A knee exoskeleton and its dual-source phase prediction control method

By employing a dual-source phase prediction control method, the problems of unreasonable mass distribution and insufficient sensor robustness in the knee exoskeleton are solved, resulting in a more natural gait and safer assist control.

CN123103587APending Publication Date: 2026-11-17CHONGQING RES INST OF BEIJING UNIV OF TECH
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611234964.1
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-08-14
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

Existing knee exoskeletons suffer from problems such as unreasonable mass distribution, increased weight on the lower leg side leading to large knee joint swing inertia, unnatural gait, insufficient robustness of sensor signals, lagging feedforward assistance, and inadequate safety.

Method used

A dual-source phase predictive control method is adopted. By verifying the consistency of kinematic constraints between thigh IMU signals and motor encoder signals, and combining a lightweight continuous phase predictive model and contact reliability gating, continuous phase and feedforward torque are generated to achieve system mass convergence to thigh, sensor anomaly compensation and safety enhancement.

Benefits of technology

It reduces lower leg mass, improves gait naturalness and dynamic response speed, reduces assist lag and safety risks, and enhances control robustness and safety.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123103587A_ABST
    Figure CN123103587A_ABST
Patent Text Reader

Abstract

This invention provides a knee exoskeleton and its dual-source phase prediction control method. The entire device integrates the drive unit, control unit, and IMU on the thigh to reduce the swing inertia of the lower leg. Within a single control cycle, the control unit synchronously acquires signals from the thigh IMU and the motor encoder, calculates the dual-source consistency metric, and maps it to phase reliability. Contact reliability is then inferred based on joint motion characteristics. A lightweight continuous phase prediction model is employed, outputting the current phase and phase advance rate to calculate the predicted phase. The predicted phase is used to retrieve the standard assist torque curve. Gating, limiting, and graded safety processing of the feedforward expected torque are performed by combining the two types of reliability. This invention's knee exoskeleton and its dual-source phase prediction control method solves the problems of large mass distribution on the lower leg side, insufficient robustness of gait phase estimation, lag in feedforward assist, and insufficient safety under abnormal signals in existing technologies.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of wearable robots and exoskeleton robots, and in particular to a knee exoskeleton and its dual-source phase prediction control method. Background Technology

[0002] As a core piece of equipment for lower limb movement assistance, the structural layout of knee exoskeletons directly determines wearing comfort and system energy efficiency. Existing knee exoskeletons generally suffer from unreasonable mass distribution, with most designs placing large components such as drive motors, reducers, control boxes, and even energy storage batteries on the lower leg support side. During walking, the lower leg swings back and forth around the knee joint with significant amplitude. The added mass on the lower leg side significantly increases the rotational inertia of the knee joint, forcing the wearer to expend extra muscle force to counteract the inertial load, disrupting gait naturalness, and increasing the output load of the drive unit, exacerbating system energy consumption and shortening battery life. Other devices distribute sensors, controllers, and drive modules in multiple locations on the thigh and lower leg, resulting in messy and redundant internal wiring, making them susceptible to wear and loosening due to limb bending and pulling. This also increases the difficulty of assembly and maintenance, and compromises long-term reliability.

[0003] At the gait perception and phase estimation level, existing control schemes generally rely on a single sensor source to achieve gait phase recognition, making robustness difficult to guarantee. Common schemes only collect motion data through a knee joint encoder or a single IMU and independently calculate gait phase, lacking a multi-source signal cross-verification mechanism. In actual wearable scenarios, IMUs are susceptible to angular shifts due to temperature drift, accumulated errors, and mechanical vibration interference; encoders are prone to signal distortion due to mechanical hysteresis and instantaneous impacts. When a single-source signal is abnormal, the controller cannot autonomously identify the data reliability and is prone to outputting incorrect phase results. At the same time, most schemes use a discretized gait phase division method, only outputting discrete states such as the support phase and swing phase, without continuous phase output capability. Phase jumps are prone to occur during phase switching, directly causing sudden changes in assist torque. Existing multi-source schemes also typically do not quantify the kinematic constraint differences between different sensor sources into continuous reliability that can directly participate in prediction time window adjustment and torque gating.

[0004] In terms of power assist control and safety protection, existing technologies also have significant shortcomings. Current feedforward power assist control generally generates the target torque based on real-time phase lookup tables, without considering the end-to-end delays in sensor sampling, processing, motor response, and mechanical transmission. This results in the output assist lagging behind the actual movement trend of the human body, making the wearer prone to a feeling of disconnection in the force exerted. Regarding safety protection mechanisms, most devices only set fixed torque hard limits, lacking hierarchical and graded anomaly handling logic. When the ground contact state is ambiguous, the phase estimation reliability is low, or the sensor signal is abnormal, the system either continues to output incorrect assist torque or directly cuts off the power output. It lacks dynamic torque gating driven by both phase reliability and contact reliability, as well as a safety mechanism that switches step by step from derating assist, zero torque following to motor de-enabling. Summary of the Invention

[0005] The purpose of this invention is to provide a knee exoskeleton and its dual-source phase prediction control method, which solves the problems of large mass distribution on the lower leg side, insufficient robustness of gait phase estimation, lag in feedforward assist, and insufficient safety under abnormal signals in the prior art.

[0006] To achieve the above objectives, the present invention provides a knee exoskeleton and its dual-source phase prediction control method, comprising the following steps: S1. Synchronously acquire thigh IMU signals and motor encoder signals in a single control cycle; S2. Based on a unified time reference, the two sensor signals are time-stamp aligned, and filtering, outlier removal, coordinate system transformation, zero bias correction, and normalization preprocessing are performed. S3. Extract dual-source kinematic features from the preprocessed two-channel sensor signals, calculate the dual-source consistency metric based on the limb motion trend constraint provided by the thigh IMU signal and the joint relative motion constraint provided by the motor encoder signal, and map the dual-source consistency metric to phase confidence. S4. Based on the extreme value of the knee joint angle, the zero point of the knee joint angular velocity and the Euler angle trajectory of the thigh, perform multi-condition joint judgment within the preset event time window, and infer gait events and ground contact state through a hysteresis state machine with different entry and exit thresholds to solve the contact confidence. S5. Input the preprocessed thigh IMU signal, motor encoder signal, dual-source consistency metric, and historical phase sliding window into the lightweight continuous phase prediction model based on the historical sliding window, output the current phase and phase advance rate, and adjust the prediction time window according to the phase reliability; calculate the predicted phase based on the current phase, phase advance rate, and prediction time window. S6. Based on the predicted phase, retrieve the preset standard assist torque curve to generate the feedforward expected torque; S7. Using phase reliability and contact reliability as gating factors, the feedforward desired torque is weighted and released, and after amplitude limiting, rate of change limiting and state hysteresis processing, the target control torque is obtained. S8. Convert the target control torque into motor commands and send them to the drive unit to perform knee joint assistance.

[0007] Preferably, the dual-source consistency metric in S3 It is obtained by weighted summation of six characteristic differences: phase difference, angle change direction difference, angular velocity sign difference, angular velocity trend difference, phase propulsion rate difference, and gait event time difference. The calculation formula is as follows: ; in, The difference between various kinematic characteristics, Let be the weight coefficient, and satisfy... ; Phase reliability According to the dual-source consistency metric Calculated using a piecewise function, the phase confidence is used for prediction time window adjustment in S5 and feedforward torque gating in S7, expressed as follows: ; in, This is the normal threshold. This is the abnormal threshold, and .

[0008] Preferably, in S4, the ground contact probability is indirectly derived only by relying on the extreme value of the knee joint angle, the zero point of the knee joint angular velocity, and the Euler angle trajectory of the thigh. No need to configure plantar pressure sensors; contact reliability The contact confidence level is calculated by weighted fusion of the inferred ground contact probability value and the prior contact confidence value, with the contact confidence level serving as the feedforward torque gating factor in S7. The calculation formula is: ; in, Inferred from the heel strike and toe-off event models. To access the prior values ​​of credibility, This is a correction factor.

[0009] Preferably, the lightweight continuous phase prediction model in S5 is a lightweight neural network based on a historical sliding window. The model input includes preprocessed thigh IMU signal, motor encoder signal, dual-source consistency metric and historical phase sliding window sequence. The model output is the current phase and phase advance rate within the normalized interval. The model adopts integer quantization deployment, and the single inference time is less than the remaining calculation time of a single control cycle. Predicted phase Based on the current phase Phase propulsion rate and prediction time window Calculations show that when the phase confidence level decreases, the prediction time window is shortened, and the predicted phase is... The expression is: .

[0010] Preferably, the output torque after the feedforward gating release in S7 It can be represented as: ; in, To ensure credibility, For phase reliability, Forward release coefficient, The feedforward expected torque is generated based on the predicted phase; contact confidence and phase confidence characterize the reliability of the ground contact state and the reliability of the dual-source phase estimation, respectively, and together determine the release ratio of the feedforward expected torque; After performing amplitude limiting processing, the target control torque is output to the motor control loop. satisfy: ; in, This is the amplitude limiting function; ; in, and These are the preset maximum output torque and the preset minimum output torque, respectively; the target control torque is also subject to a rate-of-change limit so that the change in target control torque between adjacent control cycles does not exceed a preset change threshold.

[0011] A knee exoskeleton includes a thigh support, a lower leg support, a control unit, a drive unit, a knee joint connection structure, and an inertial measurement unit. The thigh support and the calf support are rotatably connected by a knee joint connection structure; the drive unit, control unit, and inertial measurement unit are all mounted on the thigh support. The thigh support is assembled from a thigh piece and a thigh cover; the thigh piece is integrally formed and has an IMU placement part and a motor placement position; the inertial measurement unit is fixedly installed in the IMU placement part; the drive motor of the drive unit is assembled in the motor placement position; the thigh piece has an IMU wire track along its own length; the wiring harness between the inertial measurement unit and the control unit is arranged in the IMU wire track. The control unit is integrated inside the control box assembly, which consists of the control box body, the control box cover, and the control box partition. The control box assembly is provided with an inner rail connected by a slide rail. An outer rail is formed on the outside of the thigh cover and connected to the outer rail. The inner rail and the outer rail are slidably engaged and assembled. The drive unit includes a drive motor, a lower leg fixation component, and a motor output connection structure; the output end of the drive motor is connected to the motor output connection structure; the motor output connection structure is connected to the knee joint connection structure; the lower leg fixation component is rigidly fixed to the lower leg support component; the drive unit has a built-in motor encoder; the motor encoder and the inertial measurement unit are electrically connected to the control unit respectively.

[0012] Preferably, the upper end of the thigh support is integrally extended with a shoulder strap connection portion; the shoulder strap connection portion is externally connected to a matching shoulder binding assembly; both the thigh support and the calf support are equipped with a knob binding structure on their outer sides; the knob binding structure includes a binding strap body and a knob tightening mechanism; one end of the binding strap body is fixed to the support, and the other end is stored inside the knob tightening mechanism.

[0013] Preferably, the IMU placement portion is located at the end of the thigh piece away from the knee joint connection structure; the IMU wire track is a closed slotted structure integrally formed by the thigh piece.

[0014] Preferably, the drive unit is integrally arranged on the side of the thigh support member; the power output end of the drive unit extends to the knee joint connection structure.

[0015] Therefore, the present invention employs the above-mentioned knee exoskeleton and its dual-source phase prediction control method, and the technical effects are as follows: 1. The structure layout of concentrating the drive unit, control unit and inertial measurement unit on one side of the thigh support is adopted, so that the main mass of the system is concentrated in the thigh area and the mass distribution on the lower leg side is significantly reduced. This solves the problems of excessive weight-adding components on the lower leg side of existing exoskeletons, excessive knee joint swing inertia, resulting in unnatural wearing gait, high drive load and slow dynamic response.

[0016] 2. A dual-source kinematic constraint consistency verification mechanism is adopted using thigh IMU signals and motor encoder signals. The dual-source consistency metric is calculated by weighting multi-dimensional features and the metric is mapped to continuous phase reliability. The phase reliability is further used for prediction time window adjustment and feedforward torque gating, thereby establishing a direct correspondence between the degree of sensing anomaly and the control output strength, reducing phase jumps and false assists caused by single sensor drift, hysteresis or impact interference.

[0017] 3. A lightweight continuous phase prediction model based on a historical sliding window is adopted to output the current phase and phase advance rate, and the prediction time window is adaptively adjusted according to the phase reliability. Based on the predicted phase, the standard assist torque curve is retrieved in advance to generate the feedforward expected torque, thereby compensating for the end-to-end delay of sensor sampling, processing, motor response and mechanical transmission, and reducing the lag of assist output.

[0018] 4. The feedforward torque is released by dual-factor gating of contact reliability and phase reliability, combined with amplitude limiting, rate of change limiting and graded safety handling, so that the assist torque under abnormal working conditions gradually transitions from normal assist to derated assist, zero torque following or motor disconnection, reducing the safety risks caused by erroneous impact torque and sudden power failure. Attached Figure Description

[0019] Figure 1 This is a schematic diagram of the overall frontal structure of a knee exoskeleton according to the present invention; Figure 2 This is a schematic diagram of the overall reverse view structure of a knee joint exoskeleton according to the present invention; Figure 3 This is a schematic diagram of the overall side view structure of a knee exoskeleton according to the present invention; Figure 4 This is a schematic diagram of the thigh support structure of a knee joint exoskeleton according to the present invention; Figure 5 This is a schematic diagram of the structure of a knee joint exoskeleton control device box according to the present invention; Figure 6 This is a schematic diagram of a knee joint drive structure of a knee joint exoskeleton according to the present invention; Figure 7 This is a schematic diagram of a dual-source phase prediction control process for a knee exoskeleton according to the present invention.

[0020] Figure Labels 1. Thigh support; 2. Lower leg support; 3. Control unit; 4. Drive unit; 5. Knee joint connection structure; 6. Inertial measurement unit; 7. Shoulder strap connection; 8. Thigh cover; 9. IMU placement section; 10. Thigh component; 11. IMU cable track; 12. Motor placement position; 13. Slide rail connecting outer rail; 14. Slide rail connecting inner rail; 15. Control box body; 16. Control box top cover; 17. Control box partition; 18. Drive motor; 19. Lower leg fixing component; 20. Motor output connection structure. Detailed Implementation

[0021] The technical solution of the present invention will be further described below with reference to the accompanying drawings and embodiments.

[0022] Unless otherwise defined, the technical or scientific terms used in this invention shall have the ordinary meaning as understood by one of ordinary skill in the art to which this invention pertains.

[0023] Example 1 like Figures 1-3 As shown, this invention provides a knee exoskeleton, including a thigh support 1, a lower leg support 2, a knee joint connecting structure 5, a drive unit 4, a control unit 3, and an inertial measurement unit 6. The thigh support 1 is fitted and fixed to the outer side of the wearer's thigh, and the lower leg support 2 is fitted and fixed to the outer side of the wearer's lower leg. The knee joint connecting structure 5 is assembled at the connection point between the thigh support 1 and the lower leg support 2, thereby enabling the lower leg support 2 to perform knee flexion and extension movements relative to the thigh support 1. Both the thigh support 1 and the lower leg support 2 adopt an arc-shaped support design adapted to the physiological curvature of the human leg, and both are equipped with sponge pads and flexible linings on their inner sides facing the body, effectively improving the fit and comfort during long-term wear.

[0024] This invention centrally arranges the drive unit 4, control unit 3, and inertial measurement unit 6 on one side of the thigh support 1, and arranges them sequentially along the length of the thigh support 1. During normal walking, the lower leg swings with a large stroke and high amplitude; the weight distributed on the lower leg side directly increases the knee joint's swing inertia. By removing the power and sensing electronic control components from the lower leg support 2 and minimizing the overall weight distribution on the lower leg side, the swing inertia of the knee joint's reciprocating motion is significantly reduced. This reduces the output load pressure of the drive unit 4, improving the overall dynamic response speed of the device, and allows the wearer to walk more smoothly, optimizing the naturalness and coordination of the walking gait.

[0025] like Figure 4 As shown, the thigh support 1 is assembled from the thigh part 10 and the thigh cover 8, which are fastened together by bolts to form a closed mounting cavity for disassembly and assembly of components. An IMU placement part 9 is integrally formed on the thigh part 10 for placing and securing the inertial measurement unit 6. To reduce the impact of the vibration of the drive unit 4 on the detection accuracy of the inertial measurement unit 6, the IMU placement part 9 is preferably located at the far end of the thigh support 1 away from the knee joint connection structure 5. Simultaneously, the thigh part 10 has an IMU wire track 11 along its length, which serves as a wiring channel for laying the connecting wires between the inertial measurement unit 6 and the control unit 3. The wiring channel prevents exposed wires from being dragged and repeatedly bent, causing damage and improving the reliability of the entire electrical system, while also keeping the internal wiring neat and simple. A motor placement position 12 is also provided at the lower end of the thigh body for mounting the drive motor 18.

[0026] like Figure 5As shown, the control unit 3 is integrated and packaged inside the control box assembly, which includes the control box body 15, the control box cover 16, and the control box partition 17. The internal cavity of the control box body 15 can accommodate components such as a battery, a low-power edge microcontroller, a motor drive circuit board, a wireless communication module, and a power management module. The control box body 15 and the control box cover 16 are detachably assembled using screw locking, facilitating future maintenance and replacement of internal electrical components. The control unit 3 achieves electrical signal communication with the inertial measurement unit 6 and the encoder built into the drive unit 4, respectively.

[0027] The control box assembly and the thigh support 1 are connected via a quick-release sliding rail connector structure. This structure includes an inner sliding rail 14 on the control box assembly and an outer sliding rail 13 formed on the thigh cover. The inner and outer sliding rails 14 and 13 slide and engage with each other, enabling quick assembly and disassembly of the control unit 3 and fine-tuning of its position. This structure improves the convenience of equipment maintenance and allows for flexible adjustment of the control unit 3's mounting position based on different wearers' leg circumferences and overall weight distribution requirements, thus optimizing the exoskeleton's balance.

[0028] The control box assembly is adjustable along the thigh support 1, so that the control unit 3 can be placed in a stable area close to the thigh under different wearer conditions; the inertial measurement unit 6 is set away from the knee joint drive end, and with the closed wire track, it can reduce the impact of drive vibration, wire harness swing and connection loosening on dual-source synchronous acquisition and consistency calculation, so that the mechanical low inertia layout and dual-source phase predictive control form a synergistic relationship.

[0029] The control unit 3 has built-in preset control logic, which is executed in the following order: "dual-source motion constraint consistency quantization - phase confidence calculation - predictive phase compensation - dual confidence gating assist". The specific process is as follows: Within a single control operation cycle, the thigh posture IMU data fed back by the inertial measurement unit 6 and the relative rotation data of the knee joint fed back by the motor encoder are collected synchronously. The thigh IMU attitude signal and joint relative motion signal are filtered and preprocessed to extract the lower limb kinematic feature parameters corresponding to the two data sources. Based on the constraints of the overall limb motion trend reflected by the IMU signal and the constraints of the relative joint motion reflected by the encoder signal, a dual-source consistency metric is calculated and mapped to phase reliability. Based on the predicted phase, the expected feedforward torque of the knee joint is retrieved from the preset standard assist torque curve; By combining contact reliability and phase reliability, the feedforward expected torque is gated, weighted, amplitude limited, and rate of change limited to obtain the target output torque, which is then converted into a motor drive command and sent to drive unit 4 for execution.

[0030] like Figure 6 As shown, the main body of the drive unit 4 is located on the side of the thigh support 1, with its power output end extending to the knee joint connection structure 5, thereby driving the lower leg support 2 to complete rotational flexion and extension movements relative to the thigh support 1. The drive unit 4 mainly includes a drive motor 18, a lower leg fixation component 19, and a motor output connection structure 20. The drive motor 18 is a direct-drive low-inertia motor with high torque density characteristics; the motor output connection structure 20 adopts a gear transmission mechanism containing a gear set and an output disc, which transmits the output torque of the drive motor 18 to the knee joint connection structure 5 on the one hand, and amplifies the power through gear transmission on the other hand. The lower leg fixation component 19 is rigidly connected to the lower leg support 2, which can reliably apply the output torque of the drive unit 4 to the lower leg support 2, thereby providing assistance for the wearer's knee flexion and extension movements.

[0031] The drive unit 4 also includes a motor encoder, which can collect motion parameters such as the rotation angle of the drive motor 18 shaft, the relative rotation angle of the knee joint, and the rotational angular velocity of the knee joint in real time. The control unit 3 calculates the relative motion constraints of the joint based on the measured signal fed back by the motor encoder, providing data basis for lower limb phase judgment and torque output control.

[0032] Both the thigh support 1 and the calf support 2 of this invention are equipped with an adjustable knob binding structure, which mainly consists of a strap body and a knob tightening mechanism. One end of the strap body is fixed to the corresponding support, while the other end is wound and stored around the outer side of the winding wheel of the knob tightening mechanism. The knob tightening mechanism is installed entirely on the outside of the support and mainly includes a knob body, a drive shaft, a winding wheel, a mounting housing, and a one-way locking component. It has dual functions of tightening and quick release. When the knob body is rotated clockwise, the winding wheel rotates to wind up the strap, achieving precise adjustment of the strap tension. When the release button in the center of the knob is pressed, the one-way locking structure is released, and the strap automatically loosens and releases, making assembly and disassembly convenient.

[0033] The upper end of the thigh support component 1 extends to form a shoulder strap connection portion 7. This structure connects to the wearer's shoulders via a matching shoulder binding structure, distributing part of the exoskeleton's weight to the torso and effectively reducing the load on the legs. The shoulder binding structure consists of shoulder straps and a corresponding knob tightening mechanism. The shoulder straps are 5 cm wide and feature an internal memory foam cushioning layer, increasing the contact area, cushioning wearing pressure, and improving long-term comfort and fit stability.

[0034] When wearing it, first use the knob binding structure on the thigh and calf to securely fasten the thigh support 1 and calf support 2 to the corresponding limbs of the wearer. Then adjust the shoulder knob binding structure to a comfortable tightness.

[0035] Compared to traditional punched straps, this solution uses a knob-type binding structure that allows for stepless, continuous tension adjustment, completely avoiding the drawbacks of fixed hole positions and limited adjustment levels in traditional straps, thus adapting to the needs of wearers of different body types. At the same time, each knob tightening mechanism has a built-in overload protection mechanism, effectively preventing the straps from becoming too tight and compressing the limbs, preventing limb injuries, and ensuring the safety and reliability of the equipment.

[0036] like Figure 7 As shown, this method is a real-time compliant assist control approach adapted to low-inertia mechatronic knee exoskeleton. Relying on a low-power edge microcontroller, it completes dual-source signal acquisition, preprocessing, motion constraint consistency quantization, phase reliability calculation, lightweight continuous phase prediction, contact reliability calculation, dual-reliability gated torque output, and graded safety processing within a single control cycle of 1ms-20ms. This method can calculate the ground contact state without a plantar pressure sensor and smoothly reduce assist or switch to a safety mode when the signal is abnormal.

[0037] After the control cycle is triggered, the control unit 3 is electrically connected to the inertial measurement unit 6, the motor encoder, and the motor drive circuit, respectively. At the beginning of each control cycle, the control unit 3 synchronously acquires the thigh IMU signal and the motor encoder signal. The thigh IMU signal includes the thigh Euler angle, thigh angular velocity, and thigh acceleration; the motor encoder signal includes the knee joint angle and knee joint angular velocity.

[0038] Control unit 3 first performs signal preprocessing on the acquired dual-source signals. This preprocessing includes low-pass filtering, outlier removal, time synchronization, coordinate system transformation, zero-bias correction, and normalization. For angular signals such as thigh Euler angles and knee joint angles, low-pass filtering or moving average filtering is used to suppress high-frequency disturbances introduced by mechanical vibration. For velocity signals such as thigh angular velocity, thigh acceleration, and knee joint angular velocity, threshold constraints and spike removal methods are used to suppress instantaneous impact noise. The preprocessed signals are stored in a historical sliding window, providing a data foundation for subsequent feature extraction and phase prediction.

[0039] After signal preprocessing, control unit 3 performs dual-source kinematic feature extraction. For the thigh IMU signal, the extracted features include the thigh Euler angle trajectory, thigh pitch angle change rate, thigh angular velocity sign, thigh angular velocity peak value, and thigh acceleration impact characteristics. For the motor encoder signal, the extracted features include knee joint angle, knee joint angular velocity, knee joint angle extreme value, knee joint angular velocity zero crossing point, and knee joint flexion and extension direction changes. The dual-source combined features include the consistency between the thigh motion trend and the knee joint flexion and extension trend, the phase propulsion rate difference, the gait event time difference, and the amplitude relationship between the two source signals.

[0040] Subsequently, control unit 3 performs a dual-source consistency check based on the aforementioned dual-source kinematic features. The thigh IMU signal provides constraints on the overall limb movement trend, while the motor encoder signal provides constraints on the relative joint movement. Control unit 3 compares the difference between the gait phase inferred from the thigh IMU signal and the gait phase inferred from the encoder signal, compares the matching relationship between the thigh movement direction and the knee flexion / extension direction, and compares the consistency of the heel strike and toe lift events inferred from the dual-source signals. These differences are uniformly quantified into a dual-source consistency measure.

[0041] Dual-source consistency metric The difference is calculated by weighting six characteristic differences: phase difference, angle change direction difference, angular velocity sign difference, angular velocity trend difference, phase propulsion rate difference, and gait event time difference. The calculation formula is as follows: ; in, The difference between various kinematic characteristics, Let be the weight coefficient, and satisfy... ; Based on the dual-source consistency verification results, control unit 3 calculates the phase reliability, which is expressed as a percentage. The phase reliability increases when the dual-source signals exhibit consistent trends, have a small phase difference, and provide consistent gait event inferences. Conversely, the phase reliability decreases when the dual-source signals conflict in trends, the phase difference exceeds a set threshold, or an anomaly occurs in the single-source signal. The phase reliability is passed as an independent control variable to the prediction time window adjustment and feedforward torque gating stages, and is not repeatedly output by the continuous phase prediction model.

[0042] Phase reliability According to the dual-source consistency metric Calculated using a piecewise function, the expression is: ; in, This is the normal threshold. This is the abnormal threshold, and .

[0043] By calculating the phase reliability, the kinematic difference between the two sources can be converted into a continuously changing reliability metric, which establishes a direct correspondence between the degree of sensor anomaly and the prediction lead and the feedforward torque release ratio. This avoids abrupt changes in the feedforward torque due to a single sensor anomaly and allows the control unit 3 to gradually reduce the assist output according to the degree of signal anomaly, ensuring the smoothness and safety of the control process.

[0044] After completing the dual-source consistency verification, control unit 3 indirectly infers gait events and contact states based on features such as knee joint angle extremes, knee joint angular velocity zero-crossing points, and thigh Euler angle trajectories. Gait events include heel strike events and toe-off events, and contact states include the support phase and the swing phase. The control unit calculates the contact confidence level based on the gait events and contact states. The contact confidence level is used to limit the release ratio of the feedforward expected torque when the contact state is unclear.

[0045] Specifically, within an event time window of 3 to 10 control cycles, the local extreme values ​​of the knee joint angle, the zero crossing of the knee joint angular velocity, and the change in the thigh pitch angle trajectory are checked to see if they meet the preset combination conditions corresponding to the target event. When at least two of the three criteria are met and the remaining criteria do not conflict, a candidate event of heel strike or toe lift is formed; the contact state is only allowed to change after the candidate event is held continuously for more than two control cycles. Different thresholds are used for entering and exiting the support phase, and a minimum state holding time is set to suppress repeated jumps in the critical position.

[0046] Contact credibility The value obtained by fusing the inferred ground contact probability with the prior contact confidence value can be expressed as: ; in, Inferred from the heel strike and toe-off event models. To access the prior values ​​of credibility, This is a correction factor.

[0047] The above contact reliability calculation does not require the configuration of plantar pressure sensors. It indirectly determines the contact state only through the extreme value of the knee joint angle, the zero point of the knee joint angular velocity, and the Euler angle trajectory of the thigh. It is used as an independent gating factor in the calculation of the target control torque, which can reduce false assistance when the contact state is unclear.

[0048] The lightweight continuous phase prediction model based on historical sliding windows takes preprocessed thigh IMU signals, motor encoder signals, dual-source consistency metrics, and historical phase sliding window sequences as inputs, and outputs the current phase and phase advance rate to calculate the predicted phase. Both the current phase and the predicted phase are expressed as a percentage of normalized gait period, with values ​​ranging from [value range missing]. The predicted phase has a preset lead time relative to the current phase, which is determined based on the phase advance rate and the prediction time window. When the phase confidence decreases, the control unit 3 shortens the prediction time window to reduce the risk of over-prediction in the low confidence state.

[0049] In this embodiment, the continuous phase prediction model employs a lightweight neural network. To adapt to the computing power and storage conditions of low-power edge microcontrollers, the model input window is the most recent N control cycles, the number of model parameters is less than the available storage space of the edge controller, and the model's single inference time is less than the remaining computation time of a single control cycle; the model can use fixed-point or quantization methods to reduce storage usage and computation latency.

[0050] Predicted phase Based on the current phase Phase propulsion rate and prediction time window The calculation is expressed as follows: ; Control unit 3 reads and calculates the expected feedforward torque from a preset standard assist torque curve based on the predicted phase. The standard assist torque curve represents the correspondence between gait phase and knee joint assist torque, and is pre-calibrated by averaging multiple sets of sample data collected in offline gait experiments. Compared with the method of generating torque based on the current phase, reading the feedforward torque based on the predicted phase can compensate for system delays caused by sensor sampling, processing, signal communication, motor response, and mechanical transmission. When the phase reliability decreases, the prediction time window is shortened synchronously to avoid amplifying control errors due to low-reliability predictions.

[0051] Control unit 3 uses contact reliability and phase reliability as independent gating factors to perform weighted release on the expected feedforward torque, enabling it to actively reduce torque output when dual-source phase estimation is unreliable or ground contact status is unclear. The target output torque is jointly determined by the expected feedforward torque, contact reliability, phase reliability, and feedforward release coefficient; the feedforward release coefficient is dynamically adjusted based on sensor status, gait status, motor status, and safety status. To avoid sudden changes in output torque, control unit 3 also performs amplitude limiting, rate of change limiting, directional consistency verification, and low-pass filtering on the target output torque.

[0052] Output torque after feedforward gating release It can be represented as: ; in, To ensure credibility, For phase reliability, Forward release coefficient, This is the expected feedforward torque generated based on the predicted phase.

[0053] After performing amplitude limiting processing, the target control torque is output to the motor control loop. satisfy: ; in, This is the amplitude limiting function; ; in, and These are the preset maximum output torque and the preset minimum output torque, respectively. By predicting the phase and reading the standard assist torque curve, and combining dual-confidence gating, amplitude limiting, and rate of change limiting, the control unit 3 can compensate for sampling, calculation, and drive response delays while preventing the output torque from exceeding the preset safety boundary or from abruptly changing in adjacent control cycles.

[0054] The target output torque, after gating and weighting, is converted into a motor control command and output to the drive unit 4. The drive unit 4 outputs the corresponding torque according to the motor control command, driving the lower leg support 2 to complete the assisted flexion and extension movement relative to the thigh support 1, thereby providing gentle assistance to the wearer's knee joint.

[0055] This invention does not employ a strategy of completely shutting down control upon single-source anomaly. Instead, it performs graded safety processing based on phase reliability, contact reliability, and the duration of dual-source anomalies: when any reliability is below the corresponding derating threshold but the system can still maintain basic control, the control unit 3 reduces the feedforward release coefficient and limits the maximum output torque and maximum output speed; when the anomaly duration exceeds the first preset time threshold, it switches to zero torque following mode; when the degree of dual-source signal anomaly exceeds the safety threshold or the anomaly duration exceeds the second preset time threshold, it switches to motor de-enabling mode, wherein the second preset time threshold is not less than the first preset time threshold.

[0056] Therefore, this invention centrally arranges the drive unit, control unit, and inertial measurement unit in the thigh support to reduce the swing inertia of the lower leg. The control unit completes the acquisition, preprocessing, and feature extraction of signals from the thigh IMU and motor encoder within a single control cycle. It obtains the phase reliability through dual-source motion constraint consistency measurement and adjusts the prediction time window of the lightweight continuous phase prediction model based on the phase reliability. The control unit then calculates the contact reliability by combining the joint motion characteristics to predict the phase and generate the feedforward expected torque. After dual reliability gating, amplitude limiting, rate of change limiting, and graded safety processing, the assist torque is output, thus forming a complete closed-loop control chain of "dual-source consistency quantization - phase reliability - prediction compensation - dual reliability gating".

[0057] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A knee exoskeleton and its dual-source phase prediction control method, characterized in that, Includes the following steps: S1. Synchronously acquire thigh IMU signals and motor encoder signals in a single control cycle; S2. Based on a unified time reference, the two sensor signals are time-stamp aligned, and filtering, outlier removal, coordinate system transformation, zero bias correction, and normalization preprocessing are performed. S3. Extract dual-source kinematic features from the preprocessed two-channel sensor signals, calculate the dual-source consistency metric based on the limb motion trend constraint provided by the thigh IMU signal and the joint relative motion constraint provided by the motor encoder signal, and map the dual-source consistency metric to phase confidence. S4. Based on the extreme value of the knee joint angle, the zero point of the knee joint angular velocity and the Euler angle trajectory of the thigh, perform multi-condition joint judgment within the preset event time window, and infer gait events and ground contact state through a hysteresis state machine with different entry and exit thresholds to solve the contact confidence. S5. Input the preprocessed thigh IMU signal, motor encoder signal, dual-source consistency metric, and historical phase sliding window into the lightweight continuous phase prediction model based on the historical sliding window, output the current phase and phase advance rate, and adjust the prediction time window according to the phase reliability; calculate the predicted phase based on the current phase, phase advance rate, and prediction time window. S6. Based on the predicted phase, retrieve the preset standard assist torque curve to generate the feedforward expected torque; S7. Using phase reliability and contact reliability as gating factors, the feedforward desired torque is weighted and released, and after amplitude limiting, rate of change limiting and state hysteresis processing, the target control torque is obtained. S8. Convert the target control torque into motor commands and send them to the drive unit to perform knee joint assistance.

2. The knee exoskeleton and its dual-source phase prediction control method according to claim 1, characterized in that, Dual-source consistency measure in S3 It is obtained by weighted summation of six characteristic differences: phase difference, angle change direction difference, angular velocity sign difference, angular velocity trend difference, phase propulsion rate difference, and gait event time difference. The calculation formula is as follows: ; in, The difference between various kinematic characteristics, Let be the weight coefficient, and satisfy... ; Phase reliability According to the dual-source consistency metric Calculated using a piecewise function, the phase confidence is used for prediction time window adjustment in S5 and feedforward torque gating in S7, expressed as follows: ; in, This is the normal threshold. This is the abnormal threshold, and .

3. The knee exoskeleton and its dual-source phase prediction control method according to claim 1, characterized in that, In S4, the probability of ground contact is indirectly derived only by relying on the extreme value of the knee joint angle, the zero point of the knee joint angular velocity, and the Euler angle trajectory of the thigh. No need to configure plantar pressure sensors; contact reliability The contact confidence level is calculated by weighted fusion of the inferred ground contact probability value and the prior contact confidence value, with the contact confidence level serving as the feedforward torque gating factor in S7. The calculation formula is: ; in, Inferred from the heel strike and toe-off event models. To access the prior values ​​of credibility, This is a correction factor.

4. The knee exoskeleton and its dual-source phase prediction control method according to claim 1, characterized in that, The lightweight continuous phase prediction model in S5 is a lightweight neural network based on a historical sliding window. The model input includes preprocessed thigh IMU signals, motor encoder signals, dual-source consistency metrics, and historical phase sliding window sequences. The model output is the current phase and phase advance rate within the normalized interval. The model uses integer quantization deployment, and the single inference time is less than the remaining computation time of a single control cycle. Predicted phase Based on the current phase Phase propulsion rate and prediction time window Calculations show that when the phase confidence level decreases, the prediction time window is shortened, and the predicted phase is... The expression is: 。 5. The knee exoskeleton and its dual-source phase prediction control method according to claim 1, characterized in that, The output torque after feedforward gating release in S7 It can be represented as: ; in, To ensure credibility, For phase reliability, Forward release coefficient, The feedforward expected torque is generated based on the predicted phase; contact confidence and phase confidence characterize the reliability of the ground contact state and the reliability of the dual-source phase estimation, respectively, and together determine the release ratio of the feedforward expected torque; After performing amplitude limiting processing, the target control torque is output to the motor control loop. satisfy: ; in, This is the amplitude limiting function; ; in, and These are the preset maximum output torque and the preset minimum output torque, respectively; the target control torque is also subject to a rate-of-change limit so that the change in target control torque between adjacent control cycles does not exceed a preset change threshold.

6. A knee exoskeleton, based on the knee exoskeleton and its dual-source phase prediction control method according to any one of claims 1-5, characterized in that, It includes a thigh support component, a calf support component, a control unit, a drive unit, a knee joint connection structure, and an inertial measurement unit; The thigh support and the calf support are rotatably connected by a knee joint connection structure; the drive unit, control unit, and inertial measurement unit are all mounted on the thigh support. The thigh support is assembled from a thigh piece and a thigh cover; the thigh piece is integrally formed and has an IMU placement part and a motor placement position; the inertial measurement unit is fixedly installed in the IMU placement part; the drive motor of the drive unit is assembled in the motor placement position; the thigh piece has an IMU wire track along its own length; the wiring harness between the inertial measurement unit and the control unit is arranged in the IMU wire track. The control unit is integrated inside the control box assembly, which consists of the control box body, the control box cover, and the control box partition. The control box assembly is provided with an inner rail connected by a slide rail. An outer rail is formed on the outside of the thigh cover and connected to the outer rail. The inner rail and the outer rail are slidably engaged and assembled. The drive unit includes a drive motor, a lower leg fixation component, and a motor output connection structure; the output end of the drive motor is connected to the motor output connection structure; the motor output connection structure is connected to the knee joint connection structure; the lower leg fixation component is rigidly fixed to the lower leg support component; the drive unit has a built-in motor encoder; the motor encoder and the inertial measurement unit are electrically connected to the control unit respectively.

7. A knee joint exoskeleton according to claim 6, characterized in that, The upper end of the thigh support is integrally extended with a shoulder strap connection part; the shoulder strap connection part is externally connected to a matching shoulder binding component; both the thigh support and the calf support are equipped with a knob binding structure on their outer sides; the knob binding structure includes a binding strap body and a knob tightening mechanism; one end of the binding strap body is fixed to the support, and the other end is stored inside the knob tightening mechanism.

8. A knee joint exoskeleton according to claim 6, characterized in that, The IMU placement section is located at the end of the thigh piece away from the knee joint connection structure; the IMU wire track is a closed slotted structure integrally formed by the thigh piece.

9. A knee joint exoskeleton according to claim 6, characterized in that, The drive unit is integrally arranged on the side of the thigh support; the power output end of the drive unit extends to the knee joint connection structure.