Brain-like intelligent EEG flexible rehabilitation robot

Abstract

The application discloses a kind of brain-like intelligent EEG flexible rehabilitation robot, it is related to medical instrument and rehabilitation engineering technical field, through the complexity and self-adaptability of the brain-like intelligent neural network control model designed to simulate human brain neural network, through learning and optimization control strategy, realize the accurate control to flexible lower limb walking aid robot;To realize the gradual abstraction and integration of electroencephalogram signal from primary perception to senior cognition;Through the powerful computing capacity of deep learning, the fusion model can automatically learn and optimize the representation of these features, so that the feature extraction process is more efficient and accurate;And fusion model can constantly update and optimize its parameters and structure according to newly collected data, to adapt to the dynamic changes of electroencephalogram signal, also provides more personalized and accurate rehabilitation guidance for patients.

Description

Technical Field

[0001] This invention relates to the field of medical devices and rehabilitation engineering technology, specifically a brain-like intelligent EEG flexible rehabilitation robot. Background Technology

[0002] Hemiplegia following stroke is a common neurological disorder primarily caused by brain damage resulting from ruptured or blocked blood vessels in the brain. This disease severely impacts patients' motor abilities, sensory functions, and daily living skills, placing a heavy burden on patients' families and society. Stroke has become one of the major health problems in my country and globally. Currently, routine rehabilitation treatments for hemiplegia following stroke include physical therapy, occupational therapy, speech therapy, psychotherapy, and the use of assistive devices. In addition, traditional rehabilitation therapies such as massage, acupuncture, and traditional Chinese medicine are also frequently used. Drug therapy includes improving blood circulation, reducing inflammation, and promoting nerve regeneration. For severe limb dysfunction, surgical treatment may be necessary.

[0003] Rehabilitation robots have demonstrated significant potential and superiority in assisting hemiplegic patients with motor training. They not only improve training efficiency and reduce costs but also enable the integrated application of various active and passive training strategies. The robot's movement trajectory and the force applied to the affected limb exhibit excellent consistency, facilitating the development of scientific training plans and improving patient rehabilitation outcomes. Therefore, combining conventional rehabilitation therapy with intelligent, convenient, and practical rehabilitation robots is a crucial approach to enhancing the rehabilitation effects for stroke-induced hemiplegic patients. However, existing exoskeleton robot-assisted rehabilitation often employs preset trajectory tracking modes (passive rehabilitation), where patients are "forced" to follow the device's movements, lacking autonomy and a sense of participation. Existing intention recognition based on signals such as electromyography (EMG) suffers from response delays or insufficient accuracy, failing to achieve truly "brain-controlled" active rehabilitation that precisely synchronizes with the patient's movement intentions. Summary of the Invention

[0004] To address the shortcomings mentioned in the background art, the present invention aims to provide a brain-like intelligent EEG flexible rehabilitation robot.

[0005] Firstly, the objective of this invention can be achieved through the following technical solution: a brain-like intelligent EEG flexible rehabilitation robot, comprising:

[0006] The data acquisition module is used to acquire electroencephalogram (EEG) signal data and limb movement data, and send them to the preprocessing module. The EEG signal data is acquired based on the EEG acquisition unit, and the limb movement data is acquired based on the inertial measurement unit (IMU).

[0007] The preprocessing module is used to preprocess the EEG signal data to obtain processed EEG signal data, estimate the limb posture of the limb movement data to obtain limb orientation data, and send the processed EEG signal data to the intention recognition module and the limb orientation data to the control command module.

[0008] The intent recognition module is used to input the processed EEG signal data into a pre-established brain-like EEG signal recognition network model, output the patient's EEG analysis results, and send them to the control command module.

[0009] The control command module is used to generate joint control commands for the patient based on the patient's EEG analysis results and limb orientation data, and send them to the execution module;

[0010] The execution module is used to control the flexible rehabilitation robot to perform corresponding joint rehabilitation actions based on the patient's joint control commands.

[0011] In conjunction with the first aspect, in some implementations of the first aspect, the system further includes: the EEG signal data includes SSVEP signals and motor imagery EEG signals;

[0012] Limb motion data includes: angular velocity data, acceleration data, and local geomagnetic data.

[0013] In conjunction with the first aspect, in some implementations of the first aspect, the system further includes: the process by which the preprocessing module preprocesses the EEG signal data includes denoising and filtering the EEG signal data;

[0014] The process of estimating limb posture from limb motion data is as follows:

[0015] An improved complementary filtering method is used to fuse limb motion data to obtain a fusion coefficient, and the fusion coefficient is dynamically adjusted according to the deviation of acceleration and local geomagnetism; thus realizing the dynamic fusion of angular velocity integral and vector observation.

[0016] In conjunction with the first aspect, in some implementations of the first aspect, the system further includes: the pre-established brain-like EEG signal recognition network model of the intent recognition module includes an encoding module, a residual module, a pulse connection module, a pulse fully connected module, and a decoding module. The pulse encoding module converts the EEG signal into a pulse signal using Poisson coding, enabling the pulse signal to transmit information within the network. The residual module consists of convolutional layers and pulse firing layers, used for pulse information feature extraction. The pulse connection module consists of pulse firing layers and pooling layers. The pulse fully connected module transforms the two-dimensional pulse network into a one-dimensional pulse network and outputs the classification pulse result. The decoding module transforms the pulse classification result into an actual classification action.

[0017] In conjunction with the first aspect, in some implementations of the first aspect, the system further includes: the processing procedure of the pre-established brain-like electroencephalogram (EEG) signal recognition network model is as follows:

[0018] The SSVEP signal was analyzed to identify the frequency components corresponding to the patient's gaze side;

[0019] Feature extraction and classification of EEG signals related to motor imagery can identify the patient's motor intentions;

[0020] The frequency components corresponding to the patient's gaze side are fused with the patient's movement intention, and a judgment is made based on a preset decision algorithm to obtain the patient's EEG analysis results.

[0021] In conjunction with the first aspect, in some implementations of the first aspect, the system further includes: the execution module includes executing corresponding joint rehabilitation actions through three-loop closed-loop control of current, speed, and position; position feedback is obtained through an encoder; speed feedback is obtained through position feedback differentiation; current feedback is obtained through current acquisition; during position closed-loop control, the position trajectory generator provides a position reference according to the set position trajectory in a time-division manner; the position reference is compared with the position feedback and adjusted by the position loop controller to obtain the speed reference; the speed reference is compared with the speed feedback and adjusted by the speed loop controller to obtain the current reference; finally, the current reference is compared with the current feedback and adjusted by the current loop to obtain the three-phase voltage waveform controlling the PMSM.

[0022] In conjunction with the first aspect, in some implementations of the first aspect, the system further includes: the three-loop closed-loop control of the execution module is implemented based on a permanent magnet synchronous motor control algorithm with field-oriented control to achieve high-precision control.

[0023] The beneficial effects of this invention are:

[0024] This invention mimics the complexity and adaptability of the human brain's neural network through a designed brain-like intelligent neural network control model. By learning and optimizing control strategies, it achieves precise control of a flexible lower limb assistive robot. This enables the gradual abstraction and integration of EEG signals from primary perception to advanced cognition. The powerful computing capabilities of deep learning allow the fusion model to automatically learn and optimize the representation of these features, making the feature extraction process more efficient and accurate. Furthermore, based on newly acquired data, the fusion model continuously updates and optimizes its parameters and structure to adapt to dynamic changes in EEG signals, providing patients with more personalized and precise rehabilitation guidance. Attached Figure Description

[0025] 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, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0026] Figure 1 This is a schematic diagram of the system structure of the present invention;

[0027] Figure 2 This is a schematic diagram of the overall framework of the present invention;

[0028] Figure 3 This is a schematic diagram of the axis and surface of the human body of the present invention;

[0029] Figure 4 This is a schematic diagram of the working space at the end of the thigh and lower leg in an embodiment of the present invention;

[0030] Figure 5 This is a schematic diagram of motion capture within an embodiment of the present invention;

[0031] Figure 6 This is a schematic diagram illustrating the motion within an embodiment of the present invention;

[0032] Figure 7 This is a schematic diagram of the virtual ground contact phase coordinate system in an embodiment of the present invention;

[0033] Figure 8 This is a schematic diagram of the flexible exoskeleton configuration in an embodiment of the present invention;

[0034] Figure 9 This is a schematic diagram of the VR EEG head-mounted device in an embodiment of the present invention;

[0035] Figure 10 This is a schematic diagram of the ReC-SNN network structure in an embodiment of the present invention;

[0036] Figure 11 This is a schematic diagram of the algorithm flow within an embodiment of the present invention;

[0037] Figure 12 This is a schematic diagram of the functional framework of the control unit in an embodiment of the present invention;

[0038] Figure 13 This is a schematic diagram of the three-loop closed-loop structure for position servo control in an embodiment of the present invention;

[0039] Figure 14 This is a schematic diagram of the key points of the exoskeleton drive technology in an embodiment of the present invention;

[0040] Figure 15 This is a schematic diagram of the overall control method and strategy within an embodiment of the present invention;

[0041] Figure 16 This is a schematic diagram of the rehabilitation strategy framework in an embodiment of the present invention. Detailed Implementation

[0042] 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.

[0043] Example 1:

[0044] like Figure 1 As shown, a brain-like intelligent EEG flexible rehabilitation robot includes:

[0045] The data acquisition module is used to acquire electroencephalogram (EEG) signal data and limb movement data, and send them to the preprocessing module. The EEG signal data is acquired based on the EEG acquisition unit, and the limb movement data is acquired based on the inertial measurement unit (IMU).

[0046] The EEG signal data includes SSVEP signals and motor imagery EEG signals.

[0047] Limb motion data includes: angular velocity data, acceleration data, and local geomagnetic data.

[0048] The specific processing steps for limb movement data include:

[0049] Bionic structure creation and human-machine coupling system modeling of flexible wearable rehabilitation robot.

[0050] The following is an analysis of the motion of various joints in the human lower limb during the creation of a flexible robot biomimetic structure:

[0051] The bones of the lower limbs support the human body and, together with ligaments and joints, enable walking. The hip, knee, and ankle joints play a crucial role in lower limb movement. These three joints allow the lower limbs to perform flexion, extension, adduction, and rotation, thus enabling various human activities. This invention primarily addresses the lower limb walking problems of stroke patients with hemiplegia, focusing mainly on the movement of the knee and ankle joints.

[0052] First, establish the concept of the human body coordinate system in anatomy, such as Figure 3 As shown, in anatomy, the human body is often divided into three planes and three axes to describe joint movement. The three planes intersect each other perpendicularly, namely:

[0053] (1) Coronal plane: a plane that bisects the front and back of the human body in the middle and is perpendicular to the ground;

[0054] (2) Sagittal plane: a plane that bisects the left and right sides of the human body and is perpendicular to the ground;

[0055] (3) Horizontal plane: A plane parallel to the ground plane that bisects the upper and lower parts of the human body.

[0056] The three axes are the sagittal axis, the coronal axis, and the vertical axis. Each of the three axes is perpendicular to its corresponding plane, and they are mutually perpendicular and intersect at the center of the human body, thereby determining the coordinate position of each structure. Among them, the present invention mainly focuses on the movement of the knee and ankle joints in the sagittal plane.

[0057] Generally speaking, the movement patterns of the lower limb joints can be divided into four main categories:

[0058] (1) Flexion and extension movements, movements around the coronal axis;

[0059] (2) Abduction and adduction movements, movements around the sagittal axis;

[0060] (3) Internal and external rotational motion, motion around the vertical axis;

[0061] (4) Circular motion, which connects the actions of flexion and extension, abduction and adduction, with the distal side making a circular motion around the proximal side.

[0062] Based on the above classification of joint movements, specifically for the knee and ankle joints, the main aspects are as follows: (1) Knee joint: The range of flexion and extension is -130° to 0°; when flexion and extension are 90°, small-amplitude internal and external rotation of the knee can be performed. When flexion and extension exceed 90°, the range of motion in the cross section and coronal plane decreases, the knee joint is straightened, and abduction or adduction cannot be performed; (2) Ankle joint: Plantar flexion and dorsiflexion. The range of motion of the ankle joint in the sagittal plane is 45°, the plantar flexion angle is 25° to 35°, and the dorsiflexion angle is 10° to 20°.

[0063] Specifically, based on the division of upper joint movements, assuming that the thigh and lower leg are both unit lengths, the working space of the thigh end driven by the hip joint is calculated using Mathematica software; similarly, the working space surface of the lower leg driven by the knee joint is calculated as follows: Figure 4 As shown.

[0064] Humans rely on the continuous coordination of body structures such as the hip, knee, and ankle joints to move forward while walking. The posture of the lower limbs during walking is called gait. Normal walking gait has three distinct characteristics: constraint, periodicity, and repetition. However, due to knee and ankle joint dysfunction, stroke hemiplegic patients exhibit abnormal walking gait, resulting in the loss of these three characteristics during walking. Therefore, this invention first requires collecting gait parameters of the lower limbs of stroke hemiplegic patients with knee hyperextension and comparing them with the walking gait of normal individuals to obtain the gait characteristics of the lower limbs of stroke hemiplegic patients, providing input for the design of a flexible wearable rehabilitation robot.

[0065] This invention employs a three-dimensional dynamic motion capture system based on Motion Analysis for image acquisition and gait data testing. A schematic diagram of this system is shown below. Figure 5 As shown, it mainly includes: a Raptor-H digital motion capture lens, a system calibration rod device, reflective markers, a camera power supply system, and a high-speed connection cable. In addition, this invention also requires a main control computer to perform system calibration, marker capture control, and image data recording via a Cortex. Using the above system, the entire system is first calibrated; then gait acquisition begins, capturing the gait of the human lower limbs, and the results are as follows... Figure 6 As shown.

[0066] Ankle joint movement direction along the Z-axis variation graph (data exported from software):

[0067] During normal gait, the lower limbs move forward rhythmically up and down, and also shift slightly to the left and right, with a uniform and harmonious movement pattern. When walking from the side (i.e., looking at the Z-axis forward direction), the body can be seen moving up and down rhythmically, with its trajectory approximating a uniform sine wave. The highest point of the upward movement is at the end of the swing phase, and the lowest point of the downward movement is within the double support phase. When walking from the axial direction, when the support shifts from one lower limb to the other, the body's center of gravity also shifts from one supporting leg to the other. The alternation of supporting legs causes the body's center of gravity to shift left and right repeatedly, resulting in a slight shift in the trajectory distribution plane, forming a uniform approximate sine curve in the horizontal plane.

[0068] Dynamics of human lower limb walking:

[0069] The human lower limbs have six joint degrees of freedom (three for the hip joint, one for the knee joint, and two for the ankle joint), making them a typical complex multibody structure. In this invention, the Lagrange method is used to establish the differential motion equations of the system for processing.

[0070] In this invention, the movement of the knee and ankle joints is mainly considered. These movements are all within the sagittal plane. Movements in the coronal and horizontal planes consume very little energy. Therefore, when designing the mechanical structure of the flexible wearable robot, the assistance is mainly designed for movement in the sagittal plane. No drive is designed for the other two planes. The dynamic analysis of the human lower limbs in this part is also only within the sagittal plane.

[0071] Based on the above analysis and considering the patient's actual action state, this invention adopts a dynamic model of the virtual ground contact phase, such as... Figure 7 As shown.

[0072] The "virtual contact phase" refers to a state where at least one foot is not fully on the ground, such as the heel just touching the ground or the toes just off the ground. In this case, the torso (3) is divided into two parts, and the lower limb system can be decomposed into a combination of two four-bar linkage models. Assuming that both legs are in virtual contact with the ground simultaneously, a coordinate system is established with the ground plane as the horizontal axis and the vertical direction as the vertical axis. The distances between the two foot contact points and the origin are constants a and b, respectively. The mass of the left limb is m3L, and the coordinates of its center of mass are (x3L, z3L). The mass of the right limb is m3R, and the coordinates of its center of mass are (x3R, z3R). The mass distribution of the torso (3) should be calculated according to the distances of the center of mass coordinates; thus, the final Lagrange dynamic equations are obtained.

[0073] Flexible wearable robot configuration design:

[0074] The flexible wearable robot's body structure is the focus of this invention, as it directly relates to the robot's performance, wearing comfort, and rehabilitation strategies, and is also the foundation for modeling human-machine coupling systems; for example... Figure 8 The flexible exoskeleton configuration shown generally comprises a mechanical structure, sensor system, control system, actuators, and other components. However, the flexible wearable robot of this invention primarily uses restraint straps and brushless motors, allowing it to adhere perfectly to the human body surface and exhibiting excellent wearability. Force is transmitted through the forward and reverse rotation of the brushless motors, thus achieving the function of flexible wearability. The overall structure meets the following requirements:

[0075] (1) The degrees of freedom of the flexible mechanical mechanism should be similar to the number of degrees of freedom required for human movement, so that the human body can complete certain movements after wearing it without any obvious obstruction;

[0076] (2) The size can be adjusted to take into account the differences in body shape among different wearers;

[0077] (3) The connection with the human body should be comfortable and easy to use;

[0078] (4) Lightweight design reduces the burden on the human body while meeting certain strength requirements.

[0079] Based on the above analysis, taking the flexible lower limb rehabilitation robot for the left knee joint as an example, it mainly consists of a knee joint binding mechanism and a rotation locking mechanism. The expected function is achieved through the combined action of brushless servo motor drive and mechanical limit. The binding mechanism includes a thigh guard, a calf guard, and straps. In order to better fit the thigh and calf, it is made of flexible materials and has padding on the inside to increase the user's wearing comfort.

[0080] In the rotary locking mechanism, the thigh plate is fixedly connected to the rotary shaft, and the lower leg plate is fixedly connected to the winding wheel, using bearings to achieve active degrees of freedom. The Bowden rope is wound in the groove of the winding wheel and fixed at the fixed block, and the limit block prevents the Bowden rope from leaving the groove of the winding wheel; the fixed pulley is used to increase the winding length to enhance the motor's assist effect. The rotary locking mechanism is also equipped with an encoder to measure the rotation angle of the flexible skeleton robot in real time.

[0081] The process of acquiring EEG signal data is as follows:

[0082] Equipment preparation:

[0083] 1) VR EEG headset: Figure 9 It is a VR EEG head-mounted device used to display visual stimuli and provide an immersive experience, and a high-precision EEG acquisition system used to simultaneously acquire SSVEP signals and motor imagery EEG signals.

[0084] 2) Stimulus generation system: Integrated into the VR device, it generates visual stimuli of different frequencies on the left and right sides.

[0085] 3) Stimulus design: Design different visual stimuli on the left and right sides of the VR environment, such as flashing light spots or graphics, and stimulate each side at a different fixed frequency (e.g., 10Hz on the left and 15Hz on the right).

[0086] Ensure that the stimulus is comfortable and friendly, and avoid causing visual fatigue or inducing epileptiform brain activity.

[0087] The specific process is as follows:

[0088] 1) The patient puts on VR equipment and EEG acquisition equipment and enters a preset rehabilitation scenario;

[0089] 2) After the system is started, visual stimulation is applied and the patient is prompted to perform motor imagery (e.g., "Please imagine moving your left lower limb").

[0090] 3) The EEG device collects the patient's SSVEP signal and motor imagery EEG signal in real time as EEG signal data.

[0091] The preprocessing module is used to preprocess the EEG signal data to obtain processed EEG signal data, estimate the limb posture of the limb movement data to obtain limb orientation data, send the processed EEG signal data to the intention recognition module, and send the limb orientation data to the control command module.

[0092] The preprocessing module performs preprocessing on EEG signal data, including denoising and filtering.

[0093] The process of estimating limb posture from limb motion data is as follows:

[0094] An improved complementary filtering method is used to fuse limb motion data to obtain a fusion coefficient, and the fusion coefficient is dynamically adjusted according to the deviation of acceleration and local geomagnetism; thus realizing the dynamic fusion of angular velocity integral and vector observation.

[0095] The intent recognition module is used to input the processed EEG signal data into a pre-established brain-like EEG signal recognition network model, output the patient's EEG analysis results, and send them to the control command module.

[0096] like Figure 10 As shown, the pre-established brain-like EEG signal recognition network model of the intent recognition module includes an encoding module, a residual module, a pulse connection module, a pulse fully connected module, and a decoding module. The pulse encoding module converts EEG signals into pulse signals using Poisson coding, enabling the pulse signals to transmit information within the network. The residual module consists of convolutional layers and pulse firing layers, serving as a key layer for pulse information feature extraction. The pulse connection module consists of pulse firing layers and pooling layers, reducing the network structure and thus the number of training parameters. The pulse fully connected module transforms the two-dimensional pulse network into a one-dimensional pulse network and outputs the classification pulse results. The decoding module transforms the pulse classification results into actual classification actions.

[0097] The processing procedure of the pre-established brain-like EEG signal recognition network model is as follows:

[0098] The SSVEP signal was analyzed to identify the frequency components corresponding to the patient's gaze side;

[0099] Feature extraction and classification of EEG signals related to motor imagery can identify the patient's motor intentions;

[0100] The frequency components corresponding to the patient's gaze side are fused with the patient's movement intention, and a judgment is made based on a preset decision algorithm to obtain the patient's EEG analysis results.

[0101] Specifically, in this embodiment, the brain-like electroencephalogram (EEG) signal recognition network is as follows:

[0102] Brain-inspired model algorithms for recognizing electroencephalogram (EEG) signals are computational methods that draw inspiration from the information processing mechanisms and structures of the human brain. They aim to simulate the high efficiency, low power consumption, and adaptability of biological neural networks to achieve more accurate recognition of EEG signals. The following is an introduction to this type of algorithm:

[0103] 1) Algorithm basics and principles.

[0104] a. Brain-inspired computing concept: Brain-inspired computing, also known as neuromorphic computing, is a general term for computing theories, architectures, chip designs, and application models and algorithms that draw on the information processing patterns and structures of biological nervous systems; its core lies in imitating the connection, communication, and information processing methods of neurons and synapses in the brain.

[0105] b. Neuron Models: Brain-inspired model algorithms are typically based on neuron models such as spiking neural networks (SNNs), which attempt to replicate the dynamic behavior of brain neurons in receiving, processing, and transmitting information. For example, the Hodgkin-Huxley (HH) model is a classic neuron model that details the mechanism of neuronal action potential generation. In recent years, brain-inspired neuron models based on endogenous complexity, such as those simulating the HH model using time-varying parameter LIF neurons (tv-LIF), have further improved computational efficiency and resource utilization.

[0106] c. Information processing method: The brain-like model algorithm adopts an event-driven and pulse coding information processing method, which is significantly different from the traditional computing method based on the von Neumann architecture. In the event-driven mode, the computing unit is only activated and processed when it receives a specific event (such as a neuronal pulse), thereby greatly reducing energy consumption and latency.

[0107] 2) Algorithm characteristics and advantages.

[0108] a. High efficiency: By simulating the large-scale parallel processing capabilities of biological neural networks, the brain-like model algorithm can exhibit extremely high efficiency when processing complex and ever-changing information. This parallel processing capability enables the algorithm to complete a large number of computational tasks in a short time, making it suitable for EEG signal recognition scenarios with high real-time requirements.

[0109] b. Low power consumption: The neuromorphic computing model draws on the brain's low-power information processing mechanism, significantly reducing energy consumption during the computing process through event-driven and pulse coding methods. This is particularly important for the application of portable EEG devices, helping to extend device usage time and improve user experience.

[0110] c. Adaptability: The brain-like model algorithm has a strong self-learning ability. It can continuously optimize the model parameters and structure through training on a large amount of EEG data, thereby improving the recognition accuracy and generalization ability for specific tasks. This adaptability enables the algorithm to cope with individual differences and changes in the condition of different patients, providing more personalized support for the rehabilitation training of stroke hemiplegic patients.

[0111] d. Robustness: The structure and mechanism of the brain-like computing model have high flexibility and plasticity, and can tolerate the influence of noise and interference signals to a certain extent, thereby improving the robustness of EEG signal recognition. This is of great significance for improving the stability and reliability of the system in practical application environments.

[0112] The process of estimating the orientation of human limbs based on IMU is as follows:

[0113] As the foundation for estimating human kinematics information, high dynamic accuracy and hyperparameter robustness in estimating human limb orientation are crucial for high-precision estimation of human joint angles. Human limb orientation estimation is the first step in human joint angle estimation based on inertial measurement units (IMUs). Its task is to establish the rotational relationship between the sensor coordinate system fixed to the IMU and the inertial frame, i.e., the absolute attitude of the IMU. This allows IMU measurements expressed in different sensor coordinate systems to be expressed in a common coordinate space, thus laying the groundwork for subsequent descriptions of human kinematic constraints. Current IMU-based limb posture estimation algorithms lack the accuracy required for further application at human motion frequencies. This invention proposes a limb posture estimation algorithm based on improved complementary filtering, which can automatically adjust the fusion coefficients and achieves state-of-the-art accuracy under both static and dynamic conditions.

[0114] Specifically, in this invention, the measurement values ​​of the nine-axis IMU, namely the signal models of angular velocity, acceleration, and local geomagnetism, and their correspondence with known quantities in the inertial frame, are used to form a problem description. Based on the problem description, a limb attitude estimation algorithm that integrates the measurement values ​​of the nine-axis IMU (gyroscope, accelerometer, and magnetometer) is proposed. The fusion coefficient is adaptively adjusted based on the absolute attitude estimation deviation, thereby forming two forms: adaptive complementary filtering and adaptive Kalman filtering. The algorithm flow in the attitude tracking process is described.

[0115] Human limb orientation estimation based on improved complementary filtering:

[0116] Let q be the quaternion obtained by integrating the angular velocity. gyro Let q be the quaternion obtained from vector observation. gyro According to complementary filtering, these two quaternions can be fused by weighted averaging of fusion coefficients.

[0117] Traditional complementary filtering methods rely on algorithm priors or manual adjustment through trial and error to pre-set the value of the fusion coefficient. However, the value of the fusion coefficient involves a trade-off between various factors such as baseline drift of angular velocity integral, local magnetic field distortion, and acceleration generated by motion, and needs to be dynamically adjusted according to changes in measurement and environment. This invention proposes an improved complementary filtering method with dynamic adjustment of the fusion coefficient.

[0118] In summary, the improved complementary filtering method can dynamically adjust the fusion coefficient based on acceleration and local geomagnetic deviations, thereby achieving dynamic fusion of angular velocity integrals and vector observations. This effectively estimates the current position of the limb and feeds it back to the control unit. Combined with EEG analysis results, it enables motion control of the flexible lower limb exoskeleton. The entire algorithm flow is as follows: Figure 11 As shown.

[0119] The control command module is used to generate joint control commands for the patient based on the patient's EEG analysis results and limb orientation data, and send them to the execution module.

[0120] The execution module is used to control the flexible rehabilitation robot to perform corresponding joint rehabilitation actions based on the patient's joint control commands.

[0121] The execution module includes: executing corresponding joint rehabilitation actions through three-loop closed-loop control of current, speed, and position. Position feedback is obtained through an encoder, speed feedback is obtained through position feedback differentiation, and current feedback is obtained through current acquisition. During position closed-loop control, the position trajectory generator provides a position reference according to the set position trajectory in a time-division manner. The position reference is compared with the position feedback and adjusted by the position loop controller to obtain the speed reference. The speed reference is compared with the speed feedback and adjusted by the speed loop controller to obtain the current reference. Finally, the current reference is compared with the current feedback and adjusted by the current loop to obtain the three-phase voltage waveform controlling the PMSM.

[0122] The three-loop closed-loop control of the execution module is implemented based on the permanent magnet synchronous motor control algorithm of field-oriented control to achieve high-precision control.

[0123] Specifically, the present invention will be further illustrated below through embodiments:

[0124] The hardware control process of this application is as follows:

[0125] This invention uses the high-performance application processor chip RK3588 from Rockchip as the main part of the control unit. This chip adopts an advanced ARM architecture and integrates a variety of high-performance processing units, mainly targeting ARMPC, edge computing, personal mobile Internet devices and other multimedia product applications.

[0126] Processor architecture: The RK3588 adopts ARM's big.LITTLE architecture, which combines four high-performance Cortex-A76 cores and four high-efficiency Cortex-A55 cores. The A76 cores can reach a maximum frequency of 2.4GHz, which is suitable for computationally intensive tasks. The A55 cores have lower power consumption characteristics and are suitable for low-load scenarios, thus achieving a balance between performance and energy efficiency.

[0127] Memory Interface: The RK3588 integrates a quad-channel LPDDR4 / LPDDR4X / LPDDR5 external memory interface, supporting memory frequencies up to 1066MHz (LPDDR4) and 3200MHz (LPDDR5), providing up to 34GB / s of memory bandwidth. This configuration not only meets the needs of high-performance computing and multitasking, but also provides ample bandwidth support for high-resolution video processing and complex image algorithms. Built-in NPU with 6TOPS computing power, a tri-core architecture, and support for int4 / int8 / int16 / FP16 / BF16 / TF32.

[0128] like Figure 12 The diagram shows the functional framework of the control unit, which can deploy the brain-like analysis model ReC-SNN proposed in this invention and can quickly infer the EEG signal recognition results.

[0129] Driver control hardware and software design:

[0130] The following is an overview of the actuation control of the flexible exoskeleton:

[0131] Exoskeletons require precise control of the motion curves of the lower limb joints, necessitating position servo control. High-performance position servo control systems typically consist of a three-loop closed-loop control system encompassing current, speed, and position. Figure 13 The diagram shows a three-loop closed-loop structure for position servo control.

[0132] Position feedback is obtained through an encoder, speed feedback is obtained through position feedback differentiation, and current feedback is obtained through current acquisition. During position closed-loop control, the position trajectory generator provides a position reference according to the set position trajectory in a time-division manner. The position reference is compared with the position feedback and adjusted by the position loop controller to obtain the speed reference. The speed reference is compared with the speed feedback and adjusted by the speed loop controller to obtain the current reference. Finally, the current reference is compared with the current feedback and adjusted by the current loop to obtain the three-phase voltage waveform controlling the PMSM.

[0133] The invention employs a high-performance coreless brushless servo motor. This type of motor uses coreless armature technology, eliminating iron loss and eddy current loss, and uses high-performance neodymium iron boron magnets, resulting in ultra-high power density and energy utilization. Its special structure also makes the armature inductance and resistance much smaller than those of ordinary motors. The back electromotive force of the coreless motor is optimized into a sine wave, so its analysis can be equivalent to that of a permanent magnet synchronous motor. In the control of permanent magnet synchronous motors, a field-oriented control method is often used to achieve high-precision position control.

[0134] like Figure 14 The key points of exoskeleton drive technology are shown. Addressing the precise torque control required in exoskeleton control, this invention provides an in-depth analysis of the current loop model of the servo driver, including the parameter design and current sampling strategy of the current loop. Secondly, it analyzes the modeling in the speed control mode, presenting the speed loop parameters and the encoder speed sampling strategy with limited resolution. Thirdly, to achieve precise human intention tracking of the exoskeleton, this invention implements a compliant position control strategy. Finally, to prevent control step loss during exoskeleton power-on positioning, the problem of identifying the initial electrical angle position of the permanent magnet synchronous motor is analyzed.

[0135] The rehabilitation strategy of the brain-inspired intelligent EEG flexible rehabilitation robot is as follows:

[0136] Overall control methods and rehabilitation strategies:

[0137] Based on the aforementioned hardware and control systems of the flexible robot, wearable robot technology is developed. This invention primarily uses pose data for trajectory tracking, and the control strategy does not rely on the motion model of the human lower limbs or the system model of the flexible robot. Therefore, for trajectory tracking, only the target pose information and the current pose information of the controlled object need to be acquired to complete the tracking control. The target pose information can be real-time gait data or a specified gait pattern. The control method and strategy are as follows: Figure 15 The overall control method and strategy diagram is shown.

[0138] Design of a progressive rehabilitation strategy that combines active and passive approaches:

[0139] Treatment and rehabilitation for lower limb paralysis in stroke patients typically include medication, stimulation, and motor rehabilitation. This invention primarily utilizes a brain-like intelligent EEG flexible rehabilitation robot for motor rehabilitation, employing a progressive intelligent rehabilitation strategy that combines passive and active approaches. Passive rehabilitation training is mainly for patients in the early stages of rehabilitation, as their joints have high resistance torque and poor motor function, requiring the robot to traction them for appropriate training. Once patients regain some motor function, the rehabilitation training can be upgraded to an active mode. This active mode, based on multi-sensor control, encourages patients to actively participate in rehabilitation training, increasing their motivation and promoting muscle strength and motor function recovery. Based on the advice of rehabilitation physicians and considering the motor limitations of hemiplegic patients—excessive range of motion training could cause secondary injury—this invention designs different modes and parameters for the main rehabilitation training, which can be set by rehabilitation physicians or therapists as needed. The main rehabilitation strategies are as follows: Figure 16 The rehabilitation strategy framework is shown below.

[0140] Based on the same inventive concept, this invention also provides a computer device, comprising: one or more processors, and a memory for storing one or more computer programs; the programs include program instructions, and the processor executes the program instructions stored in the memory. The processor may be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. It is the computing and control core of the terminal, used to implement one or more instructions, specifically for loading and executing one or more instructions stored in a computer storage medium to implement the above-described method.

[0141] It should be further explained that, based on the same inventive concept, the present invention also provides a computer storage medium storing a computer program, which, when executed by a processor, performs the above-described method. This storage medium can be any combination of one or more computer-readable media. The computer-readable medium can be a computer-readable signal medium or a computer-readable storage medium. The computer-readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of computer-readable storage media (a non-exhaustive list) include: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In the present invention, the computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0142] In the description of this specification, references to terms such as "an embodiment," "example," "specific example," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this disclosure. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0143] The foregoing has shown and described the basic principles, main features, and advantages of this disclosure. Those skilled in the art should understand that this disclosure is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of this disclosure. Various changes and modifications can be made to this disclosure without departing from its spirit and scope, and all such changes and modifications fall within the scope of this disclosure as claimed.

Claims

1. A brain-like intelligent EEG flexible rehabilitation robot, characterized in that, include: The data acquisition module is used to acquire electroencephalogram (EEG) signal data and limb movement data, and send them to the preprocessing module. The EEG signal data is acquired based on the EEG acquisition unit, and the limb movement data is acquired based on the inertial measurement unit (IMU). The preprocessing module is used to preprocess the EEG signal data to obtain processed EEG signal data, estimate the limb posture of the limb movement data to obtain limb orientation data, and send the processed EEG signal data to the intention recognition module and the limb orientation data to the control command module. The intent recognition module is used to input the processed EEG signal data into a pre-established brain-like EEG signal recognition network model, output the patient's EEG analysis results, and send them to the control command module. The control command module is used to generate joint control commands for the patient based on the patient's EEG analysis results and limb orientation data, and send them to the execution module; The execution module is used to control the flexible rehabilitation robot to perform corresponding joint rehabilitation actions based on the patient's joint control commands.

2. The brain-like intelligent EEG flexible rehabilitation robot according to claim 1, characterized in that, The EEG signal data includes SSVEP signals and motor imagery EEG signals; Limb motion data includes: angular velocity data, acceleration data, and local geomagnetic data.

3. The brain-like intelligent EEG flexible rehabilitation robot according to claim 1, characterized in that, The preprocessing module preprocesses the EEG signal data by performing noise reduction and filtering. The process of estimating limb posture from limb motion data is as follows: An improved complementary filtering method is used to fuse limb motion data to obtain a fusion coefficient, and the fusion coefficient is dynamically adjusted according to the deviation of acceleration and local geomagnetism; thus realizing the dynamic fusion of angular velocity integral and vector observation.

4. The brain-like intelligent EEG flexible rehabilitation robot according to claim 1, characterized in that, The pre-established brain-like EEG signal recognition network model of the intent recognition module includes an encoding module, a residual module, a pulse connection module, a pulse fully connected module, and a decoding module. The pulse encoding module converts EEG signals into pulse signals using Poisson coding, enabling the pulse signals to transmit information within the network. The residual module consists of convolutional layers and pulse firing layers for pulse information feature extraction. The pulse connection module consists of pulse firing layers and pooling layers. The pulse fully connected module transforms the two-dimensional pulse network into a one-dimensional pulse network and outputs the classification pulse results. The decoding module transforms the pulse classification results into actual classification actions.

5. The brain-like intelligent EEG flexible rehabilitation robot according to claim 4, characterized in that, The processing procedure of the pre-established brain-like electroencephalogram (EEG) signal recognition network model is as follows: The SSVEP signal was analyzed to identify the frequency components corresponding to the patient's gaze side; Feature extraction and classification of EEG signals related to motor imagery can identify the patient's motor intentions; The frequency components corresponding to the patient's gaze side are fused with the patient's movement intention, and a judgment is made based on a preset decision algorithm to obtain the patient's EEG analysis results.

6. The brain-like intelligent EEG flexible rehabilitation robot according to claim 1, characterized in that, The execution module includes three-loop closed-loop control of current, speed, and position to execute corresponding joint rehabilitation actions. Position feedback is obtained through an encoder, speed feedback is obtained through position feedback differentiation, and current feedback is obtained through current acquisition. During position closed-loop control, the position trajectory generator provides a position reference according to the set position trajectory in a time-division manner. The position reference is compared with the position feedback and adjusted by the position loop controller to obtain the speed reference. The speed reference is compared with the speed feedback and adjusted by the speed loop controller to obtain the current reference. Finally, the current reference is compared with the current feedback and adjusted by the current loop to obtain the three-phase voltage waveform controlling the PMSM.

7. The brain-like intelligent EEG flexible rehabilitation robot according to claim 6, characterized in that, The three-loop closed-loop control of the execution module is implemented based on the permanent magnet synchronous motor control algorithm of field-oriented control to achieve high-precision control.

Images