Exoskeleton robot waist assisting control method and system

Abstract

This invention relates to the field of exoskeleton robot technology and discloses a method and system for controlling the lumbar support of an exoskeleton robot. The method and system include: constructing an exoskeleton robot system comprising: at least one inertial measurement unit (IMU) located at the lower back or hip joint of the human body for measuring the posture angles and angular velocities of the lower back; at least two encoders located within the left and right hip joint drive units of the exoskeleton robot for measuring hip joint angles and angular velocities; at least one pressure sensor located on the sole of the human foot for measuring the plantar pressure value; and at least one electromyography (EMG) sensor located on the arm or upper arm. This invention achieves accurate recognition and adaptive assistance of the human lumbar movement intention through dynamic weighted fusion decision-making of multi-source sensor information, effectively improving the response accuracy and control intelligence of the exoskeleton robot in complex work scenarios.

Description

Technical Field

[0001] This invention relates to the field of exoskeleton robot technology, and in particular to a method and system for controlling the waist assistance of an exoskeleton robot. Background Technology

[0002] Currently, exoskeleton robots have shown significant application potential in fields such as industry, medicine, and rehabilitation, especially in lumbar support, which aims to reduce the burden on the lumbar spine of workers and reduce the risk of strain injuries.

[0003] However, existing control methods for lumbar assist exoskeletons mostly rely on information from a single or few sensors for intent recognition and control. This makes it difficult to accurately distinguish between different movement stages and true intentions of the human body during complex tasks (such as bending over to carry, standing up, and walking), which can easily lead to problems such as delayed assistance response, abrupt mode switching, and poor human-machine coupling.

[0004] Meanwhile, existing systems often lack personalized adaptive capabilities for different user body types, work habits, and load conditions, affecting the comfort, naturalness, and overall efficiency of the assistive device. Furthermore, existing technologies (CN110802603A) suffer from inaccurate motion intent recognition and poor scene adaptability due to the reliance on single sensor information (only electromyographic signals and encoders). Summary of the Invention

[0005] This invention provides a waist-assisted control method and system for exoskeleton robots to solve existing technical problems, thereby addressing the difficulty of accurately distinguishing human bodies for complex tasks in existing waist-assisted exoskeletons.

[0006] To solve the above-mentioned technical problems, according to one aspect of the present invention, more specifically, a waist-assist control method for an exoskeleton robot, comprising the following steps:

[0007] S1. Construct an exoskeleton robot system, which includes:

[0008] At least one inertial measurement unit (IMU) is installed at the lower back or hip joint of the human body to measure the attitude angle and angular velocity of the lower back.

[0009] At least two encoders are installed in the left and right hip joint drive units of the exoskeleton robot to measure hip joint angle and angular velocity;

[0010] At least one pressure sensor is installed on the sole of the human foot to measure the pressure value of the human foot.

[0011] At least one electromyography (EMG) sensor is installed on the human arm or upper arm to collect muscle electrical signals;

[0012] A main controller is used to receive and process all sensor data;

[0013] Left and right hip joint drive motors are used to output assist torque according to the instructions of the main controller;

[0014] S2. The main controller runs a dynamic weighted adaptive fusion decision-maker, which performs the following process:

[0015] S201. Real-time acquisition of data from the inertial measurement unit, encoder, pressure sensor, and electromyography sensor;

[0016] S202. Based on the sensor data, calculate and identify the current human movement state in real time as one of normal standing, flexion following, assist preparation, or full assistance;

[0017] S203. Under the identified current motion state, calculate a dynamic confidence weight for the electromyography signal, encoder phase information, and inertial measurement unit angle information according to the preset dynamic weight allocation strategy.

[0018] S204. Based on the dynamic confidence weight, the motion intention information from different sensors is fused to generate a comprehensive expected assist trajectory.

[0019] S205. Based on the comprehensive desired assist trajectory, calculate the control torque required for the left and right hip joint drive motors;

[0020] S3. The drive motor executes the control torque to provide assistance to the wearer's waist.

[0021] Furthermore, in step S202, identifying the current human motion state specifically includes the following steps:

[0022] 1) Calculate the flexion angle of the human lower back relative to the initial upright position using data from the inertial measurement unit. and extension angular velocity ;

[0023] 2) Calculate the current pressure value using a plantar pressure sensor. Compared to the previous moment's pressure value The difference ;

[0024] 3) Determine muscle activation level E using electromyography (EMG) sensor signals;

[0025] 4) Define and calculate a current task stage index S, which is about... , , The function is used to quantify the specific stage in the current work cycle from bending over to standing up;

[0026] 5) Based on the value of the current work stage index S, and in conjunction with the angle... angular velocity Pressure difference The current movement state is determined by comparing the threshold of electromyography level E.

[0027] Furthermore, the current operation stage index S is calculated using the following formula:

[0028] ;

[0029] In the above formula, , , The preset scaling factor, and ; , These are the normalization coefficients; The reference pressure value is denoted as sigmoid and tanh, which are non-linear activation functions. The range of the exponent S is mapped to the interval [0, 1], and the larger the value, the closer it is to the stage of getting up and exerting force.

[0030] Furthermore, the dynamic confidence weight allocation strategy in step S203 is specifically as follows:

[0031] Electromyographic signal weight This indicates that a higher weight is assigned when the index S is higher in the work phase. The calculation formula is:

[0032] ;

[0033] in, Preset gain; The exponential threshold for the effectiveness of electromyographic weighting;

[0034] Encoder phase information weights This indicates that a dominant weight is assigned when a periodic walking state is detected, and its value is negatively correlated with the stability of the phase difference between the left and right legs.

[0035] Inertial Measurement Unit Angular Information Weights This indicates that a basic weight is provided throughout the entire operation, and it serves as the primary basis for state judgment when the S value is low.

[0036] The dynamic confidence weights are normalized in each control loop, such that:

[0037] .

[0038] Furthermore, prior to step S2, there is an adaptive learning phase for upper-level parameters:

[0039] When the system is used for the first time or when the wearer changes, guide the wearer to complete several standard bending and lifting movements;

[0040] Record the angle of the inertial measurement unit at the lowest point of bending over during each movement. And the maximum pressure of the soles of the feet before and after grasping an object. and Therefore, according to the formula:

[0041] ;

[0042] Computational task adaptive gain ;in, Based on the gain, For reference angle; It serves as a global scaling factor for the torque output in subsequent control.

[0043] Furthermore, in step S204, a comprehensive expected assist trajectory is generated, specifically as follows:

[0044] Target assist torque under full assist condition Calculated by the following formula:

[0045] ;

[0046] in, It is the expected torque component based on the mapping of electromyographic signal E; Based on the target perspective From the current perspective The impedance control component of the deviation; It is based on the gait phase calculated by the encoder. The generated periodic boost component is generated using an oscillator model; Adaptive gain for the task.

[0047] Furthermore, the motion states identified in step S202 constitute a mid-level finite state machine, and the transition rules between each state are as follows:

[0048] Normal standing indicates the angle of lumbar and back flexion. This state is maintained when the pressure is below the first threshold and there is no significant change in plantar pressure; if... If the first threshold is exceeded, the system transitions to a buckling-following state.

[0049] The flexion-following state refers to the state in which the exoskeleton motors adopt a position-following mode, following the movement of the human waist with near-zero impedance; if a change in plantar pressure is detected... If the pressure threshold is exceeded and electromyographic signals are activated, the system will switch to the assisted preparation state.

[0050] The assist-ready state refers to the state in which the motor outputs a small, preset base torque to eliminate system backlash and indicate the start of assist; if the lumbar and back extension angular velocity is detected... If the speed threshold is exceeded, the system will switch to full power assist mode.

[0051] Fully assisted state refers to the state in which the target assist torque is applied. The torque calculation formula outputs assistance; when the angle of the lower back... When you return to a near-upright position, return to a normal standing position.

[0052] Furthermore, the calculation of the control torque in step S205 and the execution of the drive motor in step S3 are completed by the underlying force-position hybrid controller:

[0053] The lower-level controller receives instructions from the middle-level finite state machine and the target torque calculated by the upper-level fusion decision-maker. Output torque of the underlying controller This can be achieved through the following formula:

[0054] ;

[0055] in, , The stiffness and damping coefficients are configured according to the different states of the intermediate state machine; , For the target position and velocity; , The encoder provides feedback on the current position and velocity;

[0056] In different motion states, by , Set to zero or assign a specific value, and combine with feedforward terms. It enables seamless switching between pure position servo, pure force control, and force-position hybrid control.

[0057] A waist-assist control system for an exoskeleton robot includes:

[0058] The sensing module includes:

[0059] At least one inertial measurement unit (IMU) is configured to be worn on the lower back or hip joint of a human body to measure the attitude angle and angular velocity of the lower back.

[0060] At least two encoders are built into the left and right hip joint drive units of the exoskeleton robot, respectively, to measure the rotation angle and angular velocity of the corresponding hip joints;

[0061] At least one plantar pressure sensor is configured to be placed on the sole of the human foot for measuring plantar pressure values.

[0062] At least one electromyography (EMG) sensor is configured to be located in the human arm or upper arm for collecting muscle electrical signals;

[0063] The processing and control module includes a main controller, which is communicatively connected to all sensors in the sensing module, and is used for:

[0064] Receive and process real-time data from the sensing module;

[0065] Run a dynamic weighted adaptive fusion decision engine to perform the control method described above, including state recognition, dynamic weight calculation, information fusion and control torque calculation;

[0066] Generate control commands;

[0067] The drive execution module includes drive motors corresponding to the left and right hip joints respectively. The drive motors are electrically connected to the processing and control module and are used to receive and output assist torque according to the control commands issued by the main controller.

[0068] The structural support and binding module is used to integrate the sensing module, processing and control module and drive execution module and wear them on the human body to ensure the accuracy of the sensor measurement points and the alignment of the drive execution module with the human hip joint movement axis.

[0069] The present invention provides a waist-assist control method and system for an exoskeleton robot. Compared with the prior art, the advantages achieved by this method are as follows:

[0070] ① This invention achieves accurate recognition and adaptive assistance of human waist movement intentions through dynamic weight fusion decision-making of multi-source sensor information, effectively improving the response accuracy and control intelligence of exoskeleton robots in complex work scenarios;

[0071] ② By introducing an adaptive learning mechanism for upper-level parameters, the system can automatically adjust the assist gain according to the user's individual characteristics and task load, thereby providing personalized assist support and enhancing the applicability and comfort of the exoskeleton in different users and tasks.

[0072] ③ This invention adopts a hierarchical control architecture, combining a middle-level finite state machine with a bottom-level force-position hybrid controller, to achieve smooth switching between different motion states and seamless connection of multi-mode control, ensuring the natural continuity of the assistance process and the smoothness of human-machine collaboration.

[0073] ④ By integrating multiple control components such as electromyographic feedforward, impedance feedback and gait oscillation, the system can provide timely and appropriate assistance in various movement modes such as bending over to carry, standing up to exert force and walking, which not only reduces the burden on the human waist, but also maintains the freedom and coordination of movement. Attached Figure Description

[0074] Figure 1 This is a flowchart of the present invention.

[0075] Figure 2 This is a graph showing the relationship between the current work stage index and the bending angle in this invention.

[0076] Figure 3 This is a graph showing the relationship between electromyographic signal weights and the work phase index in this invention.

[0077] Figure 4 This is a graph showing the target assist torque and its components changing over time in this invention.

[0078] Figure 5 This is a graph showing the change of angle deviation over time in this invention.

[0079] Figure 6 This is a diagram showing the relationship between the waist angle and angular velocity during real-time motion in this invention. Detailed Implementation

[0080] To make the technical solution of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0081] Example 1

[0082] like Figure 1 As shown, according to one aspect of the present invention, a method for waist assistance control of an exoskeleton robot is provided, comprising the following steps:

[0083] Step 1: Construct an exoskeleton robot system, which includes:

[0084] At least one inertial measurement unit (IMU) is installed at the lower back or hip joint of the human body to measure the attitude angle and angular velocity of the lower back.

[0085] At least two encoders are installed in the left and right hip joint drive units of the exoskeleton robot to measure hip joint angle and angular velocity;

[0086] At least one pressure sensor is installed on the sole of the human foot to measure the pressure value of the human foot.

[0087] At least one electromyography (EMG) sensor is installed on the human arm or upper arm to collect muscle electrical signals;

[0088] A main controller is used to receive and process all sensor data;

[0089] Left and right hip joint drive motors are used to output assist torque according to the instructions of the main controller;

[0090] Step 2: The main controller runs a dynamic weighted adaptive fusion decision-maker, which performs the following process:

[0091] S201. Real-time acquisition of data from the inertial measurement unit, encoder, pressure sensor, and electromyography sensor;

[0092] S202. Based on sensor data, calculate and identify the current human motion state in real time as one of the following: normal standing, flexion following, assisted preparation, or full assistance.

[0093] S203. Under the identified current motion state, calculate a dynamic confidence weight for the electromyography signal, encoder phase information, and inertial measurement unit angle information according to the preset dynamic weight allocation strategy.

[0094] S204. Based on dynamic confidence weights, motion intention information from different sensors is fused to generate a comprehensive expected assist trajectory.

[0095] S205. Based on the comprehensive expected assist trajectory, calculate the control torque required for the left and right hip joint drive motors;

[0096] Step 3: The drive motor applies control torque to provide assistance to the wearer's waist.

[0097] Before the algorithm is executed, the multi-source sensor data needs to be preprocessed to ensure data quality.

[0098] Inertial Measurement Unit (IMU): Employs quaternion-based complementary filters or Kalman filters, fusing data from triaxial accelerometers and gyroscopes to calculate the pitch angle of the human lower back in the sagittal plane (i.e., the flexion / extension plane). and its angular velocity The initial upright posture is defined as... =0°.

[0099] The encoder is a built-in encoder in the motor. The absolute angle of the hip joint is obtained by converting the readings, and the angular velocity is calculated by differential calculation. The left and right hip joint angles are used to calculate the gait phase difference.

[0100] The plantar pressure sensor collects total pressure values, which are then low-pass filtered to remove high-frequency noise. Pressure difference. =F t -F t-1 Used to detect instantaneous events of heavy object grabbing / releasing.

[0101] Electromyography (sEMG) sensors obtain an envelope signal E that characterizes the level of muscle activation by passing the raw signal through bandpass filtering (e.g., 20-500Hz), full-wave rectification, and smoothing (e.g., moving average or low-pass filtering).

[0102] In this embodiment, step S202, identifying the current human motion state, specifically includes the following steps:

[0103] 1) Calculate the flexion angle of the human lower back relative to the initial upright position using data from the inertial measurement unit. and extension angular velocity ;

[0104] 2) Calculate the current pressure value F using a plantar pressure sensor. t Compared to the previous pressure value F t-1 The difference ;

[0105] 3) Determine muscle activation level E using electromyography (EMG) sensor signals;

[0106] 4) Define and calculate a current task stage index S, which is about... , , The function is used to quantify the specific stage in the current work cycle from bending over to standing up;

[0107] 5) Based on the value of the index S for the current work stage, and in conjunction with the angle... angular velocity Pressure difference The current movement state is determined by comparing the threshold of electromyography level E.

[0108] like Figure 2 As shown, the current work phase index S is calculated using the following formula:

[0109] ;

[0110] In the above formula, , , The preset scaling factor, and ; , These are the normalization coefficients; The reference pressure value is denoted as sigmoid and tanh, which are non-linear activation functions. The range of the exponent S is mapped to the interval [0, 1], and the larger the value, the closer it is to the stage of getting up and exerting force.

[0111] 1. Purpose and physical significance of the derivation:

[0112] The index S is a scalar designed to continuously and smoothly characterize the progress of one "bend-grab-stand" work cycle. Its range is designed to be [0, 1], where:

[0113] S≈0 corresponds to the initial stage of the task (standing upright or just beginning to bend over).

[0114] S≈0.5 may correspond to bending over to the lowest point and stably grasping the object.

[0115] S≈1 corresponds to the stage of getting up and exerting force.

[0116] 2. Detailed explanation and derivation of each component, angle component :

[0117] in, It is a monotonically increasing sigmoid function with an output range of (0, 1). This characteristic is ideal for calculating the bending angle. (From 0° to possibly above 90°) Non-linearly mapped onto a growth trend.

[0118] when When the sigmoid function is relatively small (at the beginning of bending over), its growth is gradual, exhibiting slow exponential growth in the initial stage; when... Exceed (For example, after 45°), the function enters the region of approximately linear growth; when When the angle is large, the function tends to saturate (close to 1), indicating that the contribution of the bending angle to the progress of the stage has reached its upper limit. The coefficient α weighs the importance of this component in the total exponent. The above formula intuitively reflects "how much bending occurred" and is the basis for judging the stage of the operation.

[0119] angular velocity components :

[0120] in, It is an odd function, with an output range of (-1, 1). Here, we take... .

[0121] Derivation: angular velocity It can be positive or negative (positive for forward flexion, negative for backward extension). The tanh function can smoothly compress the angular velocity value to the range (-1, 1), avoiding excessively large values ​​from impacting the exponent.

[0122] when >0 (bending over). A positive value slightly increases S, but the contribution is usually small (because bending over is usually slower).

[0123] when <0 (getting up) It is negative. The key point is that this term is additive in the formula. During the rising phase, although this term is negative, the angle components are still negative at this time. It is already at a high level, and the pressure component... This provides a significant positive contribution (see below), enough to drive the S-value towards 1. The negative value of this term effectively "suppresses" excessively high S-values ​​caused by the still large angle during rapid ascent, making the rise of the S-value more closely resemble the actual "force exertion phase." Introducing dynamic information about the direction and speed of movement makes phase judgment more accurate.

[0124] Pressure change component :

[0125] in, =F t -F t-1 middle, It is the resting foot pressure benchmark (the maximum pressure value measured before grasping an object) corresponding to the wearer's weight.

[0126] The moment the heavy object is lifted off the ground There will be a significant positive jump (the increased gravity is transmitted through the body to the soles of the feet). The ratio may reach 0.2-0.5 or higher (depending on the weight of the object).

[0127] This component directly and sensitively responds to the key event of "increased load." The coefficient γ controls the weight of this event. When the grabbing event occurs, this component will cause a step increase in the exponent S, which is the key signal driving S from "bending over" to "standing up and exerting force."

[0128] Directly detecting changes in external load is the strongest evidence for determining whether force has been applied to the material.

[0129] 3. Coefficient normalization ( ):

[0130] This constraint guarantees that the theoretical range of S is within the range of [-1, 2]. Furthermore, the actual output ranges of tanh and sigmoid, as well as... The typical value of ensures that S is effectively mapped to the [0,1] interval for subsequent judgment. It makes the three components form a weighted sum, reflecting the trust allocation among different information sources.

[0131] This embodiment constructs a lumbar assist control architecture based on multi-source information fusion. Through the collaborative acquisition of data from an inertial measurement unit, encoder, pressure sensor, and electromyography sensor, combined with a dynamic weighted adaptive fusion decision-maker, it achieves multi-dimensional and highly redundant perception of the human lumbar movement intention. Its advantage lies in overcoming the limitations of traditional single-sensor control. By complementing and verifying multi-source data, it significantly improves the accuracy and robustness of motion state recognition and assist trajectory generation, enabling the exoskeleton to provide assistance more intelligently and in a more realistic manner in complex tasks.

[0132] Example 2

[0133] like Figure 3 As shown, the dynamic confidence weight allocation strategy in step S203 is as follows:

[0134] Electromyographic signal weight This indicates that a higher weight is assigned when the index S is higher in the work phase. The calculation formula is:

[0135] ;

[0136] in, Preset gain; The exponential threshold for the effectiveness of electromyographic weighting;

[0137] Encoder phase information weights This indicates that a dominant weight is assigned when a periodic walking state is detected, and its value is negatively correlated with the stability of the phase difference between the left and right legs.

[0138] Inertial Measurement Unit Angular Information Weights This indicates that a basic weight is provided throughout the entire operation, and it serves as the primary basis for state judgment when the S value is low.

[0139] The dynamic confidence weights are normalized in each control loop, such that:

[0140] .

[0141] 1. The driving idea behind the weight allocation strategy:

[0142] Different sensors have varying reliability and information content under different motion scenarios. The core of this strategy is to dynamically adjust the level of trust in each sensor's information based on the identified motion state and operation stage.

[0143] 2. Derivation of weight calculations:

[0144] Electromyographic weight :

[0145] Among them, formula middle, It can be a linear function (such as max(0, min(1, kx+b))) or another sigmoid function. th It is a threshold close to 0.5 (e.g., 0.6).

[0146] Electromyographic (EMG) signals directly reflect the neural drive intentions of muscles, making them an ideal source for "feedforward" control. However, during the bending phase (low S), muscle activity may primarily be focused on maintaining posture, and signal noise is relatively high; therefore, their weight should be reduced at this stage. When entering the standing and exertion phase (S>S), the weighting should be adjusted accordingly. th Electromyographic signals are strong and have clear intent, so they should be given high weight to quickly respond to the body's force exertion intention. This formula realizes the function of adaptively increasing the weight with the stage exponent S.

[0147] Encoder phase weight This weight primarily serves the walking assistance mode.

[0148] First, the gait phase φ is calculated using left and right hip joint encoders, and the stability of the phase difference over several consecutive steps is evaluated (e.g., the variance is calculated).

[0149] When the system detects a periodic walking pattern (e.g., determined by periodic angle and speed) and the phase difference is stable, the gait phase information provided by the encoder is considered highly reliable and should be assigned [a certain value]. High values ​​(e.g., 0.7-0.9).

[0150] In simple bending and carrying tasks, both legs move synchronously with no significant gait phase difference. It should be set to a very low value (close to 0).

[0151] Its assignment logic can be implemented using a lookup table or a function based on periodic characteristic values.

[0152] IMU Angle Weights The IMU provides a global attitude reference and serves as a fundamental source of information in any state.

[0153] During the flexion-follow-up phase (bending over) (low S value), the body moves slowly, the electromyographic intent is not obvious, and there is no walking phase. The IMU should be the dominant factor, and the control system should primarily rely on the IMU's angle for position tracking.

[0154] As S increases and possible Increase, The weight of will be relatively reduced, but the angle deviation it provides ( () is always the basis of impedance control components.

[0155] A simple implementation is: =1- - And set a minimum value (such as 0.1) to ensure that basic information is not lost.

[0156] 3. Normalization:

[0157] , , The original calculated values ​​are normalized by dividing by their sum to ensure that the fusion is a convex combination and to avoid uncontrolled output torque.

[0158] Based on the identified motion state and task phase index, the weight ratio of electromyography (EMG), encoder phase, and IMU angle information is adaptively adjusted. Its advantage lies in its ability to optimize and fuse decisions in real time based on the reliability and information content of various sensors in different motion modes (such as bending, standing, and walking). This strengthens the EMG feedforward response during the standing-up phase, provides rhythmic assistance based on encoder phase during the walking phase, and achieves compliant movement based on IMU angle during the bending phase, enhancing the system's control adaptability and natural human-machine collaboration in diverse scenarios.

[0159] Example 3

[0160] Before step two, there is also an upper-level parameter adaptive learning stage:

[0161] When the system is used for the first time or when the wearer changes, guide the wearer to complete several standard bending and lifting movements;

[0162] Record the angle of the inertial measurement unit at the lowest point of bending over during each movement. And the maximum pressure of the soles of the feet before and after grasping an object. and Therefore, according to the formula:

[0163] ;

[0164] Computational task adaptive gain ;in, Based on the gain, For reference angle; It serves as a global scaling factor for the torque output in subsequent control.

[0165] like Figure 4 , Figure 5 As shown, in step S204, a comprehensive expected assist trajectory is generated, specifically as follows:

[0166] Target assist torque under full assist condition Calculated by the following formula: ;

[0167] in, It is the expected torque component based on the mapping of electromyographic signal E; Based on the target perspective From the current perspective The impedance control component of the deviation; It is based on the gait phase calculated by the encoder. The generated periodic boost component is generated using an oscillator model; Adaptive gain for the task.

[0168] 1. This formula is the output of the upper-level fusion decision-maker, reflecting the assistance of multi-source information fusion. It is a hybrid structure of feedforward and feedback.

[0169] 2. Detailed explanation of each component:

[0170] Electromyographic feedforward component , This is a function that maps the processed electromyographic envelope signal E to the desired torque. Linear or piecewise linear mapping is typically used. ,in For gain, This represents the baseline value for muscles at rest. More advanced mappings are available via neural networks. This component directly responds to the force commands from the central nervous system, aiming to provide "what you think, you get" assistance and reduce latency.

[0171] Impedance feedback component The derivation is as follows: g(x) is usually a PD (proportional-differential) function, specifically:

[0172] .

[0173] in This is the target angle (0° for upright). This constitutes a virtual spring-damped system. When the human body deviates from the target posture, this component generates a torque that pulls it back. It ensures the stability and accuracy of control, especially in the fine adjustments during the following and assist phases.

[0174] Gait oscillation component , This is generated by a rhythmic motion generator (such as a Hopf oscillator or a phase oscillator). After the oscillator is adaptively synchronized with the human gait frequency (estimated from the encoder speed), it outputs a torque curve that is in phase with the gait cycle, for example... =A*sin( ), where A is the amplitude, This is the internal phase of the oscillator (0~2π). This component provides periodic, efficient assistance during walking, conforming to human walking biomechanics.

[0175] 3. Dynamic weight fusion:

[0176] In a fully assisted state (lifted up). and Dominant. The torque is a weighted sum of electromyographic feedforward (rapid response) and impedance feedback (stable correction), achieving compliance and accuracy in the "force interaction" between humans and machines.

[0177] In the flexion-following state ≈1, ≈0, ≈0. At this time ≈ And K p It is set very small, so that the exoskeleton is like a lightly damped virtual spring, following the body's movements with almost no resistance.

[0178] While walking, Dominant ≈ It provides rhythmic walking assistance.

[0179] 4. Task Adaptive Gain :

[0180] Where the formula middle, To standardize the depth of bending over. The deeper the bend ( The larger the angle, the greater the gravitational torque that needs to be resisted during the standing process, thus requiring a greater boost.

[0181] Standardized load weight. The heavier the object being transported (F... max The larger F is ref The greater the pressure from the body's own weight, the greater the assistance required.

[0182] To pass this normalized product, It can be personalized to adapt to the different work habits (bending depth) of different wearers and the actual load of the task at hand, so as to achieve "assistance on demand" and avoid insufficient or excessive assistance.

[0183] By guiding users to complete standard movements and recording key characteristic parameters (such as maximum bending angle and changes in plantar pressure), the system automatically calculates the task-adaptive gain as a global torque scaling factor. Its advantage lies in achieving personalized assistance adjustment. The system can dynamically adjust the assistance level based on the wearer's body shape, work habits, and actual handling load, thus providing appropriate assistance under different user and task conditions, improving wearing comfort and work efficiency, and reducing problems of insufficient or excessive assistance caused by fixed parameters.

[0184] Example 4

[0185] like Figure 6 As shown, the motion states identified in step S202 constitute a mid-level finite state machine, and the transition rules between each state are as follows:

[0186] Normal standing indicates the angle of lumbar and back flexion. This state is maintained when the pressure is below the first threshold and there is no significant change in plantar pressure; if... If the first threshold is exceeded, the system transitions to a buckling-following state.

[0187] The flexion-following state refers to the state in which the exoskeleton motors adopt a position-following mode, following the movement of the human waist with near-zero impedance; if a change in plantar pressure is detected... If the pressure threshold is exceeded and electromyographic signals are activated, the system will switch to the assisted preparation state.

[0188] The assist-ready state refers to the state in which the motor outputs a small, preset base torque to eliminate system backlash and indicate the start of assist; if the lumbar and back extension angular velocity is detected... If the speed threshold is exceeded, the system will switch to full power assist mode.

[0189] Fully assisted state refers to the state in which the target assist torque is applied. The torque calculation formula outputs assistance; when the angle of the lower back... When the body returns to a near-upright position (below the second threshold), it returns to a normal standing position.

[0190] In this embodiment, the calculation of the control torque in step S205 and the execution of the drive motor in step three are completed by the underlying force-position hybrid controller:

[0191] The lower-level controller receives instructions from the middle-level finite state machine and the target torque calculated by the upper-level fusion decision-maker. Output torque of the underlying controller This can be achieved through the following formula:

[0192] ;

[0193] in, , The stiffness and damping coefficients are configured according to the different states of the intermediate state machine; , For the target position and velocity; , The encoder provides feedback on the current position and velocity;

[0194] In different motion states, by , Set to zero or assign a specific value, and combine with feedforward terms. This enables seamless switching between pure position servo control, pure force control, and force-position hybrid control.

[0195] 1. This formula is the command torque calculation formula for the bottom motor current loop, which is a feedforward-feedback composite controller.

[0196] 2. Derivation of each item and configuration under different states:

[0197] feedforward term That is, the target torque calculated by the upper layer carries the fused motion intention information.

[0198] 1) Location feedback item *( - ):

[0199] Set by the intermediate state machine. In the buckling follower state, = (Real-time angle of human body) Set to a smaller value (e.g., 0.1-1 Nm / deg) to achieve compliant motion.

[0200] In full-assisted mode = =0°, Using the calculations from the upper layer Related gain (potentially large), and impedance components Collaborative work enhances the ability to revert to previous positions.

[0201] In zero-force / follow-up mode, you can set... Current position If the value is 0, then this item is invalid.

[0202] 2) Speed ​​Feedback Item =( - ):

[0203] It is mainly used to increase system damping and suppress overshoot and oscillation. =0 (the desired speed is zero, i.e., there is no speed deviation). Adjust the system based on its inertia and response.

[0204] 3) Seamless transition during state switching:

[0205] During state transition, , , , and feedforward terms Synchronous and smooth switching (e.g., using a first-order low-pass filter for transition).

[0206] For example, shifting from "flexion-following" to "assistive preparation". From a follower's perspective Smooth transition to 0° Transitioning from a small value to 0 (in the assist preparation state, pure force control), =0), and at the same time The feedforward torque transitions from 0 to a small base torque t. This design avoids sudden torque changes during mode switching, ensuring wearing comfort and safety.

[0207] The mid-level control logic based on a finite state machine clarifies four states: normal standing, flexion follow-up, assist preparation, and full assistance, along with their transition rules based on multi-sensor thresholds. Its advantage lies in achieving structured management and smooth switching between motion modes through the state machine, avoiding abrupt changes and conflicts in the assistance process. This allows the exoskeleton to seamlessly transition between various control modes such as follow-up, preparation, and assistance according to the human body's movement stages, ensuring the continuity, safety, and user experience of the assistance process.

[0208] Example 5

[0209] A waist-assist control system for an exoskeleton robot includes:

[0210] The sensing module includes:

[0211] At least one inertial measurement unit (IMU) is configured to be worn on the lower back or hip joint of a human body to measure the posture angle and angular velocity of the lower back; at least two encoders are respectively built into the left and right hip joint drive units of the exoskeleton robot to measure the rotation angle and angular velocity of the corresponding hip joints; at least one plantar pressure sensor is configured to be placed on the sole of the human foot to measure the plantar pressure value; at least one electromyography (EMG) sensor is configured to be placed on the arm or upper arm of a human body to collect muscle electrical signals.

[0212] The processing and control module includes a main controller, which is communicatively connected to all sensors in the sensing module. The main controller is used to: receive and process real-time data from the sensing module; and run a dynamic weighted adaptive fusion decision-maker to execute the control methods described above, including state recognition, dynamic weight calculation, information fusion, and control torque calculation.

[0213] The system generates control commands; it also includes a drive execution module, which comprises drive motors corresponding to the left and right hip joints respectively. The drive motors are electrically connected to the processing and control module and are used to receive and output assist torque according to the control commands issued by the main controller; and a structural support and binding module, which is used to integrate the sensing module, processing and control module and drive execution module and wear them on the human body to ensure the accuracy of sensor measurement points and the alignment of the drive execution module with the human hip joint movement axis.

[0214] Integrating the aforementioned methods into a complete system implementation encompasses four major modules: sensing, processing and control, drive execution, and structural binding, forming a closed-loop control system from signal acquisition and decision fusion to torque output. Its advantages lie in providing a practically implementable system solution. Modular design ensures a rational sensor layout, centralized data processing, rapid execution response, and accurate human-machine alignment, thereby effectively transforming theoretical methods into practical engineering applications and providing systematic support for the development and application of exoskeleton lumbar support products.

[0215] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention. Therefore, the scope of protection of this patent should be determined by the appended claims.

Claims

1. A method for controlling the waist assistance of an exoskeleton robot, characterized in that, Includes the following steps: S1. Construct an exoskeleton robot system, which includes: At least one inertial measurement unit is installed at the human lower back or hip joint to measure the posture angle and angular velocity of the human lower back. At least two encoders are installed in the left and right hip joint drive units of the exoskeleton robot to measure hip joint angle and angular velocity; At least one pressure sensor is installed on the sole of the human foot to measure the pressure value of the human foot. At least one electromyography (EMG) sensor is installed on the human arm to collect muscle electrical signals; A main controller is used to receive and process all sensor data; Left and right hip joint drive motors are used to output assist torque according to the instructions of the main controller; S2. The main controller runs a dynamic weighted adaptive fusion decision-maker, which performs the following process: S201. Real-time acquisition of data from the inertial measurement unit, encoder, pressure sensor, and electromyography sensor; S202. Based on the sensor data, calculate and identify the current human movement state in real time as one of normal standing, flexion following, assist preparation, or full assistance; S203. Under the identified current motion state, calculate a dynamic confidence weight for the electromyography signal, encoder phase information, and inertial measurement unit angle information according to the preset dynamic weight allocation strategy. S204. Based on the dynamic confidence weight, the motion intention information from different sensors is fused to generate a comprehensive expected assist trajectory. S205. Based on the comprehensive desired assist trajectory, calculate the control torque required for the left and right hip joint drive motors; S3. The drive motor executes the control torque to provide assistance to the wearer's waist; In step S202, identifying the current human motion state specifically includes the following steps: 1) Calculate the flexion angle of the human lower back relative to the initial upright position using data from the inertial measurement unit. and extension angular velocity ; 2) Calculate the current pressure value using a plantar pressure sensor. Compared to the previous moment's pressure value The difference ; 3) Determine muscle activation level E using electromyography (EMG) sensor signals; 4) Define and calculate a current task stage index S, which is about... , , The function is used to quantify the specific stage in the current work cycle from bending over to standing up; 5) Based on the value of the current work stage index S, and in conjunction with the angle... angular velocity Pressure difference The current movement state is determined by comparing the threshold of electromyography level E. The current work phase index S is calculated using the following formula: ； In the above formula, , , The preset scaling factor, and ; , These are the normalization coefficients; The reference pressure value is denoted as sigmoid and tanh, which are non-linear activation functions. The range of the exponent S is mapped to the interval [0, 1], and the larger the value, the closer it is to the stage of getting up and exerting force.

2. The waist-assist control method for exoskeleton robots according to claim 1, characterized in that: The dynamic confidence weight allocation strategy in step S203 is as follows: Electromyographic signal weight This indicates that a higher weight is assigned when the index S is higher in the work phase. The calculation formula is: ； in, Preset gain; The exponential threshold for the effectiveness of electromyographic weighting; Encoder phase information weights This indicates that a dominant weight is assigned when a periodic walking state is detected, and its value is negatively correlated with the stability of the phase difference between the left and right legs. Inertial Measurement Unit Angular Information Weights This indicates that a basic weight is provided throughout the entire operation, and it serves as the primary basis for state judgment when the S value is low. The dynamic confidence weights are normalized in each control loop, such that: 。 3. The waist-assist control method for exoskeleton robots according to claim 1, characterized in that: Before step S2, there is also an upper-level parameter adaptive learning stage: When the system is used for the first time or when the wearer changes, guide the wearer to complete several standard bending and lifting movements; Record the angle of the inertial measurement unit at the lowest point of bending over during each movement. And the maximum pressure of the soles of the feet before and after grasping an object. and ; Therefore, according to the formula: ； Computational task adaptive gain ;in, Based on the gain, For reference angle; It serves as a global scaling factor for the torque output in subsequent control.

4. The waist-assist control method for exoskeleton robots according to claim 3, characterized in that: In step S204, a comprehensive expected assist trajectory is generated, specifically as follows: Target assist torque under full assist condition Calculated by the following formula: ； in, It is the expected torque component based on the mapping of electromyographic signal E; Based on the target perspective From the current perspective The impedance control component of the deviation; It is based on the gait phase calculated by the encoder. The generated periodic boost component is generated using an oscillator model; Adaptive gain for the task.

5. The waist-assist control method for exoskeleton robots according to claim 4, characterized in that: The motion states identified in step S202 constitute a mid-level finite state machine, and the transition rules between each state are as follows: Normal standing indicates the angle of lumbar and back flexion. This state is maintained when the pressure is below the first threshold and there is no significant change in plantar pressure; if... If the first threshold is exceeded, the system transitions to a buckling-following state. The flexion-following state refers to the state in which the exoskeleton motors adopt a position-following mode, following the movement of the human waist with near-zero impedance; if a change in plantar pressure is detected... If the pressure threshold is exceeded and electromyographic signals are activated, the system will switch to the assisted preparation state. The assist-ready state refers to the state in which the motor outputs a small, preset base torque to eliminate system backlash and indicate the start of assist; if the lumbar and back extension angular velocity is detected... If the speed threshold is exceeded, the system will switch to full power assist mode. Fully assisted state refers to the state in which the target assist torque is applied. The torque calculation formula outputs assistance; when the angle of the lower back... When you return to a near-upright position, return to a normal standing position.

6. The waist-assist control method for exoskeleton robots according to claim 1, characterized in that: The calculation of the control torque in step S205 and the execution of the drive motor in step S3 are completed by the underlying force-position hybrid controller. The lower-level controller receives instructions from the middle-level finite state machine and the target torque calculated by the upper-level fusion decision-maker. Output torque of the underlying controller This can be achieved through the following formula: ； in, , The stiffness and damping coefficients are configured according to the different states of the intermediate state machine; , For the target position and velocity; , The encoder provides feedback on the current position and velocity; In different motion states, by , Set to zero or assign a specific value, and combine with feedforward terms. It enables seamless switching between pure position servo, pure force control, and force-position hybrid control.

7. A waist-assist control system for an exoskeleton robot, characterized in that, The exoskeleton robot waist assist control method according to any one of claims 1-6, the exoskeleton robot waist assist control system includes: a sensing module, a processing and control module, a drive execution module, and a structural support and binding module.

Images