Lower limb exoskeleton knee joint power-assisted control method and system

Abstract

The invention relates to the technical field of wearable exoskeleton assistance, in particular to a lower limb exoskeleton knee joint assistance control method and system. The lower limb exoskeleton comprises a pair of swing arms respectively connected with thighs and shanks and an actuator, and the method comprises the following steps: processing data of a motion parameter sensor, acquiring detection data input by the motion parameter sensor arranged on the lower limb exoskeleton, and processing the detection data to extract motion features; gait phase detection is conducted, the current gait phase is determined according to the motion characteristics, the gait phase comprises a supporting phase and a swinging phase, the supporting phase corresponds to the state that the lower limb touches the ground, and the swinging phase corresponds to the state that the lower limb leaves the ground; activity identification: motion features are extracted in the supporting phase to determine activity types, and the activity types comprise at least one of flat walking, upward walking, downward walking, running, squatting, backward walking and unknown; according to the control strategy, the corresponding control strategy is selected according to the gait phase and the activity type so as to control an actuator.

Description

Technical Field

[0001] The present invention relates to the field of wearable exoskeleton power-assisting technology, and in particular to a lower limb exoskeleton knee joint power-assisting control method and system. Background Art

[0002] In recent years, exoskeleton technology has made great progress in medical rehabilitation, industrial assistance and consumer fields, especially in knee joint assistance, showing significant application potential. These systems usually use motors, pneumatic or hydraulic actuators to provide external assistance to the wearer, and combine a variety of motion parameter sensors (such as accelerometers, force motion parameter sensors and angle encoders) to collect motion data in real time, and realize dynamic adjustment of assistance through precise control algorithms to adapt to different usage scenarios. With the expansion of the application scope of exoskeletons, their control methods in gait phase recognition and activity type detection have become a key direction for improving system performance. In the consumer field, especially in outdoor activities such as mountaineering, the demand for knee joint assistance exoskeletons is increasing. Mountaineering activities involve changing terrain, large slope differences and long-term physical exertion, which puts a particularly significant burden on the knee joint. Despite this, there is still room for improvement in the adaptability, control accuracy and responsiveness to dynamic changes of existing technologies in complex activity scenarios, which provides the necessity and innovation basis for the research of this patent "a method and system for controlling the use of exoskeletons based on gait phase and activity type detection".

[0003] In the prior art, several solutions have attempted to optimize and improve the performance of exoskeletons through motion state detection and control strategies. For example, US 20240077848A1 patent proposes an exoskeleton activity conversion control method and system. The system collects data such as knee joint angle, plantar pressure, and thigh and calf acceleration through motion parameter sensors, combines machine learning models to predict the wearer's activity type (such as walking, ascending, descending or running), and detects the transition between activities. Based on the predicted confidence score, the system adjusts the output force or torque of the actuator to improve adaptability and safety in multiple activity scenarios. Similarly, US 20220096249A1 patent describes a control architecture based on the comparison of predicted and actual motion parameter sensor data, which dynamically adjusts the power-assisting action (such as reducing the power-assisting or optimizing the joint range) by calculating uncertainty metrics to improve the efficiency of activity conversion. This method relies on accelerometers, force motion parameter sensors, and bending motion parameter sensors, emphasizing the importance of motion state detection in control. However, these two technologies mainly focus on the prediction of activity types, but pay insufficient attention to the detailed identification of gait phases and their deep coupling with power-assisting regulation.

[0004] In terms of the combination of structure and control, the CN111805511B patent discloses a lower limb exoskeleton system with actively adjustable leg rod length. The system dynamically adjusts the length of the thigh and calf rods through a linear displacement device to align the exoskeleton with the wearer's joints, and uses a cubic spline model to calculate the target length based on the joint angle data, while using an adaptive oscillator to identify the gait phase to optimize the power assistance. This technology demonstrates the application potential of gait phase detection in rehabilitation and walking assistance scenarios, but its control method still focuses on structural adaptation rather than a comprehensive analysis of activity types and phases. In addition, the US11931307B2 patent and related literature propose an exoskeleton control system for skiing scenarios, which collects data such as joint angles and pressure through motion parameter sensors and adjusts fluid pressure to provide power assistance. Its innovation lies in combining the user's surrounding environment data (such as other skier information) and GPS location to optimize the control strategy, but this method is highly dependent on specific scenarios and lacks a systematic solution for general gait phase and activity type detection.

[0005] Academic research has also provided inspiration for exoskeleton control methods. For example, "Assistive Control Design for Powered Lower Limb Orthosis" proposed a control architecture for powered lower limb orthosis, designed for the elderly or people with limited mobility. The system generates a reference pattern based on the gait of a healthy person, calculates the joint angles through linear or quadratic interpolation, and estimates the gait parameters using the cumulative moving average method. When the gait is abnormal, the reference pattern is adjusted online to correct the movement. This method performs well in rehabilitation training, but its control logic relies more on preset gait patterns, and its ability to detect and respond to changes in dynamic activity types in real time is limited. Summary of the invention

[0006] In order to overcome the deficiencies of the above-mentioned prior art, the present invention provides a lower limb exoskeleton knee joint power-assist control method and system, which improves the stability of the system in complex environments and abnormal conditions by dynamically identifying gait phases and activity types and performing adaptive control.

[0007] In order to achieve the above object, the present invention is implemented through the following technical solutions: A lower limb exoskeleton knee joint power-assistance control method, the lower limb exoskeleton comprising a pair of swing arms and an actuator respectively connected to the thigh and the calf, the actuator being used to apply torque to the pair of lower limb exoskeletons to assist the knee joint movement; The steps include: Motion parameter sensor data processing, acquiring detection data input by a motion parameter sensor disposed on the lower limb exoskeleton, and processing the detection data to extract motion features; Gait phase detection, determining the current gait phase according to the motion characteristics, wherein the gait phase includes a stance phase and a swing phase, wherein the stance phase corresponds to the state where the lower limbs touch the ground, and the swing phase corresponds to the state where the lower limbs leave the ground; Activity recognition: extracting motion features in the stance phase to determine the activity type, which includes at least one of: walking, ascending, descending, running, squatting, backwards, and unknown; Control strategy, select the corresponding control strategy to control the actuator according to the gait phase and activity type.

[0008] Preferably, in the method provided by the present application, the number of the motion parameter sensors is 2, which are a pair of IMUs respectively arranged on a pair of swing arms. The condition of the knee joint angle can be obtained by converting the motion data such as the tilt angle and angular velocity obtained by the pair of IMUs.

[0009] Preferably, in the method provided by the present application, the number of the motion parameter sensors is 2, which is a knee angle sensor and an IMU arranged on one of the swing arms. The motion state corresponding to the swing arm without the IMU can be obtained by converting the motion data obtained by the IMU and the knee angle change obtained by the knee angle sensor.

[0010] Furthermore, in the method provided by the present application, the control strategy includes: at least one of elastic control, damping control, and hybrid control; wherein: Elastic control is to control the torque of the actuator acting on the swing arm according to the knee joint angle to simulate the spring; Damping control is to control the torque of the actuator acting on the swing arm according to the angular velocity of the knee joint to simulate the damper; Hybrid control combines elasticity and damping control to suit the current activity type; The control strategy corresponding to the support phase includes at least one of elastic control, damping control, and mixed control; When executing the control strategy, the control parameters are adjusted in real time according to the current gait phase and / or activity type and / or active motion characteristics.

[0011] Furthermore, in the method provided by the present application, the control strategy includes: at least one of transparent control and resistance reduction control; wherein: The control strategies corresponding to the swing phase include transparent control and / or resistance reduction control.

[0012] Furthermore, in the method provided by the present application, the gait phase also includes an unknown phase, and the unknown phase corresponds to at least one of the following situations: the motion characteristics do not clearly meet the transition state during the support or swing phase, the algorithm cannot determine the current gait, and the algorithm is initialized; the control strategy of the unknown phase includes transparent control and / or resistance reduction control to ensure safety and comfort.

[0013] Furthermore, the method provided by the present application also includes adjusting the gait transition conditions according to the current activity type and / or motion characteristics, so as to ensure that the exoskeleton adapts to the changing scenarios.

[0014] Furthermore, in the method provided by the present application, the motion characteristics include knee joint angle, knee joint angular velocity, inclination angle of the calf and / or thigh, angular velocity of the calf and / or thigh, and acceleration value of the thigh and / or calf.

[0015] Furthermore, in the method provided by the present application, the process of activity identification includes: Threshold rule-based feature comparison, which uses predefined threshold rules to compare motion features to determine whether they meet the conditions for a specific activity; Extreme value detection within the time window: Use the time window to analyze the extreme values ​​of motion features to capture the dynamic changes of motion, and then filter them through time conditions to ensure the robustness of detection; Cumulative counting and state continuity check: accumulate counts of motion features that meet preset conditions, such as the number of times the calf angle peak exceeds a certain threshold, or the duration of the knee joint angular velocity remaining within a certain range. The state continuity check is used to avoid misjudgment; Dynamic update: In the support phase, the motion features are dynamically updated, and the activity types are distinguished through these dynamically updated motion features; conditional logic combination: multiple motion features are jointly analyzed using conditional logic combination. Therefore, in this application, the activity recognition process is a lightweight analysis method based on the trend of motion parameter sensor data. It does not rely on resource-intensive technologies (such as machine learning or deep learning), but is based on direct feature extraction and threshold logic judgment of real-time motion parameter sensor data. Through predefined rules and time window analysis, the computational complexity is reduced to ensure efficient operation on embedded devices. All feature calculations use low-overhead mathematical operations (such as addition, multiplication, and comparison), and are combined with cached data in a limited time window to avoid large-scale data storage and processing.

[0016] The present application also provides a lower limb exoskeleton knee joint power assist control system, comprising: The lower limb exoskeleton includes a pair of swing arms and actuators respectively connected to the thigh and the calf, and the actuators are used to apply torque to the pair of lower limb exoskeletons to assist the knee joint movement; The detection module is used to obtain detection data input by the motion parameter sensor arranged on the lower limb exoskeleton, A motion feature extraction module, used for processing the detection data to extract motion features; A gait phase judgment module is used to determine the current gait phase according to the motion characteristics, wherein the gait phase includes a support phase and a swing phase, wherein the support phase corresponds to the state where the lower limb touches the ground, and the swing phase corresponds to the state where the lower limb leaves the ground; An activity recognition module, used for extracting motion features in the support phase to determine the activity type, where the activity type includes at least one of walking, ascending, descending, running, squatting, backwards and unknown; The control strategy confirmation module is used to select the corresponding control strategy to control the actuator according to the gait phase and activity type.

[0017] It can be seen from the above technical solution that the present invention has the following beneficial effects: 1. Enhance the robustness and environmental adaptability of the system. By dynamically identifying gait phases and activity types and performing adaptive control, the stability of the system in complex environments and abnormal situations is improved, ensuring reliable assistance in dynamic scenarios such as mountain climbing and improving versatility.

[0018] 2. Simplify the configuration of motion parameter sensors and reduce system costs. The present invention only requires two motion parameter sensors, avoids the reliance on complex motion parameter sensors, simplifies the exoskeleton device, and reduces procurement and manufacturing costs.

[0019] 3. Reduce hardware computing requirements and improve popularity. The activity recognition process is lightweight and can run efficiently on a standard MCU without the need for high-performance hardware, significantly reducing hardware costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 A schematic diagram of an exoskeleton power assist device in one embodiment of the present application (using two IMUs); Figure 2 A schematic diagram of an exoskeleton power assist device in one embodiment of the present application (using an IMU and an encoder); Figure 3 Schematic diagram of the principle of corresponding hip projection difference in one embodiment of the present application.

[0021] In the figure: 1-exoskeleton; 11-swing arm; 12-actuator. DETAILED DESCRIPTION

[0022] Combination Figure 1 and Figure 2 As shown, a method for assisting the knee joint of a lower limb exoskeleton is provided, wherein the lower limb exoskeleton 1 comprises a pair of swing arms 11 and an actuator 12 respectively connected to the thigh and the calf, and the actuator 12 is used to apply torque to the pair of lower limb exoskeletons 1 to assist the knee joint movement; The steps include: Motion parameter sensor data processing, obtaining detection data input by a motion parameter sensor disposed on the lower limb exoskeleton 1, and processing the detection data to extract motion features; Gait phase detection, determining the current gait phase according to the motion characteristics. The gait phase includes the stance phase and the swing phase, where the stance phase corresponds to the state where the lower limbs touch the ground, and the swing phase corresponds to the state where the lower limbs leave the ground; Activity recognition, extracting motion features in the stance phase to determine the activity type, where the activity type includes at least one of walking, ascending, descending, running, squatting, backwards and unknown; Control strategy, select a corresponding control strategy to control the actuator 12 according to the gait phase and activity type.

[0023] Among them, the stance phase (hereinafter referred to as the "stance phase") is defined as the gait phase when the lower limbs are in contact with the ground, which usually marks the beginning of a complete gait cycle. The occurrence of this phase is universal. Regardless of the type of activity of the wearer (such as walking, ascending, squatting or descending), the gait always starts with the contact of the foot with the ground. Therefore, the stance phase provides a natural time window that can be used as the starting point of the gait cycle to capture and analyze the initial characteristics of the gait.

[0024] Furthermore, during the stance phase, motion parameter sensor data (such as acceleration and knee angular velocity from the thigh inertial measurement unit (IMU)) show a relatively stable trend. For example, the acceleration is close to the acceleration of gravity (about 9.8 m / s²), and the angular velocity is small (usually below a certain threshold, such as 0.5 rad / s). This stability provides a reliable reference signal for feature extraction, in contrast to the fluctuations in subsequent dynamic stages (such as the swing phase).

[0025] On the one hand, the extraction of motion features relies on the key representation of the motion pattern, and the support phase becomes the best time point for identifying the activity type due to its specific biomechanical and dynamic properties. 1. In the support phase, the wearer's body posture (such as knee angle, calf tilt angle, hip position, etc.) directly reflects the type of activity. For example, the knee angle decreases significantly when squatting (e.g., less than 120 degrees), while the calf tilt angle is larger when walking up (e.g., more than 30 degrees). These posture information are most obvious and stable when the foot is in contact with the ground, which is easy to capture through motion parameter sensor data. Therefore, in contrast, the foot is off the ground in the swing phase, and the posture changes are more driven by inertia, which is difficult to directly associate with a specific activity type.

[0026] Second, the motion parameter sensor signal in the support phase has a lower noise level because the contact between the foot and the ground reduces the uncertainty caused by free movement. This stability allows the algorithm to more accurately extract features, such as the smoothness of acceleration, the low amplitude of the knee angular velocity, or the static value of the knee angle. In the swing phase, the signal fluctuates greatly (for example, the acceleration may vary by 10%-50% due to the swing speed), and the features are easily disturbed by environmental factors (such as uneven ground) or individual differences (such as leg swinging habits).

[0027] 3. As the starting point of a gait cycle, the stance phase provides the earliest opportunity to identify the activity type. This early identification is crucial for the detection and control of subsequent gait phases. For example, after identifying the "upward" activity, the algorithm can adjust the switching conditions of the swing phase (such as increasing the knee joint angular velocity threshold) to optimize the phase determination of the entire gait cycle.

[0028] Combination Figure 1 As shown, in one embodiment, the number of motion parameter sensors is 2, which is a pair of IMUs respectively arranged on a pair of swing arms 11. The condition of the knee joint angle can be obtained by converting the motion data such as tilt angle and angular velocity obtained by a pair of IMUs.

[0029] Combination Figure 2 As shown, in one embodiment, the number of motion parameter sensors is 2, which is composed of a knee joint angle sensor and an IMU arranged on one of the swing arms 11, and the IMU is used to capture the angular velocity and acceleration data of the thigh and calf at least. The motion state corresponding to the swing arm without the IMU can be obtained by converting the motion data obtained by the IMU and the knee joint angle change obtained by the knee joint angle sensor. In one embodiment, the knee joint angle sensor is an encoder arranged on the actuator, and the encoder is used to monitor the angle change of a pair of relatively rotating parts on the actuator (such as the stator and rotor of the joint motor), and the pair of relatively rotating parts are connected to a pair of swing arms, so as to obtain the change signal of the knee joint angle and angular velocity.

[0030] Furthermore, in one embodiment, the motion characteristics include knee joint angle, knee joint angular velocity, inclination angle of the calf and / or thigh, angular velocity of the calf and / or thigh, acceleration of the thigh and / or calf.

[0031] Furthermore, in one embodiment, the motion feature includes a hip projection difference, which is the change in the hip position in the current state relative to the lower limb in a preset initial posture. Specifically, it refers to the difference in position change when the hip moves in the horizontal direction (front and back) and the vertical direction (up and down), which is used to capture the overall trend of leg movement. Simply put, it "projects" the movement of the hip onto the ground and the sagittal plane, including the forward and backward movement (X direction) and the up and down movement (Y direction), and then calculates the difference between the two. This difference can characterize whether the wearer's legs tend to move forward, backward, or more raised or lowered. For example, when walking forward, the horizontal projection usually changes more than the vertical direction; when moving backward, the horizontal direction may become smaller or even become negative, while the vertical direction may increase slightly. By measuring the hip projection difference, different motion states can be more clearly distinguished, such as the difference between normal walking and walking backwards. Specifically, combined with Figure 3 As shown, the lower limbs are in a preset initial posture with the thigh and calf in an upright state, that is, the knee joint angle is 180°. The hip projection difference includes the hip projection difference in the X direction and the hip projection difference in the Y direction; The hip projection difference in the X direction can be expressed as the vector sum of horizontal displacements, with the X direction being horizontally forward as the positive direction: ΔX = -[l_shk · cos(θ_shk)+ l_thg · cos(θ_thg)] where: ΔX represents the hip projection difference in the X direction (unit: meter); l_shk, l_thg are the lengths of the calf and thigh respectively (typical values ​​are both 0.42 m); θ_shk, θ_thg are the angles of the calf and thigh relative to the vertical direction (unit: radian); The Y-direction hip projection difference is combined with normalization processing, and the Y-direction is vertically upward as the positive direction, which can be expressed as: ΔY =l_shk · sin(θ_shk) + l_thg· sin(π - θ_thg) - L_total Where: ΔY represents the hip projection difference in the Y direction (unit: meter); l_shk, l_thg are the lengths of the calf and thigh respectively (typical values ​​are both 0.42 m); θ_shk, θ_thg are the angles of the calf and thigh relative to the vertical direction (unit: radian); sin(π - θ_thg); L_total = l_shk + l_thg is the total leg length, which is used for normalization. After normalization, it is suitable for users with different heights or leg lengths.

[0032] The motion features based on hip projection difference also include the rate of change of hip projection difference, that is, calculating the time derivative of hip projection difference, including the differential of hip projection difference in the X direction of hip joint and the differential of hip projection difference in the Y direction of hip joint, and then capturing the rate of change of hip projection difference through a smoothing differentiator to further enhance the description of motion trend.

[0033] In one embodiment, hip projection differences play a role in gait phase detection in the following manner: 1. Switching from support phase to swing phase: When switching from the stance phase to the swing phase, the hip projection difference provides the spatial context of the body's center of gravity and leg position. For example, when the hip projection difference in the X direction becomes negative (e.g., less than -0.12 meters) and the hip projection difference in the Y direction shows an upward deviation (e.g., greater than -0.08 meters), combined with the knee angular velocity exceeding 1.5 rad / s, the system determines that the leg begins to leave the ground and enters the swing phase.

[0034] The dynamics of the hip projection differences (their first order differentials with respect to time) further aid in determining the timing of the switch. For example, when the first order differential of the hip projection difference on the x-axis is > 0 and the first order differential of the hip projection difference on the y-axis is < -0.1, it indicates that the hip is moving forward and the leg is lifted upward, reinforcing the identification of the swing phase.

[0035] 2. Switching from swing phase to support phase: When switching from the swing phase to the stance phase, the hip projection difference is used to detect the spatial motion characteristics of the foot landing. For example, when the hip y-axis hip projection difference reaches a positive peak value (for example, greater than 0.15 meters) and its first-order differential value (for time) is greater than 0.25, it indicates that the leg has completed the swing and re-contacted the ground, triggering the judgment of the stance phase.

[0036] The extreme values ​​of the hip projection differences combined with a time window (e.g., within 150 ms after the peak) ensured robustness of the switching conditions.

[0037] 3. Enhancement of activity type context: Hip projection difference is not only used for phase switching, but also supports activity type identification through its trend and amplitude. For example, when walking up, the hip projection difference in the X direction is usually small (e.g., less than 0.05 meters) while the hip projection difference in the Y direction shows a significant negative value (e.g., less than -0.07 meters), reflecting the spatial characteristics of the leg-lifting action. These motion features are used to dynamically adjust the switching threshold to adapt phase detection to specific activities.

[0038] Compared to other movement characteristics, hip projection difference has the following unique advantages: 1. Comprehensiveness of spatial context: Hip projection difference directly reflects the two-dimensional displacement (x, y direction) of the leg relative to the body's center of mass, providing more comprehensive spatial information about the movement than a single joint angle or angular velocity. For example, the knee joint angle only describes the local joint state, while the hip projection difference captures the overall coordination of the hip and leg. This global perspective is particularly important in complex activities (such as walking up or down stairs) because these activities involve significant body center of mass movement.

[0039] 2. Sensitivity to center of gravity movement: Hip projection differences are highly sensitive to changes in the wearer's center of gravity and can effectively distinguish between stability in the stance phase and dynamics in the swing phase. For example, during walking, the change in the x-axis projection of the hip from negative to positive reflects the trend of the center of gravity moving from the back to the front, while the negative change in the y-axis projection of the hip indicates the height of the leg lift. This dual-axis motion feature is superior to relying solely on knee angular velocity, which may lose discrimination due to individual gait differences.

[0040] 3. Robustness and environmental adaptability: Hip projection difference is based on geometric calculations (trigonometric functions) and is less sensitive to sensor noise (such as acceleration drift in IMU). In contrast, acceleration motion characteristics are easily affected by external disturbances (such as uneven ground), while hip projection difference maintains high stability through standardized calculation of leg length. This robustness makes it more advantageous in dynamic and changeable scenarios such as mountain climbing.

[0041] 4. Implementation of low computational overhead: The calculation of hip projection difference only involves basic trigonometric operations and addition and subtraction, with low time complexity and no need for complex filtering or iterative optimization. Compared with the combined motion features based on acceleration and angular velocity (such as deviation ratio or curvature change), hip projection difference provides sufficient differentiation ability while maintaining high efficiency, which meets the design goal of the lightweight algorithm of the present invention.

[0042] If the hip projection difference is removed and gait phase detection is performed based on other motion features, the following undesirable consequences may occur: 1. Phase switching accuracy decreases: Without the spatial context of the hip projection difference, the system may have difficulty accurately judging the position of the legs relative to the body. For example, in the stair-walking activity, if only the angular velocity is relied upon and the positive peak of the hip x-axis projection is ignored, the brief leg adjustment may be mistakenly judged as the swing phase, resulting in the incorrect application of the control strategy (such as the incorrect application of transparent control in the stance phase).

[0043] 2. Weakened ability to distinguish between activity types: Hip projection difference provides key spatial motion features in activity recognition, such as the negative trend of hip projection difference in the Y direction when walking up. Without this motion feature, the system may not be able to effectively distinguish between walking and walking up (the angular velocity patterns of the two may be similar), thus affecting the accuracy of dynamic threshold adjustment and ultimately leading to inappropriate power output.

[0044] 3. Reduced robustness: In complex terrain (such as uneven ground in mountain climbing), angular velocity and acceleration are easily disturbed by noise, while the hip projection difference maintains relative stability through geometric relationships. If the hip projection difference is not used, the system's adaptability to environmental changes will be significantly reduced, and it may frequently enter unknown phases under abnormal conditions (such as temporary distortion of sensor signals), reducing the safety and comfort of the wearer.

[0045] 4. Impaired transition smoothness: The rate of change of the hip projection difference (first-order differential in time) provides a smooth dynamic basis for gait phase switching. Without this motion feature, the switching condition may be too dependent on instantaneous values ​​(such as angular velocity thresholds), resulting in jitter or mutation in gait phase transition, affecting the continuity of exoskeleton control and user experience.

[0046] Therefore, the hip projection difference (x, y direction) significantly improves the accuracy and adaptability of gait phase detection by providing spatial context, center of gravity sensitivity and robustness. Its unique advantage is that it achieves efficient representation of complex motion patterns at low computational cost, enabling the system to maintain stable phase switching and control output in a variety of activities (such as climbing and ascending). If the hip projection difference is not used, the system will face problems such as reduced accuracy, weakened adaptability and uneven transition, which limits its performance in dynamic scenes.

[0047] In one embodiment, the hip projection difference plays a role in activity type determination in the following ways: The hip projection difference in the X and Y directions is a spatial displacement index calculated based on the angle and length of the calf and thigh, and is used to characterize the position change of the knee exoskeleton during movement. By combining the hip projection difference and its differential (i.e., the rate of change of the hip projection difference over time), the user's activity type can be effectively determined, such as going down stairs, going up stairs, or walking. These hip projection differences reflect the relative movement trajectory of the hip in the horizontal (X direction) and vertical (Y direction), which is closely related to the leg posture and movement dynamics.

[0048] In the activity type judgment, the X-direction hip projection difference is mainly used to detect the displacement movement characteristics in the horizontal direction. For example, in the activity of going down stairs, the X-direction hip projection difference usually shows a small positive or negative offset. Combined with its differential, it can further capture the dynamic changes of the movement, thereby distinguishing the fast and slow stair descent patterns. In walking activities, the X-direction hip projection difference usually shows a periodic negative offset, reflecting the movement characteristics of the legs moving forward. In the activity of going up stairs, the X-direction hip projection difference may show a large positive offset, indicating the action of lifting the legs and moving forward.

[0049] The Y-direction hip projection difference mainly reflects the displacement change in the vertical direction and is often used to detect activities related to height changes. For example, when going down stairs, the Y-direction hip projection difference is usually negative, indicating that the hip moves downward relative to the initial position; when going up stairs, the Y-direction hip projection difference may show a change from negative to positive, reflecting the process of lifting the leg and landing. By normalizing the Y-direction hip projection difference (minus the total leg length), the influence of individual leg length differences can be eliminated, making the judgment result more universal.

[0050] By integrating the hip projection difference in the X and Y directions and its differential, the system can construct a multi-dimensional motion feature space, which is combined with other sensor data (such as knee angle, calf and thigh angle, gyroscope data, etc.) to form the discrimination conditions of the activity type. This method can capture the spatiotemporal characteristics of the movement, thereby achieving accurate classification of different activity types.

[0051] In this regard, hip projection has the following unique advantages: 1. Spatial intuitiveness: Hip projection difference directly reflects the displacement of the hip in three-dimensional space, providing more intuitive motion trajectory information than a single angle or speed, and helping to distinguish activities with similar angular motion characteristics but different spatial paths (such as walking and going down stairs).

[0052] 2. Dynamic adaptability: By introducing the differential of the hip projection difference, the acceleration and trend changes of the movement can be captured, making the system more sensitive to the detection of dynamic activities (such as rapid descent or sudden leg lifting when climbing stairs).

[0053] 3. Robustness: The hip projection difference combines the length and angle information of the calf and thigh, which can offset the influence of sensor noise or individual posture differences to a certain extent. It is more stable than directly using the original sensor data.

[0054] 4. Multidimensional synergy: The combination of hip projection differences in the X and Y directions provides a description of motion on a two-dimensional plane, which complements other motion features (such as knee angular velocity) to enhance the accuracy and specificity of activity classification.

[0055] If the hip projection difference in the X and Y directions is not used, the activity type determination may face the following problems: 1. Reduced resolution: Relying only on a single motion feature such as angle or angular velocity may not effectively distinguish activities with similar spatial trajectories but different dynamics. For example, walking down stairs and walking may have similar knee angle ranges, but the horizontal and vertical displacement patterns of the hips are significantly different. Removing the hip projection difference will lead to an increase in the misjudgment rate.

[0056] 2. Lack of dynamic information: The differential of the hip projection difference provides real-time changing information on the movement trend. If it is not used, the system may find it difficult to capture fast movements or transition states (such as switching from standing to going down stairs), thus affecting the response speed and stability of the control algorithm.

[0057] 3. Reduced adaptability: The normalized design of hip projection difference makes it suitable for users of different heights or leg lengths. If hip projection difference is not used, the system may require additional calibration steps to adapt to individual differences, increasing the complexity of use.

[0058] 4. Limited control accuracy: In the control of knee exoskeleton, hip projection difference provides a key basis for gait phase switching and torque distribution. Without this information, the control algorithm may not accurately match the user's movement intention, resulting in insufficient or excessive assistance from the exoskeleton, affecting comfort and safety.

[0059] In summary, hip projection difference plays an irreplaceable role in judging the activity type and can significantly improve the control performance of the knee exoskeleton in complex terrain and dynamic activities.

[0060] Furthermore, in one embodiment, the motion feature includes the product of thigh acceleration and knee angular velocity to characterize the mechanical interaction in motion. Specifically, the product of thigh acceleration and knee angular velocity is designed as an intermediate variable. The core idea is to extract more discriminative features by fusing two different dimensions of motion parameters (linear acceleration and angular velocity) to enhance the accuracy of activity type recognition and gait phase detection. Specific applications in some embodiments include the following aspects: 1. Characterization of Mechanical Interactions This feature directly reflects the dynamic interaction between the linear motion of the thigh and the rotational motion of the knee joint. For example, in the upward activity, when the wearer raises his leg, the thigh acceleration may show a positive peak (e.g., 12 m / s²), while the knee angular velocity increases rapidly (e.g., 2.0 rad / s), and the product can reach 24 m·rad / s³, indicating strong mechanical coupling. In level walking, due to the lower angular velocity of the knee joint (e.g., 0.8 rad / s), even if the acceleration is similar, the product value is smaller (e.g., 9.6 m·rad / s³). This difference provides a significant distinction for activity classification.

[0061] 2. Enhancement of dynamic trends Through the product operation, this feature amplifies the dynamic changes during movement. For example, when the swing phase quickly transitions to the support phase (such as running landing), the thigh acceleration may drop rapidly from a high value (for example, from 15 m / s² to 9.8 m / s²), and the knee joint angular velocity changes from a positive value to near zero (for example, from 1.5 rad / s to 0.2 rad / s). The rapid change of the product (from 22.5 m·rad / s³ to 1.96 m·rad / s³) can sensitively capture this trend, thereby assisting in the accurate determination of gait phase switching.

[0062] 3. Robustness of noise suppression When using thigh acceleration or knee angular velocity alone, the signal may be affected by environmental noise (such as acceleration jitter caused by uneven ground) or sensor drift. By multiplying the two, this engineering feature smoothes the transient anomalies of a single variable to a certain extent. For example, if the acceleration has an abnormal peak value (such as 20 m / s²) due to a short-term disturbance, but the knee angular velocity remains low (such as 0.3 rad / s), the product value (6 m·rad / s³) remains within a reasonable range, avoiding misjudgment.

[0063] In a specific application in activity recognition, when extracting motion features in the stance phase to determine the activity type, the product of thigh acceleration and knee joint angular velocity is used as a key discriminant indicator. In some embodiments, for example: Level walking: The product value shows periodic fluctuations, with a typical range of 5 to 15 m·rad / s³, reflecting a stable gait rhythm.

[0064] Upward: This is manifested by a significant increase in the product value during leg lifting (e.g. 20 to 30 m·rad / s³) because both acceleration and angular velocity reach their peak values ​​simultaneously.

[0065] Squat: This is manifested by a small product value and a long duration (e.g. 2 to 5 m·rad / s³ for about 500 milliseconds), reflecting the low-speed, high-load characteristics.

[0066] Downward: characterized by moderate and slowly varying product values ​​(e.g., 10 to 20 m·rad / s³), due to low angular velocity and stable acceleration.

[0067] By setting threshold rules and combining them with time window analysis (such as average value within 300 milliseconds), the system can effectively distinguish different types of activities.

[0068] Role in control strategies The product of thigh acceleration and knee angular velocity also directly affects the actuator control strategy selection. For example: When it is above a certain threshold (such as 25 m·rad / s³), it indicates that the wearer is in a high-dynamic activity (such as running or walking), and the system can choose hybrid control to combine elasticity and damping effects to provide strong assistance.

[0069] When it is low and stable (e.g. 3 to 8 m·rad / s³), indicating that the wearer is in a quasi-static state (such as squatting or standing), the system can use elastic control to simulate the spring effect.

[0070] Compared with using only thigh acceleration or knee angular velocity, the product of thigh acceleration and knee angular velocity has the following advantages: 1. Information fusion: Integrate linear and rotational motion information in the form of products to provide a richer mechanical description than a single variable.

[0071] 2. Dynamic sensitivity: The product operation amplifies the coordinated changes of the two variables, making the feature more responsive to motion intentions.

[0072] 3. Computational efficiency: Only one multiplication operation is required to generate features, which meets the design goal of lightweight algorithms and is suitable for real-time processing in embedded systems.

[0073] 4. Strong adaptability: The range and trend of the product value can adapt to the movement habits of different wearers (such as stride or speed differences) without the need for additional calibration.

[0074] If the product of thigh acceleration and knee angular velocity is removed and only the original variables are relied upon, the following problems may occur: Discrimination ability is reduced: A single variable cannot capture the coupling effects of linear and rotational motion simultaneously, and may confuse dynamically similar activities (such as horizontal walking and slow ascent).

[0075] Insufficient dynamic detection: Lacking the amplification effect of the product feature on the changing trend, the system may miss the key moments of rapid transitions (such as swings to support).

[0076] Imprecise control: The control strategy may not accurately match the movement requirements due to the lack of comprehensive mechanical indicators, resulting in insufficient or excessive assistance.

[0077] Further, in one embodiment, the motion feature includes a normalized thigh acceleration-knee joint angular velocity ratio. The feature robustness is enhanced by applying an upper threshold to limit the influence of outliers.

[0078] The core of the normalized thigh acceleration-knee angular velocity ratio lies in the positive and negative angular velocity of the knee joint, which reflects whether the knee joint is in flexion (for example, the angular velocity is positive when the leg is raised) or extension (for example, the angular velocity is negative when the leg is lowered). By distinguishing this directionality, the system can better understand the wearer's intention of leg movement. For example, in the swing phase, the knee joint usually flexes quickly, while in the stance phase, the knee joint may slowly extend or remain stable.

[0079] On this basis, the relationship between thigh acceleration and knee angular velocity is further combined. Specifically, it generates a ratio by comparing the linear acceleration of the thigh with the rotational speed of the knee joint. This ratio is not simply the division of the two, but is normalized to ensure that its value is within a controllable range to avoid inconsistencies caused by differences in the magnitude of the original data (such as acceleration is usually measured in meters per second squared, while angular velocity is measured in radians per second). The normalized ratio can more intuitively reflect the relative strength of the two. For example, when the leg is raised quickly, the thigh acceleration and knee angular velocity may be high at the same time, while when squatting slowly, both may be low. In addition, an upper threshold is introduced to limit extreme values ​​caused by sensor noise, external interference, or abnormal movements of the wearer. For example, when the wearer is suddenly impacted by external force, the thigh acceleration may have a short abnormal peak, while the knee angular velocity remains normal. The upper threshold can limit the ratio to a reasonable range to avoid system misjudgment.

[0080] This feature performs well in many application scenarios of lower limb exoskeletons, especially for activity type recognition and gait phase detection. For example: Gait analysis during horizontal walking: During normal horizontal walking, the knee angular velocity is positive during the swing phase (knee flexion) and close to zero or negative during the stance phase (knee extension). The normalized ratio can highlight the synergy between the rapid increase in thigh acceleration and angular velocity during the swing phase, helping the system to accurately distinguish between the swing and stance phases.

[0081] Dynamic capture when walking up: When walking up, the knee joint angular velocity shows a significant positive value during the leg lifting phase, and the thigh acceleration increases significantly due to the leg lifting. The normalized ratio reflects this high dynamic state and avoids abnormal fluctuations caused by uneven ground or excessive force by the wearer through the upper threshold.

[0082] Stability judgment when squatting or standing: During squatting, the knee angular velocity is usually negative (the knee extends slowly), and the thigh acceleration is also low and stable. The value of the normalized ratio is small and changes slowly, indicating a low-speed movement state, and the system can be adjusted to a low-power mode accordingly.

[0083] Abnormal motion filtering: When the wearer's thigh acceleration suddenly surges due to a fall or external impact, the knee joint angular velocity may not change synchronously, and the normalized ratio may appear abnormally high. After the upper threshold is intervened, the ratio is limited to a reasonable range to prevent the system from mistaking it for some kind of high-dynamic activity.

[0084] The reasons for improving robustness include: Positive and negative knee angular velocity: By focusing on the directionality of the angular velocity, the feature can naturally distinguish different phases of motion and avoid the ambiguity of a single numerical value. For example, even if the absolute value of the angular velocity is the same, the positive and negative differences can clearly distinguish between flexion and extension intentions.

[0085] Normalization: Normalization ensures that the ratio is not directly affected by the dimension of the sensor data or individual differences of the wearer (such as weight, stride length). For example, a heavier wearer may produce greater acceleration, but after normalization, the ratio is still consistent with the trend of a light wearer, enhancing the versatility of the feature.

[0086] The protective function of the upper threshold: The upper threshold acts as a safety valve to prevent abnormal values ​​from interfering with the system's judgment. For example, if the acceleration suddenly soars due to a bump on the ground during running, the threshold mechanism can prevent the ratio from getting out of control and ensure that the system still operates in normal running mode.

[0087] The externally normalized thigh acceleration-knee joint angular velocity ratio can support the optimization of the control strategy of the lower limb exoskeleton. In some embodiments, for example: When the ratio is high and the knee angular velocity is positive, indicating that the wearer is in a state of rapid leg lifting (such as running or walking), the system can increase the actuator's power output to provide greater thrust.

[0088] At low ratios and negative angular velocity, indicating that the wearer may be slowly lowering their legs (such as when squatting or walking down stairs), the system can switch to damping mode, enhancing stability and reducing impact.

[0089] When the ratio triggers the upper threshold due to an abnormal value, the system can temporarily maintain the current control strategy to avoid frequent mode switching due to short-term interference, thereby improving the continuity of the wearing experience.

[0090] Furthermore, in one embodiment, the motion characteristics include accumulating the positive knee joint angular velocity to quantify the continuous trend of the knee joint motion; when the knee joint angular velocity is negative, the accumulated value is reset.

[0091] Furthermore, in one embodiment, the motion characteristics include the maximum / minimum values ​​of the angles and angular velocities of the knee joint, thigh, and calf, characterizing the peak trend of the motion.

[0092] Furthermore, in one embodiment, the control strategy includes at least one of elasticity control, damping control, and hybrid control; wherein: The elastic control is to control the torque of the actuator 12 acting on the swing arm 11 according to the knee joint angle to simulate the spring; The damping control is to control the torque of the actuator 12 acting on the swing arm 11 according to the angular velocity of the knee joint to simulate a damper; Hybrid control combines elasticity and damping control to suit the current activity type; The control strategy corresponding to the support phase includes at least one of elasticity control, damping control, and mixed control.

[0093] When executing the control strategy, the control parameters are adjusted in real time according to the current gait phase and / or activity type and / or active motion characteristics, such as adjusting the elastic force of the simulated spring or the damping force of the simulated damper.

[0094] Regarding the control strategy corresponding to standing support, it should be noted that: During the stance phase (hereinafter referred to as the "stance phase"), the system detects the motion state of the foot when it contacts the ground and applies specific control strategies to optimize the wearer's stability, comfort and efficiency. These strategies include elasticity control, damping control and hybrid control, each designed for different motion needs and scenarios.

[0095] For elastic control, the system applies torque based on the joint angle to simulate the behavior of a virtual spring. For example, when the wearer is in a half-squat posture (knee angle is about 90 to 120 degrees, which usually occurs from standing to squatting), the system detects this angle change, indicating that the leg is in a static load or quasi-static transition state. At this time, the system applies an auxiliary torque proportional to the angle deviation (for example, the torque size is 5 to 15 Nm, depending on the wearer's weight and joint stiffness requirements) to relieve the burden on the leg muscles and simulate the natural rebound effect of the spring. The benefit of this control is that it not only enhances the wearer's strength by providing assistance, but also reduces energy consumption when recovering the posture (such as getting up from squatting), while avoiding discomfort caused by excessive joint bending. The joint angle is detected by an inertial measurement unit (IMU), which combines accelerometer and gyroscope data to calculate the relative angle of the legs in real time to ensure detection accuracy and response speed.

[0096] For damping control, the system applies torque based on the angular velocity of the knee joint to simulate the cushioning effect of the damper. For example, when the wearer walks on uneven ground or squats quickly, the angular velocity of the knee joint may reach 0.5 to 1.5 rad / s, indicating that the leg is undergoing dynamic adjustment or external disturbances (such as ground impact). In this scenario, the system applies a reverse torque proportional to the angular velocity of the knee joint (for example, 3 to 10 Nm, depending on the angular velocity of the knee joint) to slow down the joint movement and avoid instability or muscle fatigue caused by excessive bending. The advantage of this control is that it can smooth the motion trajectory and suppress unnecessary oscillations (such as jitter during walking), thereby improving the safety and comfort of the wearer.

[0097] Hybrid control combines elasticity and damping effects and dynamically adjusts according to the type of activity. For example, in the scenario of "carrying heavy objects and squatting", the wearer may first enter a half squat (triggering elastic control to provide 10 Nm of spring torque assistance), and then produce rapid joint adjustments due to the shaking of the heavy object (knee joint angular velocity reaches 1.0 rad / s, triggering damping control to apply 5 Nm of buffering torque). This combined strategy ensures that the system adaptively matches the needs of the wearer in complex activities by providing assistance and stability at the same time. Its advantage is that it combines the strength enhancement of elasticity with the smoothness of damping movement, and can adapt to a variety of support phase scenarios (such as walking up, squatting or walking with weight), thereby improving overall movement efficiency and wearing experience.

[0098] The fundamental reason for adopting these control strategies in the stance phase is that this gait phase is the key time point for activity type identification and load support. The wearer's movements (such as walking, squatting, or standing adjustment) usually start from the stance phase, when the joint angles and knee angular velocities provide rich motion information, allowing the system to accurately judge the activity intention and optimize the torque output.

[0099] In addition, the stability and power-assistance requirements of the stance phase directly affect the smoothness of the subsequent swing phase, so through the synergy of elasticity, damping, and hybrid control, the system not only improves the performance of the current gait, but also lays the foundation for phase transition.

[0100] Furthermore, in one embodiment, the control strategy includes: at least one of transparent control and resistance reduction control; wherein: The control strategies corresponding to the swing phase include transparent control and / or resistance reduction control.

[0101] Regarding the control strategy corresponding to the swing, it should be noted that: During the swing phase, when the foot leaves the ground and the leg is in free motion, the system is designed to minimize intervention or provide mild enhancement to support natural movement and improve efficiency. Control strategies include transparent control and drag reduction control, each targeting different dynamic needs.

[0102] Transparent control allows the wearer to move their legs in a natural way by setting the torque output to zero. For example, in normal walking, when the leg enters the swing phase (such as the process from heel off the ground to toe touching the ground), the knee joint swings freely at an angular velocity of approximately 0.8 to 1.2 rad / s. At this time, the system does not apply any active torque and relies only on the low friction design of the device (such as achieved through efficient motors and transmission systems) to ensure that the wearer does not feel any additional resistance. The advantage of this control is that it maximizes the preservation of the wearer's natural gait, avoiding discomfort or energy waste caused by unnecessary intervention, while maintaining the high transparency of the device, making it suitable for daily activities (such as walking or jogging on flat ground).

[0103] Resistance reduction control offsets the effects of friction or gravity by applying a small assist torque. For example, when walking long distances or climbing a slope, the knee joint in the swing phase may increase the wearer's burden due to the weight of the device (usually 2 to 5 kg) or terrain resistance. At this time, the system detects a decrease in the angular velocity of the knee joint (such as less than 0.5 rad / s, indicating that the swing is hindered) and applies a slight assist torque (for example, 1 to 3 Nm, in the same direction as the swing) to compensate for the resistance caused by device friction and gravity. The benefit of this control is that it reduces the wearer's muscle fatigue without changing the natural trajectory of movement, and significantly improves efficiency, especially in long-term activities or high-intensity tasks (such as mountain climbing).

[0104] The reason for choosing these control strategies for the swing phase is that the core requirements of this gait phase are freedom and efficiency. Transparent control meets the requirements of natural movement through zero torque output, while drag reduction control optimizes the wearer's experience by fine-tuning the torque to compensate for the inherent physical limitations of the device (such as mass or friction). In addition, the control of the swing phase directly affects the continuity of the gait cycle. Lightweight intervention strategies can ensure seamless connection with the support phase and avoid gait disorders caused by excessive control.

[0105] Furthermore, in one embodiment, the gait phase also includes an unknown phase, and the unknown phase corresponds to at least one of the following situations: the motion characteristics do not clearly meet the transition state during the support or swing phase, the algorithm cannot determine the current gait, and the algorithm is initialized; the control strategy of the unknown phase includes transparent control and / or resistance reduction control to ensure safety and comfort.

[0106] Regarding the control strategy corresponding to the unknown, it should be noted that: When the system cannot clearly determine the current gait phase (i.e., it is in an unknown phase), transparent control or resistance reduction control is preferred to ensure safety and comfort. This strategy is suitable for scenarios with ambiguous gait transitions or uncertain motion parameter sensor signals.

[0107] In the unknown phase, the system defaults to transparent control and maintains the torque output at zero. For example, when the wearer suddenly starts walking from standing, but the initial motion parameter sensor data (such as IMU signals) cannot accurately distinguish between support and swing due to noise or temporary occlusion, the system avoids applying any active torque and only relies on the passive characteristics of the device to support movement. The benefit of this method is that it avoids the risk of misjudgment through zero intervention (such as tripping due to incorrect application of damping torque in the swing phase), thereby ensuring the safety of the wearer and the smoothness of movement.

[0108] If a slight movement trend is detected (such as a knee angular velocity between 0.2 and 0.5 rad / s, indicating that a swing or fine-tuning posture may be entered), the system can switch to resistance reduction control and apply a small torque (for example, 0.5 to 2 Nm) to support the potential movement intention. For example, when the wearer slowly adjusts his stance or prepares to take a step, this light assistance can offset the friction of the device and ensure smooth movement initiation. The advantage of this strategy is that it can still provide moderate support under uncertainty while keeping the intervention to a minimum to avoid disturbing the wearer.

[0109] The reason for adopting these control strategies in the unknown phase is safety and robustness. When the gait phase is unclear, any aggressive control (such as high torque output) may cause discomfort or even danger, so priority transparent control can effectively avoid risks. The drag reduction control serves as a conservative supplement to ensure that the system is still responsive under possible movement intentions. This design not only improves the fault tolerance of the system, but also maintains its reliability and wearing comfort in complex or non-standard scenarios (such as irregular gait or sudden movements).

[0110] Furthermore, in one embodiment, the gait transition condition and / or the control parameter corresponding to the control strategy are adjusted accordingly according to the activity type and / or movement characteristics, so as to ensure that the exoskeleton can adapt to the changing scenes.

[0111] The motion features extracted in the stance phase are not only used to classify the current activity, but also directly affect the logic of subsequent gait switching: specifically, 1. Transition from the stance phase to the swing phase. The algorithm dynamically adjusts the switching conditions according to the activity type identified in the stance phase. For example, if it is identified as "squatting", a longer acceleration stabilization time (such as 500 milliseconds) is required to confirm the end of the stance phase to avoid misjudging the swing start due to a short stand; if it is identified as "going up", the stabilization time may be shortened (such as 150 milliseconds) and the knee joint angular velocity threshold (such as 1.5 rad / s) may be increased to accommodate rapid leg lifting. This adjustment ensures the accuracy of gait phase determination and the relevance of gait switching.

[0112] 2. Implementation of seamless control: By extracting motion features in the support phase, the system can establish motion context early in the gait cycle, thereby providing continuous instructions for the control of the exoskeleton or walking aid. For example, features identified in the "downward" activity can trigger a gentler actuator response to avoid sudden changes.

[0113] It should be noted that phase transition refers to the process in which the exoskeleton system switches from one gait phase (such as the stance phase) to another gait phase (such as the swing phase or unknown phase) according to the wearer's real-time motion state. This transition is achieved by detecting specific gait features (such as joint angles, knee joint angular velocity) and comparing them with predefined thresholds or conditions. The core logic of gait phase transition relies on multi-motion parameter sensor data fusion and dynamic threshold judgment to ensure that the system can accurately identify the wearer's motion intentions and adjust the control strategy in a timely manner.

[0114] In the stance phase, the system continuously monitors the filtered value of the knee angular velocity. When this parameter exceeds the preset dynamic threshold (typical value is 0.8 rad / s) and the hip forward projection offset reaches 0.25m, the transition to the swing phase is triggered. This process uses a triple verification mechanism: first, the continuity of the movement trend is captured by the thigh inertial measurement unit, secondly, the joint angle extreme value locking technology is used to suppress instantaneous noise interference, and finally, the phase prediction model of the historical gait data is combined for confidence verification to ensure the reliability of the conversion judgment.

[0115] The adjustment of the conversion logic by the activity type is reflected in two aspects: threshold dynamic compensation and control domain remapping. Taking the ascending scenario as an example, the system increases the knee joint angular velocity judgment threshold to 1.5 rad / s to compensate for the response delay caused by the increase in joint load. At the same time, the kinematic compensation algorithm of the hip forward projection offset is introduced to dynamically correct the forward shift of the body's center of gravity when climbing stairs.

[0116] Furthermore, in one embodiment, the motion characteristics also include frequency domain characteristics of knee joint motion, so that in complex terrain scenarios, the system characterizes the ground roughness through the frequency domain characteristics of knee joint motion (such as tremor components of 2 to 4 Hz), the elastic coefficient of the simulated spring in the dynamic coupling elastic control, and the damping coefficient of the simulated damper in the damping control. For example, when the terrain complexity is too large, that is, the ground roughness exceeds the critical value, the damping coefficient is proportionally increased to the maximum value, and the elastic coefficient is attenuated to 60% of the baseline value, thereby effectively suppressing joint oscillations caused by irregular impacts.

[0117] For low-speed and high-load activities such as squats, the system builds a nonlinear elastic control strategy: when the knee angle enters the 60°-100° working range, an auxiliary torque is applied that is exponentially related to the angle change rate. This torque generates peak assistance (typical value 12Nm) when the angle reaches 90°. In this process, the Y-direction hip projection difference is used as a support phase maintenance condition. When it lasts for more than 0.1m for 200ms, the system will forcefully lock the current control parameters to prevent mis-switching due to unexpected center of gravity fluctuations.

[0118] The abnormal condition handling mechanism adopts a simple safety strategy: when entering an unknown phase, the system automatically switches to transparent control mode.

[0119] In some embodiments, the method of gait phase detection and gait phase switching condition control is specifically performed as follows: Before the switching conditions are adjusted according to the detected activity type, the gait phase switching conditions are based on predefined rules and characteristic trends of the sensor data, as follows: The conditions for switching from the swing phase to the stance phase are: when the sensor detects that the foot is in contact with the ground, for example, the thigh acceleration value tends to be stable (for example, close to the typical range of gravity acceleration, such as around 9.8 m / s²), and the amplitude of the knee joint angular velocity decreases significantly (for example, below a preset threshold, such as 0.5 rad / s), it is determined to enter the stance phase. This is because stable acceleration and low knee joint angular velocity reflect the static characteristics of the movement after the foot contacts the ground, which contrasts with the dynamic changes in the swing phase.

[0120] The switching condition from the stance phase to the swing phase is: when the sensor detects that the foot leaves the ground, for example, the thigh acceleration fluctuates significantly (such as ±10%), and the amplitude of the knee joint angular velocity increases (for example, exceeds a certain preset threshold, such as 1.0 rad / s), it is determined to enter the swing phase. Because the fluctuation of acceleration and the increase of knee joint angular velocity indicate that the foot begins to leave the ground and enters a dynamic motion state.

[0121] The conditions for determining an unknown gait phase are: when the sensor data does not clearly meet the characteristic conditions of the stance phase or swing phase, for example, the combination of the acceleration measured at the thigh and the knee angular velocity is in the fuzzy range of the predefined threshold (for example, the knee angular velocity fluctuates between 0.5-1.0 rad / s and the acceleration is not stable), or the data is interfered by noise, resulting in unclear characteristics, it is determined to be an unknown gait phase. This is the default processing of the algorithm for transitional or uncertain states, which usually occurs at the boundary of gait phase switching or when the algorithm is initialized.

[0122] These unadjusted switching conditions are often based on fixed thresholds and direct trend analysis of sensor data and are applicable to general gait patterns (such as level ground walking).

[0123] Furthermore, when the activity type changes, the gait phase switching condition will be adjusted. That is, when a specific activity (such as walking up, walking down, squatting, etc.) is detected, the gait phase switching condition will be dynamically adjusted according to the activity characteristics to improve the accuracy and adaptability of the detection. In some embodiments, the specific adjustment method is as follows: The activity features extracted by the activity recognition module (such as knee angle, calf tilt angle, hip projection difference, etc.) are used to characterize the context of the current movement.

[0124] Adjust the threshold or feature weights based on the activity type. For example, walking up may require a higher knee angular velocity threshold, while squatting may require a longer stance phase duration.

[0125] In some embodiments, the switching condition is as follows: The switching conditions from swing to support require that the acceleration stabilization time measured at the thigh be shorter (for example, from 200 milliseconds to 150 milliseconds), and the peak angular velocity of the knee joint be higher (for example, from 1.0 rad / s to 1.5 rad / s) to adapt to faster foot landing.

[0126] Switching condition from stance to swing: A larger knee angle change (e.g., more than 60 degrees) was detected as an additional condition, reflecting the leg lifting action.

[0127] In some embodiments, the switching condition is as follows: Switching conditions from swing to support: reduce the knee joint angular velocity threshold (for example, from 1.0 rad / s to 0.8 rad / s) to adapt to a slower falling action.

[0128] Switching condition from support to swing: Extend the acceleration fluctuation detection window (e.g. from 200 ms to 300 ms) to capture a smooth transition.

[0129] For example, when squatting, in some embodiments, the switching condition table is: Prolonged stance phase: The acceleration is required to be stable and the knee angle is significantly reduced (for example, less than 120 degrees) for a period of time exceeding a certain threshold (such as 500 milliseconds) to avoid being misjudged as a short pause.

[0130] By dynamically adjusting the threshold and time window, the switching condition can better match the motion characteristics of a specific activity and avoid misjudgment. For example, when walking up, it can prevent the rapid landing from being misjudged as the swing phase, or when squatting, it can prevent the long support time from being misjudged as the start of the swing.

[0131] To further ensure seamless control of gait phase switching, the following control logic is adopted: Smooth transition mechanism: Use overlapping time windows (e.g., 50-100 milliseconds) at the gait phase switching boundary to smooth motion features. For example, by taking a sliding average (e.g., 5-point or 10-point average) of thigh acceleration and knee angular velocity, switching jitter caused by noise or instantaneous fluctuations can be reduced.

[0132] When a potential transition is detected (e.g., from stance to swing), the algorithm briefly holds the current gait phase state (e.g., an additional 50 milliseconds) before confirming the new gait phase to verify the persistence of the characteristic trend.

[0133] State continuity check: The feature is required to meet the switching condition for a minimum duration (e.g. 100 milliseconds) to avoid triggering false switching due to short-term anomalies (such as sensor jitter). For example, when switching from the stance phase to the swing phase, the switch is confirmed only when the knee joint angular velocity continuously exceeds the threshold and the acceleration fluctuation continues to occur.

[0134] Dynamic threshold update: adjusts the threshold in real time based on the activity type to ensure that the switching condition is consistent with the current motion context. For example, the angular velocity threshold is increased to match the faster swing frequency when running. If the activity type changes (such as switching from walking to walking), the algorithm updates the threshold in the first full gait cycle after detecting the new activity to avoid misjudgment of gait phase during the transition period.

[0135] Fallback mechanism: If an abnormality is detected after switching (for example, the acceleration suddenly stabilizes after switching to the swing phase), the algorithm can fall back to the previous gait phase in a short time (for example, within 100 milliseconds) and mark it as an unknown gait phase, waiting for the data to stabilize before re-determining. This mechanism is particularly suitable for transition states or unstable sensor signals.

[0136] Result of seamless control: Through the above logic, this method ensures the continuity and accuracy of gait phase switching and avoids interruption of control signals. For example, in the application of exoskeleton or walking aid, the smoothness of gait phase switching can be directly converted into seamless response of motor control, improving the comfort and safety of the wearer.

[0137] Furthermore, in one embodiment, the process of activity recognition includes: Threshold rule-based feature comparison, which uses predefined threshold rules to compare motion features to determine whether they meet the conditions for a specific activity; Extreme value detection within a time window: Use a time window (e.g., 100 to 350 milliseconds) to analyze the extreme values ​​of motion features (such as maximum or minimum values) to capture dynamic changes in motion. For example, by detecting the maximum value of the calf inclination angle and the peak value of the hip projection difference, it is determined whether there is a characteristic pattern related to upward or downward movement, and then filtered by time conditions (e.g., the duration after the peak occurs) to ensure the robustness of the detection; Cumulative counting and state continuity check: accumulate counts of motion features that meet preset conditions, such as the number of times the calf angle peak exceeds a certain threshold, or the duration of knee joint angular velocity remaining within a certain range. Through state continuity check (for example, requiring a certain combination of motion features to continue to meet the conditions for more than a certain period of time), misjudgment caused by short-term noise or abnormal data can be avoided; Dynamic update,During the stance phase, the motion features are dynamically updated, such as calculating the product of knee angular velocity and hip projection difference, or feature integration based on time windows, to characterize the trend and intensity of the movement. Through these dynamically updated motion features, the activity types can be distinguished; for example, ascending may show greater knee angle changes and negative hip projection difference, while descending may show more stable knee and thigh angular velocities and a specific range of shank angles.

[0138] Conditional logic combination, using conditional logic combination (such as "and" and "or" logic) to jointly analyze multiple motion features. For example, if the knee angle exceeds a certain threshold, the knee angular velocity shows a positive trend, and the hip projection difference is less than a negative value, it may be judged as an upward activity. This logic combination avoids complex calculation models and only relies on basic mathematical operations (such as addition, subtraction, multiplication, division, and comparison), thereby achieving lightweight.

[0139] Therefore, in this application, the process of activity recognition is a lightweight analysis method based on the trend of motion parameter sensor data. It does not rely on resource-intensive technologies (such as machine learning or deep learning), but is based on direct feature extraction and threshold logic judgment of real-time motion parameter sensor data. Through predefined rules and time window analysis, the computational complexity is reduced to ensure efficient operation on embedded devices. All feature calculations use low-overhead mathematical operations (such as addition, multiplication, comparison), combined with cached data in a limited time window, to avoid large-scale data storage and processing.

[0140] The above lightweight analysis method can effectively distinguish various types of activities. For example: Level walking: manifested by stable fluctuations in knee angular velocity and changes in knee angle.

[0141] Ascending: manifested by greater knee angle variation, negative Y-direction hip projection difference, and specific shank tilt angle.

[0142] Downward movement: characterized by a steady knee angular velocity and a smaller calf tilt angle.

[0143] Running: It is manifested by high-frequency fluctuations in the angular velocity of the knee joint and a significant increase in the acceleration of the thigh. Specifically, during running, the angular velocity of the knee joint shows rapid and periodic alternation between positive and negative, reflecting the rapid flexion of the leg during the swing phase (leg lifting) and the brief extension during the support phase (landing). At the same time, the thigh acceleration reaches its peak in the swing phase, especially when the leg steps forward, showing a large positive change, and is accompanied by a short deceleration shock when landing. The knee joint angle changes in a large amplitude and a stable rhythm, and the calf inclination angle switches frequently with the pace, presenting a highly dynamic movement pattern as a whole.

[0144] Squat: Demonstrated by a significant decrease in knee angle and a longer stance phase duration.

[0145] Reverse movement: The difference in hip projection in the X direction tends to be negative and the difference in hip projection in the Y direction tends to be positive, and the cumulative value of the knee joint angular velocity shows a positive trend.

[0146] Unknown: When the characteristic pattern does not match any known activity criteria, it is classified as unknown.

[0147] Corresponding to the method of the above embodiment, in one embodiment, a lower limb exoskeleton knee joint power assist control system is provided, comprising: The lower limb exoskeleton 1 includes a pair of swing arms 11 and an actuator 12 respectively connected to the thigh and calf. The actuator 12 is used to apply torque to the pair of lower limb exoskeletons 1 to assist the movement of the knee joint; specifically, the actuator 12 is a joint motor located at the knee joint, and the pair of swing arms 11 are respectively connected to a pair of components of the joint motor that rotate relative to the knee joint.

[0148] The detection module is used to obtain detection data input by the motion parameter sensor arranged on the lower limb exoskeleton 1, A motion feature extraction module, used for processing the detection data to extract motion features; A gait phase judgment module is used to determine the current gait phase according to the motion characteristics. The gait phase includes a support phase and a swing phase, wherein the support phase corresponds to the state where the lower limbs touch the ground, and the swing phase corresponds to the state where the lower limbs leave the ground; An activity recognition module, used to extract motion features in the support phase to determine the activity type, where the activity type includes at least one of: walking, ascending, descending, running, squatting, backwards and unknown; The control strategy confirmation module is used to select a corresponding control strategy to control the actuator 12 according to the gait phase and the activity type.

[0149] The technical principles of the present invention are described above in conjunction with specific embodiments. These descriptions are only for explaining the principles of the present invention and cannot be interpreted as limiting the scope of protection of the present invention in any way. Based on the explanations herein, those skilled in the art can associate other specific implementations of the present invention without creative work, and these methods will fall within the scope of protection of the present invention.

Claims

1. A method for assisting control of a lower limb exoskeleton knee joint, characterized in that: The lower limb exoskeleton (1) comprises a pair of swing arms (11) and an actuator (12) respectively connected to the thigh and the calf, wherein the actuator (12) is used to apply torque to the pair of swing arms (11) to assist the movement of the knee joint; The steps include: Motion parameter sensor data processing, acquiring detection data input by a motion parameter sensor arranged on the lower limb exoskeleton (1), and processing the detection data to extract motion features; Gait phase detection, determining the current gait phase according to the motion characteristics, wherein the gait phase includes a stance phase and a swing phase, wherein the stance phase corresponds to the state where the lower limbs touch the ground, and the swing phase corresponds to the state where the lower limbs leave the ground; Activity recognition: extracting motion features in the stance phase to determine the activity type, which includes at least one of: walking, ascending, descending, running, squatting, backwards, and unknown; A control strategy is used to select a corresponding control strategy to control the actuator (12) according to the motion characteristics and / or gait phase and / or activity type.

2. The method according to claim 1, characterized in that: The number of the motion parameter sensors is 2, which are a pair of IMUs respectively arranged on a pair of swing arms (11).

3. The method according to claim 1, characterized in that: The number of the motion parameter sensors is 2, which are a knee joint angle sensor and an IMU arranged on one of the swing arms (11).

4. The method according to claim 1, characterized in that: The control strategy includes: at least one of elastic control, damping control, and mixed control; wherein: The elastic control is to control the torque of the actuator (12) acting on the swing arm (11) according to the knee joint angle to simulate the spring; The damping control is to control the torque of the actuator (12) acting on the swing arm (11) according to the angular velocity of the knee joint to simulate a damper; Hybrid control combines elasticity and damping control to suit the current activity type; The control strategy corresponding to the support phase includes at least one of elastic control, damping control, and mixed control; When executing the control strategy, the control parameters are adjusted in real time according to the current gait phase and / or activity type and / or active motion characteristics.

5. The method according to claim 4, characterized in that: The control strategy includes: at least one of transparent control and resistance reduction control; wherein: Transparent control allows the wearer to move their legs in a natural way by setting the torque output to zero; The drag reduction control is performed by applying an auxiliary torque for at least partially counteracting friction or gravity; The control strategies corresponding to the swing phase include transparent control and / or resistance reduction control.

6. The method according to claim 5, characterized in that: The gait phase also includes an unknown phase, and the unknown phase corresponds to at least one of the following situations: the motion characteristics do not clearly meet the transition state during the support or swing phase, the algorithm cannot determine the current gait, and the algorithm is initialized; the control strategy of the unknown phase includes transparent control and / or resistance reduction control.

7. The method according to claim 1, characterized in that: It also includes adjusting gait transition conditions accordingly according to the current activity type and / or movement characteristics.

8. The method according to claim 1, characterized in that: The motion characteristics include knee joint angle, knee joint angular velocity, inclination angle of calf and / or thigh, angular velocity of calf and / or thigh, acceleration value of thigh and / or calf.

9. The method according to claim 8, characterized in that: The activity identification process includes: Threshold rule-based feature comparison, which uses predefined threshold rules to compare motion features to determine whether they meet the conditions for a specific activity; Extreme value detection within a time window: using the time window to analyze the extreme values ​​of motion features to capture the dynamic changes of motion, and then filtering by time conditions; Cumulative counting and state continuity check: the motion features that meet the preset conditions are cumulatively counted and checked through state continuity to avoid misjudgment; Dynamic update: In the support phase, the motion characteristics are dynamically updated, and the activity types are distinguished through the dynamically updated motion characteristics; Conditional logic combination, use conditional logic combination to jointly analyze multiple motion features.

10. A lower limb exoskeleton knee joint power-assist control system, characterized in that: include: The lower limb exoskeleton (1) comprises a pair of swing arms (11) and an actuator (12) respectively connected to the thigh and the calf, wherein the actuator (12) is used to apply torque to the pair of swing arms (11) to assist the movement of the knee joint; A detection module is used to obtain detection data input by a motion parameter sensor arranged on the lower limb exoskeleton (1), A motion feature extraction module, used for processing the detection data to extract motion features; A gait phase judgment module is used to determine the current gait phase according to the motion characteristics, wherein the gait phase includes a support phase and a swing phase, wherein the support phase corresponds to a state where the lower limbs touch the ground, and the swing phase corresponds to a state where the lower limbs leave the ground; An activity recognition module, used to extract motion features in the support phase to determine the activity type, where the activity type includes at least one of: walking, ascending, descending, running, squatting, backwards and unknown; The control strategy confirmation module is used to select a corresponding control strategy to control the actuator (12) according to the gait phase and the activity type.

Images