contrast to the prior art, which lacks an effective multi-sensor fusion framework and relies only on joint encoders for simple angle tracking. By combining the first gait phase and multi-modal motion data to predict the second gait phase, the present application can more accurately grasp the position in the current gait cycle, providing accurate input for subsequent gait generation. Further, the present application uses visual data to dynamically adjust the reference joint angle vector and time scaling factor of the adaptive joint angle vector generator. This dynamic adjustment mechanism enables the exoskeleton robot to generate adaptive gait according to the actual environment (such as uphill) rather than a rigid preset pattern. The adaptive joint angle vector generator uses a dynamic motion primitive (DMP) architecture and optimizes the weights by collecting demonstration gait data of healthy people in different environments, ensuring that the generated desired joint angle vector has high bionics and naturalness. This is in contrast to the prior art, which simply averages or fits healthy person gait data and lacks modeling of gait dynamic adaptability, and can better activate the patient's damaged neural pathways and avoid the generation of abnormal gait patterns. Therefore, the technical solution of the present application can provide patients with a safer, more efficient and personalized rehabilitation training experience, significantly improving the practicality and rehabilitation effect of lower limb exoskeleton robots.

[0035] It can be seen that the joint angle vector method provided by the application has the advantages that the adaptive joint angle vector generator is provided with the deep learning and adaptive adjustment capability for the gait characteristics of healthy people by using the teaching gait data of healthy people in different environments to pre-generate the optimal weight combination of the adaptive joint angle vector generator, the phase prediction based on multi-sensor data is realized by predicting the second gait phase according to the first gait data and the multi-modal motion data, the accuracy and reliability of phase estimation are effectively improved, the parameters of the adaptive joint angle vector generator are adjusted according to the surrounding environment by first adjusting the reference joint velocity vector and the time scaling factor of the adaptive joint angle vector generator according to the visual data, the adjusted adaptive joint angle vector generator is used to generate the expected joint angle according to the second gait phase, the expected joint angle vector is dynamically adjusted according to the surrounding environment, and therefore the application provides a joint angle vector that can generate an adaptive current environment and has high gait bionics, thereby effectively meeting the rehabilitation needs of patients in real-life scenarios, avoiding the increase in abnormal gait due to the lack of environmental perception capability, inducing non-physiological motion patterns, and affecting the reestablishment of damaged neural pathways, and effectively improving the rehabilitation training effect and expanding the application range of rehabilitation training.

[0036] In some preferred embodiments, step A2 comprises: A21, generating a first initial gait phase according to the first gait phase by using the state transition equation of the extended Kalman filter; the first initial gait phase is a predicted value of the gait phase at the current time generated according to the first gait phase; A22, generating a second initial gait phase according to the multi-modal motion data by using the observation equation of the extended Kalman filter; the second initial gait phase is an observed value of the gait phase at the current time generated according to the multi-modal motion data phase; A23, correcting the first initial gait phase according to the second initial gait phase by using the extended Kalman filter to obtain the second gait phase.

[0037] The extended Kalman filter of this embodiment is a state estimation algorithm suitable for nonlinear systems, which estimates the system state through an iterative prediction and update process. In step A21, first, the state transition equation of the extended Kalman filter is used to predict the gait state at the current time according to the first gait phase of the lower extremity exoskeleton robot at the last time, combined with the dynamic model of the system, thereby generating a first initial gait phase, which represents the priori estimation based on the motion law of the system itself. Subsequently, in step A22, the observation equation of the extended Kalman filter is used to actually observe the gait state at the current time according to the multi-modal motion data in the multi-sensor data obtained at the current time, thereby generating a second initial gait phase, which reflects the observation result based on the actual measurement data. Finally, in step A23, the extended Kalman filter fuses and corrects the first initial gait phase (predicted value) and the second initial gait phase (observed value), specifically, step A23 calculates the Kalman gain to weight and average the predicted value and the observed value, thereby obtaining an optimal gait phase estimation (i.e., the second gait phase) that is free of noise influence. It should be understood that the extended Kalman filter used in the present application belongs to the prior art, and its working principle will not be discussed in detail here.

[0038] The embodiment effectively solves the problems of noise interference and model uncertainty in traditional gait phase prediction by introducing an extended Kalman filter. Specifically, the extended Kalman filter first uses a state transition equation to predict the gait phase at the current time based on the first gait phase at the previous time, generating a first initial gait phase, which provides a priori estimate based on the system dynamic model. Subsequently, through an observation equation, the gait phase is actually observed in combination with the multi-modal motion data at the current time, generating a second initial gait phase. Finally, the extended Kalman filter performs weighted fusion of the predicted value (first initial gait phase) and the observed value (second initial gait phase) to correct the prediction result. This iterative process of prediction and correction enables the system to adjust the gait phase estimate in real time based on the latest sensor data, thereby obtaining a high-precision gait phase prediction result even in the presence of noise and uncertainty. Thus, even in complex motion environments or poor sensor data quality, accurate identification and tracking of gait phase can be ensured. Through the above technical solution, the present application uses the extended Kalman filter to predict and correct the gait phase, which can significantly improve the accuracy and robustness of gait phase estimation. Compared with simple model-based or purely sensor data-based prediction methods, the extended Kalman filter can effectively suppress the influence of sensor noise and optimally estimate the system state, thereby ensuring that the lower extremity exoskeleton robot can more accurately perceive and track the user's gait cycle. This not only improves the real-time performance and accuracy of the exoskeleton robot gait assistance, but also enhances the stability and safety of the system, providing a more natural and comfortable wearing experience for the user, thereby effectively avoiding problems such as assistance lag or excessive assistance caused by inaccurate gait phase prediction.

[0039] In some preferred embodiments, the multi-modal motion data includes a plurality of motion parameters, and step A22 includes: A221, for each motion parameter, generating a third initial gait phase corresponding to the motion parameter according to the motion parameter using the observation equation of the extended Kalman filter; A222, obtaining the second initial gait phase according to all third initial gait phases.

[0040] The multi-modal motion data of this embodiment refers to a set of motion parameters collected by different types of sensors on the lower extremity exoskeleton robot, which can include but are not limited to heel pressure and toe pressure collected by plantar pressure sensors, angular velocity, acceleration, etc. provided by inertial measurement units (IMUs), knee joint angle, hip joint angle measured by joint encoders, and muscle activity signals measured by electromyography sensors (EMGs), etc. First, for each motion parameter, the system generates a third initial gait phase corresponding to the motion parameter according to the motion parameter using the observation equation of Kalman filtering. This step aims to map and convert each independent motion parameter observation value into a preliminary gait phase estimate related to the parameter through the observation model of Kalman filtering, to allow the system to independently evaluate the contribution of each sensor data and prepare for subsequent fusion. For example, for heel pressure data, an observation equation can be established to associate heel pressure values with gait phases (e.g. through a nonlinear function to describe the periodic change of heel pressure with gait phase), and for knee joint angle data, an observation equation can also be constructed to associate knee joint angle values with gait phases (e.g. through a nonlinear function to describe the periodic change of knee joint angle values with gait phase). Subsequently, the system obtains a second initial gait phase according to all third initial gait phases to effectively fuse the preliminary gait phase estimates obtained from various independent motion parameters, thereby obtaining a more comprehensive and robust second initial gait phase. Specifically, the system can take the average of all third initial gait phases as the second initial gait phase, or the system can use a weighted average method to assign different weights according to the reliability or importance of each motion parameter, and then perform a weighted average of all third initial gait phases to obtain the final second initial gait phase.

[0041] The embodiment effectively solves the problem of inaccurate multi-parameter data fusion by decomposing the multi-modal motion data into independent motion parameters and fusing them on this basis. Specifically, when predicting the second gait phase, the system no longer directly inputs the multi-modal motion data as a whole into the observation equation of Kalman filtering, but for each independent motion parameter, such as the heel pressure or knee joint angle, the observation equation of Kalman filtering is used to generate a third initial gait phase corresponding to the motion parameter. This processing method ensures that the characteristics of each motion parameter can be independently and accurately modeled and utilized to avoid the interference of possible dimensional differences, noise characteristics differences or nonlinear relationships between different types of sensor data on the overall observation model, thereby enhancing the relevance and reliability of data processing. Subsequently, the system processes all these independently generated third initial gait phases to obtain the final second initial gait phase. This comprehensive process can effectively fuse information from different sensors, making up for the limitations or errors of a single sensor, such as when a certain sensor data is temporarily unreliable, other sensor information can still provide effective phase estimation. In this way, the scheme realizes effective fusion of multi-source data, making the final gait phase prediction result more robust and accurate. As can be seen, the embodiment avoids direct conflict and mutual interference between different sensor data by independently generating a third initial gait phase for each motion parameter, so that the observation information of each parameter can be more accurately utilized. The embodiment realizes effective fusion of multi-modal motion data by synthesizing the second initial gait phase from all third initial gait phases to obtain a more robust and accurate second initial gait phase, so the embodiment can provide high-quality observation input for the subsequent correction step of Kalman filtering, so that the lower limb exoskeleton robot can more accurately perceive the real-time gait state of the patient, thereby providing a more reliable basis for adaptive joint angle vector generation and compliant gait assistance, and effectively improving the adaptability and gait bionics of the lower limb exoskeleton robot in complex environments.

[0042] In some preferred embodiments, the motion parameter is the heel pressure or the knee joint angle, and the calculation formula of the third initial gait phase corresponding to the heel pressure is as follows: ; wherein, represents the heel pressure, represents a preset coefficient corresponding to the heel pressure, represents the third initial gait phase corresponding to the heel pressure; The calculation formula of the third initial gait phase corresponding to the knee joint angle is as follows: ; wherein, denotes the knee joint angle, denotes the preset coefficient corresponding to the knee joint angle, denotes the third initial gait phase corresponding to the knee joint angle, denotes the preset phase offset corresponding to the knee joint angle.

[0043] The heel pressure of this embodiment can be measured by a pressure sensor installed on the foot bottom, which can reflect the start and end of the support phase of the gait. The knee joint angle of this embodiment can be measured by a joint encoder installed at the knee joint, and the change pattern of the knee joint angle is closely related to the swing phase and the support phase in the gait cycle.

[0044] In some preferred embodiments, the calculation formula of the desired joint angle vector is as follows: ; wherein, t denotes time, denotes the time scaling factor, and z(t) denotes the auxiliary vector, denotes the derivative of the auxiliary vector, and both denote gain constants, denotes the desired joint angle vector, denotes the derivative of the desired joint angle vector, and g denotes the reference joint angle vector, denotes the second gait phase, denotes the nonlinear forcing term corresponding to the second gait phase; The calculation formula of the nonlinear forcing term is as follows: ; wherein, denotes the Gaussian radial basis function, and N denotes the number of Gaussian radial basis functions, denotes the width of the i-th Gaussian radial basis function, denotes the center of the i-th Gaussian radial basis function, denotes the weight corresponding to the i-th Gaussian radial basis function.

[0045] The weight corresponding to the i-th Gaussian radial basis function of this embodiment is equivalent to the optimal weight of the above-mentioned embodiment, that is, this embodiment is equivalent to obtaining the optimal weight of the adaptive joint angle vector generator by optimizing the weight corresponding to the Gaussian radial basis function using the demonstration gait data.

[0046] In some preferred embodiments, step S2 comprises: S21, generating the optimal weight set of the adaptive joint angle vector generator using a locally weighted regression algorithm according to all the demonstration gait data.

[0047] The local weighted regression algorithm used in this embodiment belongs to the prior art, and its working principle and working process will not be discussed in detail here.

[0048] As can be seen, the joint angle vector generation method provided by the application has the following advantages: the adaptive joint angle vector generator is provided with the deep learning and adaptive adjustment capability for the gait characteristics of healthy people by using the teaching gait data of healthy people in different environments to pre-generate the optimal weight combination of the adaptive joint angle vector generator, the phase prediction based on multi-sensor data is realized by predicting the second gait phase according to the first gait data and the multi-modal motion data, the accuracy and reliability of phase estimation are effectively improved, the parameters of the adaptive joint angle generator are adjusted according to the surrounding environment by first adjusting the reference joint velocity vector and the time scaling factor of the adaptive joint angle vector generator according to the visual data, and then the expected joint angle vector is dynamically adjusted according to the surrounding environment by using the adjusted adaptive joint angle vector generator to generate the expected joint angle according to the second gait phase, so the application provides a joint angle vector that can adapt to the current environment and has high gait bionics, thereby effectively meeting the rehabilitation needs of patients in real-life scenarios and avoiding the problems of increased abnormal gait due to lack of environmental perception capability, induced non-physiological motion mode, and affected re-establishment of damaged neural pathways, thereby effectively improving the rehabilitation training effect and expanding the application range of rehabilitation training.

[0049] In a second aspect, the application further provides a gait control method applied in a lower extremity exoskeleton robot, which comprises the following steps: B1, acquiring a first gait phase and multi-sensor data at a current time; the first gait phase is the gait phase of the lower extremity exoskeleton robot at a previous time, and the multi-sensor data includes multi-modal motion data and visual data; B2, predicting a second gait phase according to the first gait phase and the multi-modal motion data; the second gait phase is the gait phase of the lower extremity exoskeleton robot at the current time; B3, adjusting the reference joint angle vector and the time scaling factor of the adaptive joint angle vector generator with pre-optimized weights according to the visual data, and then using the adjusted adaptive joint angle vector generator to generate an expected joint angle vector according to the second gait phase; B4, determining a current gait phase according to the second gait phase and the multi-modal motion data; B5, acquiring an expected joint torque according to the expected joint angle vector; B6, adjusting the stiffness and damping of the lower extremity exoskeleton robot according to the current gait phase while keeping the expected joint torque unchanged; The adaptive joint angle vector generator adopts a dynamic motion primitive architecture, and the pre-optimized weight process of the adaptive joint angle vector generator includes: S1, collecting demonstration gait data of healthy people in different environments; S2, generating an optimal weight set of the adaptive joint angle vector generator according to all the demonstration gait data, the optimal weight set including optimal weights corresponding to different lower limb joints.

[0050] The gait control method provided by the application combines a multi-sensor data fusion framework with an adaptive joint angle vector generator in a real-time environment perception manner, and introduces a dynamic stiffness and damping adjustment mechanism based on gait phase, thereby realizing real-time perception of a complex environment and compliant adaptation of gait. Specifically, the fusion of multi-sensor data provides a comprehensive basis for environmental perception, solving the problem of perception missing caused by the dependence on joint encoders in the prior art; the prediction of gait phase combines historical phase and real-time motion information, improving the accuracy of estimation; the adaptive joint angle vector generator dynamically adjusts the reference joint angle vector and the time scaling factor using visual data, and optimizes the weights based on the dynamic motion primitive architecture and the demonstration data of healthy people, ensuring that the generated desired joint angle vector has high bionics and environmental adaptability; and the dynamic adjustment of stiffness and damping provides just the right assistance according to the gait phase, providing high stiffness and low damping in the early support phase to ensure stability, and providing low stiffness and high damping in the swing phase to promote the active participation of the patient. Through the above technical solutions, the lower limb exoskeleton robot can generate a natural and adaptive gait in a complex environment such as uphill and downhill scenarios. For example, when the visual data detects that there is an uphill road section in front, the system adjusts the reference joint angle vector and the time scaling factor according to the visual information, so that the adaptive joint angle vector generator outputs the desired joint angle vector adapted to the uphill terrain, and dynamically adjusts the stiffness and damping parameters according to the current gait phase. Specifically, in the early support phase, the stiffness is adjusted to be high and the damping is adjusted to be low to provide strong support and prevent the knee from collapsing; in the swing phase, the stiffness is reduced and the damping is adjusted to be high to encourage the patient to actively swing the leg. As can be seen, the application can effectively improve the naturalness and effectiveness of rehabilitation training, help to better activate the damaged neural pathways of the patient, avoid the generation of abnormal gait patterns, and thus provide the patient with a safer, more efficient and personalized rehabilitation experience.

[0051] Specifically, the mapping relationship between the gait phase and the stiffness and damping of the lower limb exoskeleton robot of this embodiment is shown in the following table.

[0052] Mapping relationship table of gait phase and stiffness and damping of lower limb exoskeleton robot

[0053] In some preferred embodiments, step B6 comprises: B61, obtaining human-robot interaction torque; B62, adjusting the stiffness and damping of the lower extremity exoskeleton robot according to the current gait phase and the human-robot interaction torque while keeping the desired joint torque unchanged.

[0054] The human-robot interaction torque of this embodiment refers to the torque between the lower extremity exoskeleton robot and the patient's limb, which acts to quantify the active or passive force exerted by the patient on the robot during movement in real time, thus reflecting the patient's movement intention and participation level. This embodiment can obtain the human-robot interaction torque by installing force sensors or torque sensors on the lower extremity exoskeleton robot to directly measure the torque at the joint. In the case of keeping the desired joint torque unchanged, the system adjusts the stiffness and damping of the lower extremity exoskeleton robot according to the current gait phase and the human-robot interaction torque to provide compliant gait assistance. This step is equivalent to dynamically adjusting the assistance characteristics of the lower extremity exoskeleton robot according to the patient's real-time interaction state and the gait phase they are in, to achieve more compliant and personalized gait assistance. Keeping the desired joint torque unchanged means that when adjusting the stiffness and damping, the lower extremity exoskeleton robot still targets the preset desired joint torque for control, but changes the stiffness and damping parameters to affect its response characteristics to the desired joint torque and its compliance to the patient's active force. Stiffness and damping are key parameters that describe the interaction characteristics of the exoskeleton with the patient. Stiffness reflects the exoskeleton's resistance to deviation from the desired position, while damping reflects the exoskeleton's resistance to changes in its movement speed. Adjusting these parameters can change the "softness" of the exoskeleton and the way it follows or guides the patient's movement, for example, when detecting an increase in the patient's active force, the system can automatically reduce the stiffness, thereby reducing the exoskeleton's intervention. The amount of reduction can be through a pre-set lookup table or a fuzzy logic-based controller.

[0055] The working principle of this embodiment is that, first, the human-robot interaction torque is obtained, which directly reflects the active or passive force exerted by the patient on the lower extremity exoskeleton robot at the current moment, thereby quantifying the patient's movement intention and participation level. Subsequently, the system combines the human-robot interaction torque with the current gait phase information as the basis for adjusting the stiffness and damping of the lower extremity exoskeleton robot. During the entire adjustment process, the expected joint torque remains unchanged, which means that the exoskeleton system is still committed to achieving the predetermined gait trajectory, but its implementation becomes more flexible and personalized. Specifically, when the system detects an increase in the patient's active force (human-robot interaction torque), indicating a stronger movement intention, the system will accordingly reduce the stiffness of the lower extremity exoskeleton robot and increase the damping of the lower extremity exoskeleton robot, thereby reducing the exoskeleton's intervention in the patient's movement, making the exoskeleton joints "softer" and allowing the patient to dominate the movement, encouraging active participation. Conversely, when the system detects that the patient has no force or has a small active force (human-robot interaction torque is less than the preset torque threshold), the system will increase the stiffness of the lower extremity exoskeleton robot and reduce the damping of the lower extremity exoskeleton robot, making the exoskeleton joints "harder" and providing more auxiliary support to make up for the patient's lack of strength. This embodiment introduces the human-robot interaction torque into the adjustment strategy of stiffness and damping, overcoming the limitations of relying solely on gait phase adjustment, so that the lower extremity exoskeleton robot's assistance is no longer a rigid preset mode, but can dynamically respond to the patient's real-time state. This adaptive adjustment mechanism that combines gait phase information and patient active intention enables the lower extremity exoskeleton robot to more accurately match the individual needs of the patient, providing more flexible, personalized, and efficient gait assistance, thereby significantly improving the naturalness and effectiveness of rehabilitation training.

[0056] In some preferred embodiments, the formula for calculating the expected joint torque is as follows: ; wherein, represents the expected joint torque, J represents the Jacobian matrix of the lower extremity skeletal robot, represents the transpose of the Jacobian matrix of the lower extremity skeletal robot, represents the preset virtual stiffness, represents the preset virtual damping, represents the expected joint angle vector, q(t) represents the actual joint angle vector, represents the derivative of the expected joint angle vector, represents the derivative of the actual joint angle vector (measured by the joint encoder), represents the preset gravity compensation term.

[0057] In some preferred embodiments, step B4 comprises: B41, taking the gait phase corresponding to the preset phase mapping interval to which the second gait phase belongs and the preset condition satisfied by the multi-modal motion data as the current gait phase.

[0058] The preset phase mapping interval refers to a range in which continuous second gait phases are divided into discrete gait sub-phases with specific physiological meanings. The preset condition refers to a logical rule or threshold value set based on multi-modal motion data, which is used to assist or correct the gait phase judgment. In this embodiment, the gait phase corresponding to the preset phase mapping interval to which the second gait phase belongs and the preset condition satisfied by the multi-modal motion data is taken as the current gait phase, for example, when the second gait phase ∈ [11π / 6, 2π) and the angular velocity of the inertial sensor arranged at the lower leg is zero, the swing phase end is taken as the current gait phase.

[0059] As can be seen from the above, the joint angle vector generation method and the gait control method provided by the present application make the adaptive joint angle vector generator have the deep learning and adaptive adjustment capability for the gait characteristics of healthy people by using the teaching gait data of healthy people in different environments to pre-generate the optimal weight combination of the adaptive joint angle vector generator, realize phase prediction based on multi-sensor data by predicting the second gait phase according to the first gait data and the multi-modal motion data, effectively improve the accuracy and reliability of phase estimation, and adjust the parameters of the adaptive joint angle vector generator according to the surrounding environment by first adjusting the reference joint velocity vector and the time scaling factor of the adaptive joint angle vector generator according to the visual data, and then generating the expected joint angle according to the second gait phase by using the adjusted adaptive joint angle vector generator to dynamically adjust the expected joint angle vector according to the surrounding environment. Therefore, the present application provides a joint angle vector that can adapt to the current environment and has high gait bionics, thereby effectively meeting the rehabilitation needs of patients in real-life scenarios and avoiding the problems of increased abnormal gait due to lack of environmental perception ability, induced non-physiological motion patterns, and affected re-establishment of damaged neural pathways, thereby effectively improving the rehabilitation training effect and expanding the application range of rehabilitation training.

[0060] In the embodiments provided by the present application, it should be understood that, in this document, relationship terms such as first and second are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply that these entities or operations have any such actual relationship or order.

[0061] The above merely provides an example of the present application, and is not used to limit the protection scope of the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included in the protection scope of the present application.

Claims

1. A method for generating joint angle vectors, applied in a lower limb exoskeleton robot, characterized in that, The joint angle vector generation method includes the following steps: A1. Acquire the first gait phase and multi-sensor data at the current moment; the first gait phase is the gait phase of the lower limb exoskeleton robot at the previous moment, and the multi-sensor data includes multimodal motion data and visual data; A2. Predict the second gait phase based on the first gait phase and the multimodal motion data; the second gait phase is the gait phase of the lower limb exoskeleton robot at the current moment; A3. Adjust the reference joint angle vector and time scaling factor of the pre-optimized adaptive joint angle vector generator according to the visual data, and then use the adjusted adaptive joint angle vector generator to generate the desired joint angle vector according to the second gait phase. The adaptive joint angle vector generator adopts a dynamic motion primitive architecture, and the pre-optimization weight process of the adaptive joint angle vector generator includes: S1. Collect teaching gait data of healthy individuals in different environments; S2. Generate the optimal weight reassembly of the adaptive joint angle vector generator based on all the taught gait data, wherein the optimal weight reassembly includes the optimal weights corresponding to different lower limb joints.

2. The joint angle vector generation method according to claim 1, characterized in that, Step A2 includes: A21. Using the state transition equation of the extended Kalman filter, a first initial gait phase is generated based on the first step state phase; the first initial gait phase is a predicted value of the gait phase at the current moment generated based on the first step state phase; A22. Using the observation equation of the extended Kalman filter, a second initial gait phase is generated based on the multimodal motion data; the second initial gait phase is the observed value of the gait phase at the current moment generated based on the phase of the multimodal motion data. A23. The first initial gait phase is corrected using the extended Kalman filter based on the second initial gait phase to obtain the second gait phase.

3. The joint angle vector generation method according to claim 2, characterized in that, The multimodal motion data includes multiple motion parameters, and step A22 includes: A221. For each of the motion parameters, the third initial gait phase corresponding to the motion parameter is generated based on the observation equation of the extended Kalman filter. A222. Obtain the second initial gait phase based on all the third initial gait phases.

4. The joint angle vector generation method according to claim 3, characterized in that, The motion parameters are heel pressure or knee angle, and the formula for calculating the third initial gait phase corresponding to the heel pressure is as follows: ； in, Indicates heel pressure. This indicates the preset coefficient corresponding to heel pressure. This indicates the third initial gait phase corresponding to heel pressure; The formula for calculating the third initial gait phase corresponding to the knee joint angle is as follows: ； in, Indicates the knee joint angle. This represents the preset coefficient corresponding to the knee joint angle. This represents the third initial gait phase corresponding to the knee joint angle. This indicates the preset phase offset corresponding to the knee joint angle.

5. The joint angle vector generation method according to claim 1, characterized in that, The formula for calculating the desired joint angle vector is as follows: ； Where t represents time. Let z(t) represent the time scaling factor, and z(t) represent the auxiliary vector. The derivative of the auxiliary vector, and Both represent gain constants. This represents the desired joint angle vector. Let denote the derivative of the desired joint angle vector, and g denote the reference joint angle vector. Indicates the second gait phase. This represents the nonlinear forcing term corresponding to the second step phase; The formula for calculating the nonlinear forcing term is as follows: ； in, Let N represent the Gaussian radial basis functions, and let N be the number of Gaussian radial basis functions. This represents the width of the i-th Gaussian radial basis function. This represents the center of the i-th Gaussian radial basis function. This represents the weight corresponding to the i-th Gaussian radial basis function.

6. The joint angle vector generation method according to claim 1, characterized in that, Step S2 includes: S21. The optimal weight reorganization of the adaptive joint angle vector generator is generated based on all the taught gait data using a local weighted regression algorithm.

7. A gait control method, characterized in that, When applied to lower limb exoskeleton robots, the gait control method includes the following steps: B1. Acquire the first gait phase and multi-sensor data at the current moment; the first gait phase is the gait phase of the lower limb exoskeleton robot at the previous moment, and the multi-sensor data includes multimodal motion data and visual data; B2. Predict the second gait phase based on the first gait phase and the multimodal motion data; the second gait phase is the gait phase of the lower limb exoskeleton robot at the current moment; B3. Adjust the reference joint angle vector and time scaling factor of the pre-optimized adaptive joint angle vector generator according to the visual data, and then use the adjusted adaptive joint angle vector generator to generate the desired joint angle vector according to the second gait phase. B4. Determine the current gait phase based on the second gait phase and the multimodal motion data; B5. Obtain the desired joint torque based on the desired joint angle vector; B6. While keeping the desired joint torque constant, adjust the stiffness and damping of the lower limb exoskeleton robot according to the current gait stage; The adaptive joint angle vector generator adopts a dynamic motion primitive architecture, and the pre-optimization weight process of the adaptive joint angle vector generator includes: S1. Collect teaching gait data of healthy individuals in different environments; S2. Generate the optimal weight reassembly of the adaptive joint angle vector generator based on all the taught gait data, wherein the optimal weight reassembly includes the optimal weights corresponding to different lower limb joints.

8. The gait control method according to claim 7, characterized in that, Step B6 includes: B61. Obtain the torque for human-computer interaction; B62. While keeping the desired joint torque constant, adjust the stiffness and damping of the lower limb exoskeleton robot according to the current gait stage and the human-machine interaction torque.

9. The gait control method according to claim 7, characterized in that, The formula for calculating the desired joint torque is as follows: ； in, Let J represent the desired joint torque, and J represent the Jacobian matrix of the lower limb skeletal robot. This represents the transpose of the Jacobian matrix of a lower limb skeletal robot. This represents the preset virtual stiffness. This represents the preset virtual damping. Let represent the desired joint angle vector, and q(t) represent the actual joint angle vector. This represents the derivative of the desired joint angle vector. The derivative of the actual joint angle vector. This indicates the preset gravity compensation item.

10. The gait control method according to claim 7, characterized in that, Step B4 includes: B41. The gait stage corresponding to the preset phase mapping interval to which the second gait phase belongs and the preset conditions satisfied by the multimodal motion data is taken as the current gait stage.

Images