A kind of bionic gait generation system and method for gantry type lower limb exoskeleton

Abstract

The application discloses a kind of gantry type lower limb exoskeleton-oriented bionic gait generation system and method, it is related to bionic gait generation technical field, including variation controller, state observer, oscillator sequence, gait generator, wherein variation controller, state observer, oscillator sequence are sequentially connected, the output of gait generator is connected with the input of oscillator sequence, the output of oscillator sequence is connected with gantry type lower limb exoskeleton;Gait generator is used to generate basic gait;Variation controller is used to receive user demand and environmental information, and parameter optimization is carried out to oscillator sequence;State observer is used to transmit the parameter optimization result to oscillator sequence;Oscillator sequence is based on basic gait and parameter optimization result, and gait reconstruction is carried out.The application can stably adapt to different user demand and external environment, realize the online adjustment of walking speed, and the wearer can be statically started on treadmill.

Description

Technical Field

[0001] This invention relates to the field of biomimetic gait generation technology, and more specifically to a biomimetic gait generation system and method for a benchtop lower limb exoskeleton. Background Technology

[0002] With the development of intelligent robotics technology, lower limb exoskeleton robots have gradually attracted attention and become a research hotspot, especially in gait rehabilitation and motor function enhancement. Currently, many commercially available lower limb exoskeletons exist, such as Lokomat, Gait Trainer, and Auto Ambulator. However, research on lower limb exoskeletons is still in its early stages, and several issues make motion control a challenging task. As a motion assistive device, it needs to maintain constant contact with the human body, therefore human-machine coordination should be a primary consideration. This requires the lower limb exoskeleton to be flexible and stable in control and suitable for different wearers. A reasonable gait is a necessary condition for achieving these requirements, meaning that the gait generator can match changes in the environment and make corresponding changes according to the wearer's individual needs. Currently, most research on gait generation focuses on methods based on fixed trajectories. This method can achieve complex and precise movements, and various methods such as fuzzy control, artificial neural networks, genetic algorithms, and reinforcement learning exist. However, these methods are often only applicable to specific devices and environments and are difficult to replicate on other exoskeletons. Take Lokomat as an example. Although its control system is mature, it still needs to lift the wearer up with weight ropes first, and then wait until the speed of the treadmill matches the speed of the exoskeleton before it can put the wearer down.

[0003] Currently, an increasing number of studies are mimicking rhythmic movement mechanisms to control the motion of legged robots. Due to their good performance, researchers have also begun to apply CPG control strategies to lower limb exoskeletons. R. Ronsse et al. investigated the role of adaptive CPG oscillators in the LOPES exoskeleton-assisted lower limb movement. K. Gui et al. designed a multimodal human-computer interaction method to enhance the active participation of subjects during rehabilitation training. RMLuo et al. established a CPG-based impedance control framework that can adapt to the impedance characteristics of the human body. However, the nervous system is highly complex and difficult to simulate accurately, thus presenting many challenges in practical applications. Currently, commonly used CPG models can be broadly classified into two categories: biomimetic half-center oscillator (HCO) models and abstract oscillator (AO)-based CPG models. The output of the HCO model is torque; when wearing an exoskeleton, the relevant dynamic parameters are difficult to measure, therefore, related research results are often applied to prostheses or ankle exoskeletons. The output of the AO model is trajectory or velocity, and the adaptive oscillator model based on it is also a commonly used model. Its key feature is its ability to learn instruction signals. However, when this method is applied to exoskeletons, guidance from a therapist is essential, clearly indicating that this method has limited functionality and is difficult to adapt to the actual needs of human-computer interaction.

[0004] Therefore, how to achieve personalized gait customization suitable for different users, different needs, and different environments is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0005] In view of this, the present invention provides a biomimetic gait generation system and method for a benchtop lower limb exoskeleton.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] A biomimetic gait generation system for a benchtop lower limb exoskeleton includes a variational controller, a state observer, an oscillator sequence, and a gait generator. The output of the variational controller is connected to the input of the state observer, the output of the state observer is connected to the input of the oscillator sequence, the output of the gait generator is connected to the input of the oscillator sequence, and the output of the oscillator sequence is connected to the benchtop lower limb exoskeleton.

[0008] The gait generator is used to generate basic gait;

[0009] The variational controller is used to receive user requirements and environmental information, and to optimize the parameters of the oscillator sequence;

[0010] The state observer is used to receive the parameter optimization results of the variational controller and transmit the parameter optimization results to the oscillator sequence;

[0011] The oscillator sequence is used to receive the basic gait from the gait generator and the parameter optimization results transmitted from the state observer, and to perform gait reconstruction.

[0012] Optionally, the gait generator includes a musculoskeletal model and a CPG network;

[0013] The musculoskeletal model includes 7 segments and 6 joints. The 7 segments include a trunk and two three-segmented lower limbs. Each lower limb has 9 virtual skeletal muscles attached to each segment.

[0014] The CPG network includes 12 oscillators, with each set of 6 oscillators corresponding to 9 virtual skeletal muscles of a lower limb.

[0015] Optionally, the driving torque at the joint is generated by 18 virtual skeletal muscles, with the following functional relationship:

[0016]

[0017] Among them, T ai C is the driving torque of the i-th joint; ij It is a (6×18) constant matrix; T mj The torque generated for the j-th virtual skeletal muscle.

[0018] Optionally, the expression for the oscillator used in the CPG network is:

[0019]

[0020] Among them, u i It refers to the nerve membrane potential and the output of the CPG network; i These are fatigue characteristic parameters of neurons; τ is the rise time constant; τ′ is the fatigue constant; ω ij Let be the connection weight from neuron j to neuron i; β be a constant coefficient; u0 be the constant stimulus input from the higher-level module; F i The expression for the feedback value from the outside world is:

[0021]

[0022] Where, q aij q bij q ci q di and q ei All are constant parameters; θ j and θ 0jThese are the joint angle and reference angle of the j-th joint, respectively; The posture angle of the torso. yes The derivative with respect to time.

[0023] Optionally, the variational controller is used to optimize the parameters of three stages of the oscillator sequence, the three stages being the lower limb exoskeleton activation stage, the treadmill speed matching stage, and the comfort adjustment stage.

[0024] Optionally, the state observer includes a reference oscillator and an amplitude oscillator;

[0025] When the foot touches the bottom, the phase of the reference oscillator is reset to 0, as follows:

[0026]

[0027]

[0028] Where, φ ref and ω ref This indicates the phase and frequency of the reference oscillator; p is the manually set percentage of the phase; ε1 and ε2 are positive constants; ω m ω is a measured value of walking frequency. m =π / T m T m Indicates the cycle of walking; For φ ref The derivative with respect to time; For ω ref The derivative with respect to time; ω1 is the frequency of oscillator 1 in the oscillator sequence; φ1 is the phase of oscillator 1 in the oscillator sequence; It is a deceleration function;

[0029] The amplitude oscillator is in the following form:

[0030]

[0031] Where, ζ i For the input of the amplitude oscillator, parameter l i Restricted to a positive constant, G(i) is the set of oscillators that the i-th oscillator receives coupled to, and R ij The bias of the amplitude is expressed as follows:

[0032]

[0033] in, and These are the intrinsic amplitudes of the i-th and j-th oscillators, respectively.

[0034] Optionally, the oscillator sequence consists of 4×4 oscillators coupled to their nearest neighbors. The 4×4 oscillators represent four blocks, each containing four oscillators. Each joint angle of the lower limb exoskeleton is generated by combining the four oscillators from each block. The joint angles include the left lower limb hip joint angle. Left lower limb knee joint angle Right lower limb hip joint angle and the angle of the right lower limb knee joint The expression is:

[0035]

[0036]

[0037]

[0038]

[0039] in, Let a be the amplitude of the i-th oscillator. i The stable values ​​are r1(t)~r4(t)∈[0,1], which are state variables.

[0040] Optionally, the expression for the oscillator in the oscillator sequence is:

[0041]

[0042] Where, ν ij ω is a constant coefficient; i Convertable to ω(t)∈[0,1] is another state variable. It is the maximum allowable frequency of the i-th oscillator, and the scaling term Ω j / Ω i Used to couple oscillators of different frequencies, φ j -Ω j / Ω i ·φ i A proportional phase difference is defined, and the discrete set D(i) includes the i-th oscillator receiving coupled oscillators, with parameter l. i Restricted to a positive number, It is ω i The second derivative with respect to time; Described by the equation:

[0043]

[0044] φ i (0) and Ω i The value depends on the convergence result of the gait learning system.

[0045] A biomimetic gait generation method for a benchtop lower limb exoskeleton, using a biomimetic gait generation system for a benchtop lower limb exoskeleton as described in any of the preceding claims, generates a biomimetic gait, including the following steps:

[0046] Obtain user needs and environmental information;

[0047] Based on the user requirements and environmental information, the oscillator sequence parameters are optimized.

[0048] Based on the parameter optimization results, the basic gait is reconstructed.

[0049] Optionally, the process of optimizing the parameters of the oscillator sequence includes three stages: lower limb exoskeleton activation stage, treadmill speed matching stage, and comfort adjustment stage.

[0050] As can be seen from the above technical solution, the present invention provides a biomimetic gait generation system and method for benchtop lower limb exoskeletons, which has the following advantages compared with the prior art:

[0051] This invention mainly consists of two parts: one is a gait generator based on a musculoskeletal model and a CPG network, which generates a stable gait by simulating the human motion control system; the other part is an oscillator sequence used to learn the basic gait, and a variational controller is used to optimize the function of each oscillator to adapt to different user needs and environments. Another set of oscillators is used to construct a state observer, which receives modulation from the variational controller and inputs the results into the oscillator sequence. This scheme ensures that the stable limit cycle is not broken, enabling the control system to stably adapt to user needs and the external environment. This invention uses a benchtop lower limb exoskeleton as its application, employing a biomimetic gait generation system to match the speed of the exoskeleton and the treadmill, allowing the wearer to start statically on the treadmill without being attached to a rope.

[0052] This invention applies a CPG control strategy to a desktop lower limb exoskeleton to achieve online adjustment of walking speed, and combines it with a preference control algorithm to achieve personalized control of the exoskeleton. This invention can adapt to the gait characteristics of different wearers, achieving gait customization. Attached Figure Description

[0053] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0054] Figure 1This is a schematic diagram of the overall structure of a biomimetic gait generation system for a benchtop lower limb exoskeleton according to the present invention.

[0055] Figure 2(a) is a schematic diagram of the gait generator;

[0056] Figure 2(b) is a schematic diagram of the musculoskeletal model;

[0057] Figure 2(c) is a schematic diagram of the CPG network structure;

[0058] Figure 3 A schematic diagram of a nonlinear system used to learn basic gait;

[0059] Figure 4(a) is a schematic diagram of the oscillator sequence;

[0060] Figure 4(b) is a schematic diagram of the variational controller;

[0061] Figure 5(a) is a schematic diagram of a four-bar exoskeleton model;

[0062] Figure 5(b) is a schematic diagram of the joint space of the four-bar linkage model;

[0063] Figure 6 This is a schematic diagram of the exoskeleton posture of the five-bar model under different speed relationships. Detailed Implementation

[0064] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0065] This invention discloses a biomimetic gait generation system for benchtop lower limb exoskeletons. (See also...) Figure 1 The system includes a variational controller, a state observer, an oscillator sequence, and a gait generator. The output of the variational controller is connected to the input of the state observer, the output of the state observer is connected to the input of the oscillator sequence, the output of the gait generator is connected to the input of the oscillator sequence, and the output of the oscillator sequence is connected to a benchtop lower limb exoskeleton.

[0066] I. Gait Generator

[0067] The gait generator is used to generate a basic gait. The gait generator includes a musculoskeletal model and a CPG network, as shown in Figure 2(a). The output signal of the CPG network induces body movement by activating muscles, while the current state of the musculoskeletal model and the environment is received by the sensory system and sent to the CPG network.

[0068] The musculoskeletal model is a simplified simulation of the human lower limb. To accommodate the structural features of a benchtop lower limb exoskeleton, the musculoskeletal model is defined as a two-dimensional system comprising 7 segments and 6 joints. The 7 segments include a trunk and two three-segmented lower limbs. Each lower limb has 9 virtual skeletal muscles (gluteus maximus, iliopsoas, hamstrings, rectus femoris, biceps femoris, femoris, gastrocnemius, tibialis anterior, and soleus), which are attached to the respective segments, as shown in Figure 2(b).

[0069] The driving torque at the joint is generated by these 18 virtual skeletal muscles, and the functional relationship is as follows:

[0070]

[0071] Among them, T ai C is the driving torque of the i-th joint; ij It is a (6×18) constant matrix; T mj The torque generated for the j-th virtual skeletal muscle.

[0072] Each muscle generates a driving torque at a specific joint, proportional to the output of the CPG network. Thus, a CPG network was constructed, comprising 12 oscillators, and virtual skeletal muscles associated with each oscillator are shown. Every 6 oscillators correspond to 9 virtual skeletal muscles of a lower limb, as shown in Figure 2(c). Oscillator 1 of the CPG network corresponds to the iliopsoas and rectus femoris muscles of one lower limb, oscillator 2 corresponds to the gluteus maximus and hamstring muscles, oscillator 5 corresponds to the biceps femoris, oscillator 6 corresponds to the rectus femoris, oscillator 9 corresponds to the tibialis anterior, and oscillator 10 corresponds to the gastrocnemius and soleus muscles. Similarly, oscillators 3, 4, 7, 8, 11, and 12 correspond to 9 virtual skeletal muscles of another lower limb.

[0073] The expression for the oscillator used in the CPG network is:

[0074]

[0075] Among them, u i It represents the neural membrane potential of the i-th oscillator and the output of the CPG network; υ i These are fatigue characteristic parameters of neurons; is u i The derivative with respect to time; It is υ i The derivative with respect to time; u j τ is the neural membrane potential of the j-th oscillator; g is the function represented by the third line of equation (2); u represents the independent variable of function g; τ i τ′ is the rise time constant. i ω is the fatigue constant. ijLet be the connection weight from neuron j to neuron i; β be a constant coefficient; u0 be the constant stimulus input from the higher-level module; F i The expression for the feedback value from the outside world is:

[0076]

[0077] Where, q aij q bij q ci q di and q ei All are constant parameters; θ j and θ 0j These are the joint angle and reference angle of the j-th joint, respectively; The posture angle of the torso. yes The derivative with respect to time.

[0078] II. Oscillator Sequence

[0079] The oscillator sequence is used to receive the basic gait from the gait generator and the parameter optimization results transmitted from the state observer, and to perform gait reconstruction.

[0080] To broaden the application range of the acquired basic gait and adapt it to various scenarios, a CPG is introduced to provide a stable limit cycle for the system. Therefore, the angle of each joint can be described by an oscillator, exhibiting limit cycle behavior. Subsequently, another oscillatory network is introduced to estimate the basic gait. The basic gait is learned through this set of oscillators, and then the oscillator parameters are adjusted to meet practical requirements.

[0081] In order for the oscillator to learn the basic gait, the present invention employs, as follows: Figure 3 The nonlinear system shown is based on an adaptive frequency Hopf oscillator. These oscillators are characterized by their ability to learn a periodic input signal y without fixing any external optimization process. input The learned frequencies remain encoded in the system even after the input signal disappears, once the oscillators have converged. Since the oscillations of the Hopf oscillators are harmonic, a suitable linear combination of several Hopf oscillators can reproduce any periodic input signal. It will act as a dynamic Fourier decomposition, with each oscillator encoding one frequency component of the input signal. Furthermore, by using a negative feedback loop, the learned frequencies can be subtracted from the input signal, and oscillators that have not yet converged to a stable frequency can be adapted to the remaining frequency components. After the system stabilizes, synchronization occurs among these oscillators, and each oscillator then generates an inherent bias phase. The following equation describes this system:

[0082]

[0083] Where, φ i and ω i These represent the phase and frequency of the i-th oscillator, respectively; ε is the coupling strength; m is the number of oscillators; α i The amplitude of the i-th oscillator is η; η is the learning constant. It is φ i The derivative with respect to time; e(t) is the bias value; It is ω i The derivative with respect to time; y input (t) is the input signal; It is an estimate of the input signal; It is α i The derivative with respect to time. The output of this system is the fundamental frequency of the oscillator, denoted as Ω. i And the bias phase of the oscillator, denoted as Φ i =φ i (0).

[0084] Based on the above theory, a benchtop lower limb exoskeleton oscillator sequence as shown in Figure 4(a) was constructed, and then the gait was reconstructed.

[0085] The oscillator sequence consists of 4×4 oscillators coupled to their nearest neighbors. Each 4×4 oscillator represents four blocks, with four oscillators in each block. Each joint angle of the lower limb exoskeleton is generated by combining the four oscillators from each block. The joint angles include the left lower limb hip joint angle. Left lower limb knee joint angle Right lower limb hip joint angle and the angle of the right lower limb knee joint Defined as

[0086]

[0087] in, Let a be the amplitude of the i-th oscillator in the oscillator sequence. i The stable values; r1(t)~r4(t)∈[0,1] are state variables. The expression for the oscillator in the oscillator sequence is:

[0088]

[0089] Where, ν ij ω is a constant coefficient; i Convertable to ω(t)∈[0,1] is another state variable. It is the maximum permissible frequency of the i-th oscillator in the oscillator sequence, and the scaling term Ω is... j / Ω i Used to couple oscillators of different frequencies, φj -Ω j / Ω i ·φ i A proportional phase difference is defined, and the discrete set D(i) includes the i-th oscillator receiving coupled oscillators, with parameter l. i Restricted to a positive number, It is ω i The second derivative with respect to time; Described by the equation:

[0090]

[0091] φ i (0) and Ω i The value depends on the convergence result of the gait learning system. Therefore, no feedback loop is needed in the oscillator sequence, which prevents the oscillators from being attracted by small-amplitude oscillations. The angle of each joint is generated by a fixed number of oscillators. Although it cannot accurately reproduce the basic gait, it can effectively avoid the influence of noise in the original gait. It is only necessary to make the output of the oscillation sequence close to the trajectory of the gait to reflect the main characteristics of the gait.

[0092] Then, by modifying the state variables ω and r i This allows the gait to adapt to various practical needs. This is equivalent to performing functional programming on these two state variables, while the role of the variational controller (as shown in Figure 4(b)) is to perform offline or online function optimization on these two parameters according to the needs of the user and the external environment.

[0093] III. Variational Controller

[0094] The variational controller is software running on an industrial control computer, used to receive user requirements and environmental information, and to optimize the parameters of the three stages of the oscillator sequence. The three stages include the lower limb exoskeleton activation stage, the treadmill speed matching stage, and the comfort adjustment stage.

[0095] (1) Lower limb exoskeleton activation phase

[0096] Based on the structural characteristics of the benchtop lower limb exoskeleton, a simplified four-bar exoskeleton model was established, as shown in Figure 5(a). It consists of four links. l1~l4, d1~d4, and m1~m4 are the lengths, centers of mass, and masses of each link, respectively; θ1~θ4 are the deflection angles of each link relative to the direction of gravity.

[0097] Assuming zero friction at each joint, one degree of freedom for each joint, and that the exoskeleton can only move in the sagittal plane, the dynamic characteristics of this model can be derived using the standard Lagrange formula:

[0098]

[0099] in yes The derivative with respect to time, yes The second derivative with respect to time, p 12 =m2l1d2,p 34 =m4l3d4,

[0100] As shown in Figure 5(b), the definition is... The joint angle in the joint space. If the driving torque of the joint is given, then...

[0101]

[0102] in,

[0103] The optimization objective selected in this invention is the equivalent energy consumption performance index J:

[0104]

[0105] The upper limit of the points, T, is the time spent in the startup phase; k0 is a penalty coefficient used to avoid sudden changes in joint angles during the later stages of startup.

[0106] From the dynamic equations and equations of the exoskeleton model, τ i It can be represented in the following abstract form:

[0107]

[0108] It can fully describe the movement process and optimal target of the exoskeleton. According to the equation, these variables can be represented by state variables ω and r. i This can be described as follows. Therefore, solving the equation is essentially about finding the optimal state variables. This is a typical variational problem. Furthermore, since there is no interaction involved, there is no need to build a state observer.

[0109] The Raleigh-Ritz method is used here to solve this variational problem. Its core principle is to bypass the differential boundary value problem and transform the variational problem into a nonlinear programming problem. After optimization programming, an approximate curve is directly obtained. Assume the optimal solution for the state variables is approximately:

[0110]

[0111] in

[0112]

[0113]

[0114] c ij The coefficients to be solved are denoted as .

[0115] s i The following boundary conditions must be met:

[0116]

[0117] Variational problems can be transformed into the following abstract form:

[0118] minJ=J(C 5×3 (16)

[0119] Where C = [c ij ] 5×3 .

[0120] (2) Treadmill speed matching stage

[0121] When the exoskeleton walks on a treadmill, the speed relationship between the exoskeleton and the treadmill affects the wearer's torso deflection angle, thus establishing a... Figure 6 The five-bar model shown.

[0122] Assume the exoskeleton's terminal velocity remains constant upon landing. Let υ0 represent the treadmill's velocity, and let υ and υ x This represents the velocity of the exoskeleton's foot and horizontal velocity, while θ represents the angle between the torso and the vertical direction. For example... Figure 6 As shown, there are three scenarios:

[0123] 1.υ x =υ0. This is the ideal situation, where θ = 0, meaning the torso will not shift during walking.

[0124] 2.υ x >υ0. In this situation, the torso will fall backward. The body will remain in a backward posture and will not change.

[0125] 3.υ x <υ0. In this situation, the torso will lean forward. The angle of the torso forward will become larger and larger until it falls over.

[0126] Speed ​​matching has two main objectives: one is to keep the torso upright, which can be achieved by resetting the phase of the oscillator. The other is to make the speed of the exoskeleton end roughly match the speed of the treadmill.

[0127] To reset the oscillator's phase, a reference oscillator is introduced as a state observer; its phase is reset to 0 when the foot touches the ground. It takes the following form:

[0128]

[0129] Where, φ ref and ω ref This indicates the phase and frequency of the reference oscillator; p is the manually set percentage of the phase; ε1 and ε2 are positive constants; ω m ω is a measured value of walking frequency. m =π / T m T m This indicates the period of travel (relative to half the oscillator period). For φ ref The derivative with respect to time, For ω ref The derivative with respect to time, ω1 is the frequency of oscillator 1 in the oscillator sequence, and φ1 is the phase of oscillator 1 in the oscillator sequence. For the deceleration function to ensure φ ref ≤π can be viewed as an exponential function, as shown below:

[0130]

[0131] Where β=ω0 / (1-p)π, η=ln(ω0 / πβ)+β·pπ / ω0, ω0 represents the oscillator 1 in the oscillator sequence at φ ref The value of ω1 when pπ. Therefore, the expression for oscillator 1 in the oscillator sequence becomes:

[0132]

[0133] Among them, ε3 and ε4 are positive constants.

[0134] When the exoskeleton contacts the ground, it is difficult to maintain a constant horizontal velocity of the feet. Therefore, adjusting the frequency can only ensure that the exoskeleton is in the correct position at the moment of landing. To achieve approximate velocity matching throughout the contact period, an amplitude oscillator is introduced as another state observer, in the following form:

[0135]

[0136] Where, ζ i Let G(i) be the input of the amplitude oscillator, and G(i) be the set of oscillators that the i-th oscillator receives coupled to. ij The bias of the amplitude is expressed as follows:

[0137]

[0138] in, and These are the intrinsic amplitudes of the i-th and j-th oscillators, respectively.

[0139] Because of the periodicity of walking, it is not necessary to analyze the entire walking process; instead, a single gait period is used as the solution interval. Therefore, ζ can be considered... i It is a function of the phase of the reference oscillator, ζ i =ζ i (φ ref Because φ ref ∈[0,π],ζ i The boundary condition is ζ i (0)=ζ i (π).

[0140] (3) Comfort adjustment stage

[0141] Personalized gait customization essentially involves adjusting the state variables ω and r. i The goal is to maximize the wearer's comfort. The problem is that human comfort cannot be quantified, so the objective function cannot be explicitly defined. Instead, humans are adept at comparing options and expressing a preference for one of them. This insight allows us to approach the optimization function in a different way. Inspired by this, we can use preference learning (PL) algorithms to construct the optimization objective.

[0142] The core of PL lies in learning a latent function from preferences. Suppose that the wearer has been shown a set of N distinct instances x. i ∈R d (represented as χ={x i Given M pairs of items (i = 1, ..., N), and considering that the wearer selected the most preferred item from each pair, a ranking pair dataset can be created.

[0143]

[0144] The symbol > indicates that the wearer is... and China is more inclined to From the dataset, the posterior distribution of the latent objective function f can be obtained:

[0145]

[0146] Assume the prior distribution of the objective function is a zero-mean Gaussian process (GP), P(f) = N(0, Σ), where Σ is an N×N covariance matrix, and its i-th row and j-th column is the covariance equation k(x). i ,x j ) = exp(-x i -x jP(D|f) is the overall probability of preference given a specific objective function value.

[0147] Superior The probability is

[0148]

[0149] Where ε is Gaussian noise: Φ(·) is the CDF of the standard normal distribution. Likelihood is the joint probability of observing a preference relationship given the latent function values; it can be evaluated as the product of likelihood functions:

[0150]

[0151] To obtain the latent function, it is necessary to estimate the posterior distribution of the latent utility function for given discrete data. In other words, it is necessary to maximize:

[0152]

[0153] By using the Laplace approximation, let The gradient matrix, Given the Hessian matrix, we can obtain:

[0154]

[0155] Where f MAP This is the maximum a posteriori (MAP) estimate.

[0156] The Hessian matrix is ​​a positive semi-definite matrix. Therefore, the MAP at f = f can be found using a simple Newton-Raphson recursion. MAP Estimate:

[0157]

[0158] therefore

[0159] P(f|D)≈N(f MAP H -1 (29)

[0160] The objective to be derived is the prediction distribution P(f) t |D), follows the direct convolution of two Gaussian functions:

[0161] P(f t |D)=∫P(f t |f)P(f|D)df, (30)

[0162] Where ft =[f(r),f(s)] T To test the zero-mean latent variable (r,s) of the pair, the final preference prediction distribution between the two test points r and s is as follows:

[0163]

[0164] in, as well as

[0165] Another embodiment of the present invention provides a biomimetic gait generation method for a benchtop lower limb exoskeleton, which uses a biomimetic gait generation system for a benchtop lower limb exoskeleton as described above to generate a biomimetic gait, including the following steps:

[0166] Obtain user needs and environmental information;

[0167] Based on the user requirements and environmental information, the oscillator sequence parameters are optimized.

[0168] Based on the parameter optimization results, the basic gait is reconstructed.

[0169] Optionally, the process of optimizing the parameters of the oscillator sequence includes three stages: the lower limb exoskeleton activation stage, the treadmill speed matching stage, and the comfort adjustment stage. For details, please refer to the relevant description in the above-mentioned bionic gait generation system.

[0170] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other.

[0171] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A biomimetic gait generation system for benchtop lower limb exoskeletons, characterized in that, It includes a variational controller, a state observer, an oscillator sequence, and a gait generator. The output of the variational controller is connected to the input of the state observer, the output of the state observer is connected to the input of the oscillator sequence, the output of the gait generator is connected to the input of the oscillator sequence, and the output of the oscillator sequence is connected to a benchtop lower limb exoskeleton. The gait generator is used to generate basic gait; The variational controller is used to receive user requirements and environmental information, and to optimize the parameters of the oscillator sequence; The state observer is used to receive the parameter optimization results of the variational controller and transmit the parameter optimization results to the oscillator sequence; The oscillator sequence is used to receive the basic gait of the gait generator and the parameter optimization results transmitted by the state observer, and to perform gait reconstruction; The state observer includes a reference oscillator and an amplitude oscillator; When the foot touches the bottom, the phase of the reference oscillator is reset to 0, as follows: in, and This indicates the phase and frequency of the reference oscillator; The percentage of the phase that is manually set; and It is a positive number; This is a measurement of walking frequency. ,in Indicates the cycle of walking; for The derivative with respect to time; for The derivative with respect to time; This represents the frequency of oscillator number 1 in the oscillator sequence. This represents the phase of oscillator number 1 in the oscillator sequence; It is a deceleration function; The amplitude oscillator is in the following form: in, For the input of the amplitude oscillator, parameters Restricted to a positive number, Let i be the set of oscillators that receive coupling from the i-th oscillator. The bias of the amplitude is expressed as follows: in, and These are the intrinsic amplitudes of the i-th and j-th oscillators, respectively.

2. The biomimetic gait generation system for a benchtop lower limb exoskeleton according to claim 1, characterized in that, The gait generator includes a musculoskeletal model and a CPG network; The musculoskeletal model includes 7 segments and 6 joints. The 7 segments include a trunk and two three-segmented lower limbs. Each lower limb has 9 virtual skeletal muscles attached to each segment. The CPG network includes 12 oscillators, with each set of 6 oscillators corresponding to 9 virtual skeletal muscles of a lower limb.

3. The biomimetic gait generation system for a benchtop lower limb exoskeleton according to claim 2, characterized in that, The driving torque at the joint is generated by 18 virtual skeletal muscles, and the functional relationship is as follows: , in, For the first i The driving torque of each joint; for( constant matrix; For the first j The torque generated by a virtual skeletal muscle.

4. The biomimetic gait generation system for a benchtop lower limb exoskeleton according to claim 2, characterized in that, The expression for the oscillator used in the CPG network is: in, It refers to the nerve membrane potential and the output of the CPG network; These are fatigue characteristic parameters of neurons; The rise time constant; It is the fatigue constant; For the first j The first neuron to the second i The connection weights of each neuron; The coefficient is constant. For constant stimulus input from the higher-level module; The expression for the feedback value from the outside world is: in, , , , as well as All are constant parameters; and The first j The joint angles and reference angles of each joint; The posture angle of the torso. yes The derivative with respect to time.

5. A biomimetic gait generation system for a benchtop lower limb exoskeleton according to claim 1, characterized in that, The variational controller is used to optimize the parameters of three stages of the oscillator sequence, namely the lower limb exoskeleton activation stage, the treadmill speed matching stage, and the comfort adjustment stage.

6. The biomimetic gait generation system for a benchtop lower limb exoskeleton according to claim 1, characterized in that, The oscillator sequence consists of 4×4 oscillators coupled to their nearest neighbors. Each 4×4 oscillator represents four blocks, with four oscillators in each block. Each joint angle of the lower limb exoskeleton is generated by combining the four oscillators from each block. The joint angles include the left lower limb hip joint angle. left lower limb knee joint angle Right lower limb hip joint angle and the angle of the right lower limb knee joint The expression is: in, For the first i The amplitude of each oscillator represents The stable value; It is a state variable.

7. A biomimetic gait generation system for a benchtop lower limb exoskeleton according to claim 6, characterized in that, The expression for the oscillator in the oscillator sequence is: in, The coefficient is constant. Convertable to , For another state variable, It is the first i Maximum permissible frequency for each oscillator, scaling term Used to couple oscillators of different frequencies. The proportional phase difference and discrete set are defined. Including the i Each oscillator receives a coupled oscillator, with parameters... Restricted to a positive number, yes The second derivative with respect to time; Described by equation (7): and The value depends on the convergence result of the gait learning system.

8. A biomimetic gait generation method for benchtop lower limb exoskeletons, characterized in that, Using the biomimetic gait generation system for a benchtop lower limb exoskeleton as described in any one of claims 1-7, the generation of biomimetic gait includes the following steps: Obtain user needs and environmental information; Based on the user requirements and environmental information, the oscillator sequence parameters are optimized. Based on the parameter optimization results, the basic gait is reconstructed.

9. A biomimetic gait generation method for a benchtop lower limb exoskeleton according to claim 8, characterized in that, The process of optimizing the parameters of the oscillator sequence includes three stages: the lower limb exoskeleton activation stage, the treadmill speed matching stage, and the comfort adjustment stage.

Images