Exoskeleton lifting and carrying collaborative assistance control method

Abstract

The application discloses a kind of exoskeleton lifting and conveying synergistic power control method, lifting control method includes bending, lifting waist power;Conveying synergistic power includes the walking power control based on gait prediction.Waist power is bent by hip joint angle Real-time identification bending, lifting action, in combination with load mass and upper body inclination angle, according to action type Adaptive configuration gain, dynamically adjust waist power gain, lifting action introduces load gravity compensation torque, realize continuous, accurate, on-demand power in whole process.Based on timing gait parameter, walking power control constructs multi-speed gait template library, uses LSTM-DTW algorithm to complete step speed mode recognition and gait timing accurate alignment, predicts next time gait parameter, realizes walking leading follow-up power.The application solves the problems of traditional exoskeleton constant power mismatch, poor man-machine cooperation, walking without prediction guide, etc., and can significantly reduce the waist load, improve the walking flexibility and operation safety in the heavy lifting and industrial operation scene.

Description

Technical Field

[0001] This invention generally relates to the field of exoskeleton assistive control technology, and specifically to an exoskeleton lifting and transport collaborative assistive control method. Background Technology

[0002] In industrial production and logistics sorting, manual handling of heavy objects is a common task. Prolonged, high-intensity handling can easily lead to occupational diseases such as lumbar muscle strain and joint damage for workers. Traditional handling equipment, such as forklifts and hydraulic pallet trucks, suffers from poor operational flexibility and limited applicability. Exoskeleton robots, as wearable human-machine collaborative equipment, can provide auxiliary torque to the human body through motor drive, becoming a key technology for solving these problems.

[0003] Under the existing exoskeleton assist control method, when a worker wears an exoskeleton and bends over to lift heavy objects, during the process of moving from an upright standing position to a bent-over position (hereinafter referred to as "bending action") or from a bent-over position to an upright standing position (hereinafter referred to as "lifting action"), the exoskeleton controller can only recognize that the worker is currently in a bent-over position and decide whether to provide a fixed amount of assistance based on whether the worker is in a bent-over position.

[0004] In reality, when workers are bending or lifting heavy objects, the amount of force they exert on their lower back varies significantly depending on the total weight of the object and the angle of their upper body tilt. If the exoskeleton continues to provide constant assistance to the lower back in a fixed manner, it will result in either insufficient or excessive assistance. Summary of the Invention

[0005] In view of the above-mentioned defects or deficiencies in the prior art, it is desirable to provide a method for coordinated assistance control of exoskeleton lifting and transportation.

[0006] On one hand, the present invention provides an exoskeleton lifting assistance control method, comprising: S1: Obtain the current movement type of the exoskeleton wearer, the mass of the object being carried, and the upper body tilt angle of the exoskeleton wearer; S2: Calculate the gain correction coefficient based on the action type, the mass of the object being moved, and the upper body tilt angle of the exoskeleton wearer; S3: Use the gain correction coefficient to correct the waist assist gain when standing upright, and obtain the corrected waist assist gain; S4: Calculate the lumbar torque based on the action type and the corrected lumbar assist gain; S5: Provide assistance to the wearer based on the waist torque.

[0007] According to the technical solution provided by the present invention, the exoskeleton includes a lower limb part and a back plate part, and the lower limb part and the back plate part are rotatably connected by a hip joint; the angle of the hip joint is the angle between the back plate part and the vertical direction; the movement types include: bending movement, lifting movement and walking movement; Obtain the current movement type of the exoskeleton wearer, including: If the sum of the hip joint angle at the previous moment and the angle error threshold is less than the hip joint angle at the current moment, it is determined to be a bending motion. If the difference between the hip joint angle at the previous moment and the angle error threshold is greater than the hip joint angle at the current moment, it is determined to be a lifting motion. If the absolute value of the difference between the hip joint angle at the previous moment and the hip joint angle at the current moment is less than the angle error threshold, it is determined to be a walking motion.

[0008] According to the technical solution provided by the present invention, the exoskeleton further includes a foot device connected to the lower limbs; a pressure sensor is installed on the foot device for detecting the total mass of the wearer and the object. The mass of the object being transported is the difference between the total mass and the mass of the wearer; the upper body tilt angle is equal to the angle of the hip joint.

[0009] According to the technical solution provided by the present invention, the action type, the mass of the object being transported, the upper body tilt angle of the exoskeleton wearer, and the gain correction coefficient satisfy the following relationship:

[0010] in, Y This is the gain correction factor. a 0、 a 1. a 2 represents the gain parameter. θ The angle of the upper body tilt. m For the mass of the object being transported, m 0 represents the ratio of the exoskeleton's rated load weight to gravitational acceleration; the gain parameter is set according to the action type; the gain parameter is set according to the action type; The waist support gain is the product of the waist support gain when standing upright and the gain correction coefficient.

[0011] According to the technical solution provided by the present invention, the gain parameter is set according to the action type in the following way: When the action type is a bending motion, the gain parameter satisfies: a 0、 a 1 and a The sum of 2 is less than 1 and greater than 0, so that the gain function can be used to characterize the functional relationship between the mass of the object being carried and the upper body tilt angle and the gain of the exoskeleton on the waist during the bending motion; When the action type is a lifting action, the gain parameter satisfies: a 0 is greater than 1 and a 1 and a 2 are all greater than 0, so that the gain function can be used to characterize the functional relationship between the mass of the object being carried and the upper body tilt angle and the gain of the exoskeleton on the waist during the lifting action; When the action type is walking, the gain parameter satisfies: a 0 equals 1 and a 1 and a 2 are all greater than 0, so that the gain function is used to characterize the functional relationship between the mass of the object being carried and the upper body tilt angle and the gain of the exoskeleton on the waist during walking.

[0012] According to the technical solution provided by the present invention, the lumbar torque is calculated based on the action type and the corrected lumbar assist gain, including: If the action type is a bending or walking action, the base torque is calculated based on the waist assist gain in the current state; the base torque is used as the waist torque.

[0013] According to the technical solution provided by the present invention, the lumbar torque is calculated based on the action type and the corrected lumbar assist gain, including: If the action type is a lifting action, the basic torque is calculated based on the waist assist gain in the current state; the load gravity compensation torque is calculated based on the mass of the object being moved, and the sum of the basic torque and the load gravity compensation torque is used as the waist torque.

[0014] On the other hand, the present invention provides an exoskeleton transport coordination and assistance control method. During walking, the exoskeleton sensing system detects the current movement of the exoskeleton in real time and obtains the movement signal. The movement signal includes multiple gait parameters that satisfy a temporal relationship. The gait parameters include: gait cycle duration, average angular velocity and amplitude of hip joint angle, and average angular velocity and amplitude of knee joint angle. When the exoskeleton wearer's current movement type is identified as walking based on motion signals, it includes: Q1: Calculate the lumbar torque during walking according to any of the exoskeleton lifting assistance control methods described above; Q2: Construct a gait template library; the gait template library contains time-series curves of various standard gait parameters corresponding to different walking speeds; the time-series curves of standard gait parameters are used to calculate gait parameters at corresponding walking speeds; Q3: During the walking process, multiple real-time gait parameters that satisfy the temporal relationship are extracted from the action signal using a sliding window method; Q4: Based on a two-layer LSTM network and the DTW algorithm, determine the gait parameters of the time series curves of the standard gait parameters corresponding to the multiple real-time gait parameters in the gait template library at the next time step; Q5: Provide assistance to the wearer based on the waist torque during walking and the gait parameters at the next moment.

[0015] According to the technical solution provided by the present invention, a gait template library is constructed, comprising: Q2-1: Collect multiple initial gait parameters from the motion signal; the multiple initial gait parameters satisfy a temporal relationship; Q2-2: Extract standard gait parameters based on the initial gait parameters; Q2-3: Using clustering, the multiple standard gait parameters are divided into various gait modes according to different paces; the gait modes include: slow mode, medium speed mode, and fast mode; Q2-4: Use a high-order Fourier series to fit the standard gait parameters belonging to the same step speed mode to obtain the time series curve of the standard gait parameters; Q2-5: Compile the time-series curves of standard gait parameters for each gait mode into a gait template library.

[0016] According to the technical solution provided by the present invention, based on a two-layer LSTM network and the DTW algorithm, the gait parameters of the time series curves of standard gait parameters corresponding to multiple real-time gait parameters in the gait template library are determined at the next time step, including: Q4-1: Input multiple real-time gait parameters into a two-layer LSTM network to obtain the current gait mode; the two-layer LSTM network is used to predict and output the corresponding gait mode based on the input gait parameters. Q4-2: Using the DTW algorithm, the reference time on the time series curve of the standard gait parameter corresponding to the current gait mode is matched for multiple real-time gait parameters; Q4-3: Based on the reference time and the time series curve of the standard gait parameters, calculate the gait parameters at the next time step after the reference time.

[0017] The beneficial effects of this invention are as follows: To address the technical problem of existing exoskeleton-assisted control methods, which can only recognize bending states and provide constant assistance, but cannot dynamically adjust the assistance based on changes in the mass of the object being carried, the upper body tilt angle, and the type of movement, resulting in insufficient or excessive assistance, this invention adopts the following technical solution: First, the wearer's current movement type, the mass of the object being carried, the upper body tilt angle, and the lumbar assistance gain when standing upright are obtained; then, a gain correction coefficient is calculated based on the movement type, the mass of the object being carried, and the upper body tilt angle, and the current lumbar assistance gain is calculated in conjunction with the lumbar assistance gain when standing upright; finally, the lumbar torque is obtained based on the movement type to provide assistance to the wearer. This solution achieves adaptive calculation of the required lumbar assistance torque based on the wearer's real-time movement type, the mass of the object being carried, and the upper body tilt angle, avoiding the assistance mismatch problem caused by constant assistance; it can provide accurate and continuous assistance support in different working conditions such as bending, lifting, and walking, effectively reducing the burden on the wearer's lumbar muscles and lowering the risk of lumbar injury. Attached Figure Description

[0018] Other features, objects, and advantages of the invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings: Figure 1 This is a flowchart illustrating an exoskeleton lifting assistance control method. Figure 2 This is a schematic diagram of the exoskeleton from a side view. Figure 3 A schematic diagram illustrating the execution process of the control method; The components are: 1. Backplate; 2. Lower limbs; 3. Hip joint; 4. Knee joint; 5. Footplate; 6. Ankle joint. Detailed Implementation

[0019] The present invention will now be described in further detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and not intended to limit it. Furthermore, it should be noted that, for ease of description, only the parts relevant to the invention are shown in the accompanying drawings.

[0020] It should be noted that, unless otherwise specified, the embodiments and features described in the present invention can be combined with each other. The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0021] The process of a worker wearing an exoskeleton to move heavy objects includes: bending over, squatting, stopping in a deep squat, lifting, and straightening up.

[0022] Among them, when a worker wearing an exoskeleton bends over to lift heavy objects, the movement from a vertical standing position to a bent-over position is called "bending movement", the movement from a bent-over position to a vertical standing position is called "lifting movement", and the state in which the upper body tilt angle remains basically unchanged (i.e. the change value is less than the angle error threshold) is called "walking movement".

[0023] Example 1 refer to Figure 2 In this embodiment, the exoskeleton structure includes a back plate 1 and a lower limb 2. The back plate 1 and the lower limb 2 are rotatably connected by a hip joint 3, so that the back plate 1 and the wearer's upper body can rotate forward together, thereby enabling the wearer to bend over and carry objects.

[0024] The lower limb portion 2 has a knee joint 4 corresponding to the knee of the human body, allowing the exoskeleton to move with the wearer's legs. The lower end of the lower limb portion 2 is rotatably connected to the foot device 5 via an ankle joint 6, allowing the exoskeleton to move with the wearer's feet.

[0025] The exoskeleton is also equipped with a sensor system for collecting human motion data, including multiple inertial measurement units (IMUs), absolute encoders at the hip and knee joints (for identifying movement types), and pressure sensors mounted on the foot devices 5. The knee absolute encoders detect the angle between the thigh and calf, which comprise the lower limb. When both knee angles are less than 30 degrees, it is considered a squat; when both angles are within the range of 175° to 180°, and the pressure difference detected by the pressure sensors on the two foot devices 5 is less than 5%, it is considered a standing state. When the pressure difference detected by the pressure sensors on the two foot devices 5 is greater than 5%, it indicates that one foot is not providing support, i.e., a walking state. In this embodiment, the walking state is assumed to have minimal change in upper body tilt angle.

[0026] Multiple inertial sensors (IMUs) are deployed on the back panel 1 corresponding to the back of the human torso, hip joints, and knee joints to collect position information of each part of the human body during movement.

[0027] Absolute encoders are installed at the hip and knee joints to collect joint angle and angular velocity data.

[0028] The pressure sensors correspond to the soles and heels, collecting data on the changes and distribution characteristics of the contact force on the soles of the feet during human movement. When a person wears an exoskeleton and steps on the foot device 5, the pressure sensors can measure the mass of the person. When a person moves an object, the mass of the object is transmitted through the person to the pressure sensors on the foot device 5, thus allowing the measurement of the total mass of the object and the person.

[0029] In this embodiment, the vertical direction is used as a reference, so the angle between the back panel 1 and the vertical direction during the bending process is the upper body tilt angle.

[0030] Based on the above, in order to transmit the detected data to the controller (installed on the exoskeleton), a signal transceiver device is also included (installed on the exoskeleton, transmitting data via conventional electrical signals or wireless signals; for example, Bluetooth signals). The signal transceiver device combines various signals (including time alignment, data format unification, and packaging into frame data) to generate a unified motion signal, which is then transmitted to the processor (in this embodiment, a computing chip or computer) for further processing.

[0031] The motion signal includes at least: the angles of the hip and knee joints at various times, the total mass of the human body and the object, and multiple gait parameters that satisfy a temporal relationship. These gait parameters include: the gait cycle duration, the average angular velocity and amplitude of the hip joint angle, and the average angular velocity and amplitude of the knee joint angle. Therefore, the motion signal includes the angles of the hip joint at the previous and current times.

[0032] Upon receiving the action signal, the processor executes the following steps: refer to Figure 1 This invention provides an exoskeleton lifting assistance control method, comprising: S1: Obtain the current movement type of the exoskeleton wearer, the mass of the object being carried, and the upper body tilt angle of the exoskeleton wearer; Specifically, motion signals are acquired, and the current motion type of the exoskeleton wearer is identified based on the motion signals; the motion signals are obtained by the exoskeleton sensing system detecting the current motion of the exoskeleton; the motion types include: bending motion, lifting motion, and walking motion.

[0033] Specifically, since the exoskeleton includes a lower limb portion and a backplate portion, the lower limb portion and the backplate portion are rotatably connected via a hip joint; the motion signal includes the angle of the hip joint at the previous moment and the current moment. (Reference) Figure 2 The lower limbs are in a vertical position; the hip joint angle is the angle between the backplate and the vertical direction. In this embodiment, to avoid the rotation of the lower limbs affecting the detection of the hip joint angle, the hip joint angle is made consistent with the upper body tilt angle, i.e. Figure 2 Therefore, when the upper body leans backward, the angle of the hip joint is negative.

[0034] Obtain the current movement type of the exoskeleton wearer, including: If the sum of the hip joint angle at the previous moment and the angle error threshold is less than the hip joint angle at the current moment, it is determined to be a bending motion. If the difference between the hip joint angle at the previous moment and the angle error threshold is greater than the hip joint angle at the current moment, it is determined to be a lifting motion. If the absolute value of the difference between the hip joint angle at the previous moment and the hip joint angle at the current moment is less than the angle error threshold, it is determined to be a walking motion.

[0035] Specifically, the angle error threshold typically ranges from 1 to 10 degrees. In some implementations, the angle error threshold is set to 3 degrees.

[0036] Because the movements of the wearer when carrying objects are relatively complex, there may be slight changes in the angle of the hip joint. If assistance is provided for bending or lifting due to slight changes, it may lead to misjudgment and excessive assistance to the waist, resulting in a fall; or it may provide too little assistance to the waist, causing the wearer's waist to have to exert a large force instantly, resulting in waist muscle injury.

[0037] In this embodiment, setting an angle error threshold allows the exoskeleton to adapt to the swaying during the process of carrying objects, providing more accurate waist assistance and reducing misjudgments.

[0038] The mass of the object being moved and the upper body tilt angle of the exoskeleton wearer are identified based on the motion signals. Specifically, the exoskeleton also includes a foot device connected to the lower limbs; the foot device is equipped with a pressure sensor for detecting the total mass of the wearer and the object; the motion signal also includes the total mass. The mass of the object being transported is the difference between the total mass and the mass of the wearer; the upper body tilt angle is equal to the angle of the hip joint.

[0039] Specifically, the mass of the wearer and the object can be input by the wearer into the controller, or detected in the following ways: The pressure sensor is set to zero when the exoskeleton is first activated. After being worn by the human body, the system automatically detects the pressure applied to the pressure sensor by the wearer and obtains the wearer's mass.

[0040] During the process of a wearer carrying an object, the pressure sensor detects a slight change in the total weight of the person and the object. Therefore, the pressure values ​​obtained from multiple measurements will satisfy a distribution (e.g., a normal distribution). In this embodiment, the average of the pressure value distribution is taken as the total mass of the person and the object, and then the mass of the person is subtracted to calculate the mass of the object being carried.

[0041] Based on the above, the object's mass can be automatically detected during movement, without the need for separate detection or manual input of the object's mass.

[0042] S2: Calculate the gain correction coefficient based on the action type, the mass of the object being moved, and the upper body tilt angle of the exoskeleton wearer; The following relationships are satisfied between the action type, the mass of the object being moved, the upper body tilt angle of the exoskeleton wearer, and the gain correction coefficient:

[0043] in, Y This is the gain correction factor. a 0、 a 1. a 2 represents the gain parameter. θ The angle of the upper body tilt. m For the mass of the object being transported, m 0 represents the ratio of the exoskeleton's rated load weight to gravitational acceleration (obtained experimentally during the exoskeleton design process and used as a preset fixed value); the gain parameter is set according to the action type; the gain parameter is set according to the action type.

[0044] Specifically, the gain parameter is set according to the action type as follows: To avoid negative calculated gains, all gain parameters need to be greater than 0; based on this condition, the following situations exist: Case 1: When the action type is a bending motion, the gain parameter satisfies: a 0、 a 1 and a The sum of 2 is less than 1 and greater than 0, so that the gain function can be used to characterize the functional relationship between the mass of the object being carried and the upper body tilt angle and the gain of the exoskeleton on the waist during the bending motion; In some implementations, in case 1, a 0 = 0.6 a 1 = 0.05 a 2=0.05. This setting ensures that the gain is less than the waist assistance gain when standing upright (set according to the wearer's movement habits and manually input into the processor). This avoids excessive assistance, which could suppress the problem of the center of gravity shifting due to excessive bending speed; it also allows the exoskeleton to follow the wearer's downward bending movement, reducing bending resistance.

[0045] Case 2: When the action type is a lifting action, the gain parameter satisfies: a 0 is greater than 1 and a 1 and a 2 are all greater than 0, so that the gain function can be used to characterize the functional relationship between the mass of the object being carried and the upper body tilt angle and the gain of the exoskeleton on the waist during the lifting action; In some implementations, under condition 2, a 0 = 1.2a 1 = 0.15 a 2 = 0.15. Because it's necessary to move objects, sufficient assistance is required. Therefore... a A value greater than 1 ensures that the amount of assistance is close to the actual required amount. The gain is then further adjusted based on the object's mass and the upper body's tilt angle to ensure precise and continuous assistance, effectively reducing the burden on the wearer's lower back muscles and lowering the risk of lower back injury.

[0046] Case 3: When the action type is walking, the gain parameter satisfies: a 0 equals 1 and a 1 and a 2 are all greater than 0, so that the gain function is used to characterize the functional relationship between the mass of the object being carried and the upper body tilt angle and the gain of the exoskeleton on the waist during walking.

[0047] When carrying objects, the wearer may find it difficult to maintain an upright posture due to the size and mass of the object and the center of gravity between the wearer and the object. When carrying an object on their back, the wearer's upper body will often lean forward; when hugging an object, the upper body will lean backward (the angle is negative when leaning backward). If this forward or backward leaning posture is maintained for an extended period without the exoskeleton providing assistance, the wearer's lower back will be under continuous strain, potentially leading to lower back injury.

[0048] Unlike the first two scenarios, in scenario 3, the upper body tilt angle doesn't actually change significantly. This embodiment can provide assistance in a similar manner to the first two methods. For example, specifically regarding the setting of the gain parameter, it can be... a 0=1, so that the assistance provided by the exoskeleton is close to the assistance required to maintain the current state; at the same time, a 1 and a With 2 = 0.1, the gain can be further adjusted based on the upper body tilt angle and the mass of the object, so that the exoskeleton provides the waist force required when moving objects, reducing the wearer's waist exertion.

[0049] S3: Use the gain correction coefficient to correct the waist assist gain when standing upright, and obtain the corrected waist assist gain; Specifically, the lumbar support gain is the product of the lumbar support gain when standing upright and the gain correction coefficient, therefore step S3 includes: S3-1: Determine the gain function based on the action type. The gain function includes a gain correction coefficient and a waist assistance gain when standing upright. It is used to characterize the functional relationship between the mass of the object being carried and the upper body tilt angle and the waist assistance gain of the exoskeleton. Specifically, the gain function is expressed as:

[0050] in, K p Provides support and enhancement for the lower back. K p0 It provides a boost to the waist when standing upright.

[0051] S3-2: Substitute the mass of the object being transported and the upper body tilt angle into the gain function to calculate the current waist assist gain.

[0052] S4: Calculate the lumbar torque based on the described action type and the corrected lumbar assist gain, including: If the action type is a bending or walking action, the base torque is calculated based on the waist assist gain in the current state; the base torque is used as the waist torque.

[0053] If the action type is a lifting action, the basic torque is calculated based on the waist assist gain in the current state; the load gravity compensation torque is calculated based on the mass of the object being moved, and the sum of the basic torque and the load gravity compensation torque is used as the waist torque.

[0054] Specifically, the basic torque τ pd The calculation method is as follows:

[0055] Load gravity compensation torque The calculation method is as follows:

[0056] in, e The posture error is equal to the preset value of the upper body tilt angle minus the actual measured upper body tilt angle. For angular velocity error, K p To provide current support and enhancement for the lower back. K d This is the damping gain (typically taken as 3 to 30 N·m·s / rad). mg For the mass of the object being transported, L The lever arm for the waist (generally set as the distance from the midpoint of the side edge of the back panel to the hip joint). The preset value for the upper body tilt angle is determined by averaging historical data of the upper body tilt angle under the same conditions.

[0057] Therefore, waist torque τ The actual result is:

[0058] S5: Provide assistance to the wearer based on the waist torque.

[0059] In actual assistance process, if the back panel is towards Figure 2 If the body is tilted to the right (i.e., in a bent-over position), the direction of the torque in the waist is... Figure 2 The counter-clockwise direction in the middle; if the back panel part is towards Figure 2 If the body is tilted to the left (i.e., leaning backward), the direction of the torque in the waist is... Figure 2 The clockwise direction in the middle.

[0060] When bending over, the waist torque is relatively small. When not carrying an object, it is only used to overcome its own weight. When carrying an object, it also has to overcome the mass of the object.

[0061] When performing a lifting motion, the waist torque is relatively large. When not carrying an object, it overcomes its own weight and resistance while also providing upward movement power. When carrying an object, in addition to overcoming resistance and its own weight, it also overcomes the mass of the object and provides upward movement power.

[0062] Based on the preceding text, the process of moving heavy objects also includes squatting.

[0063] The squatting motion employs a virtual model control (VMC) approach. By constructing a coordinate system transformation matrix, the spatial position mapping of the multi-joint lower limb parts (including the tibia and femur connected by the knee joint rotation) is realized. Combined with the PD control algorithm, virtual forces in the horizontal and vertical directions are generated. Then, the Jacobian matrix transpose operation is used to convert the virtual forces into torques of the knee and hip joints.

[0064] The specific steps of VMC include: By constructing a coordinate system transformation matrix to map the spatial position of the multi-joint lower limb, the spatial position of the body in the reference coordinate system can be accurately mapped. Figure 2 The horizontal direction is the X-axis and the vertical direction is the Z-axis, providing position feedback for closed-loop control of the lifting height. This transformation matrix... The core calculation formula for the hip joint position vector is as follows:

[0065] in, L 1 represents the length of the tibia. L 2 represents femur length. θ a For the ankle angle, θ k For knee angle, θ h Hip joint angle, k x This represents the positional offset of the lower limbs along the X-axis. k zThis represents the positional offset of the lower limbs along the Z-axis, which directly corresponds to the body height of the robot in its raised state.

[0066] Secondly, virtual forces in the X and Z axes are generated using the PD control algorithm. Vertical virtual force... f z Virtual force in the horizontal direction used to maintain a constant height during the lifting process. f x Both are used to suppress speed disturbances and are achieved through PD control.

[0067] Finally, the virtual force is converted into knee and hip joint torques using the Jacobian matrix transpose operation. First, a single-leg Jacobian matrix is ​​constructed. :

[0068] This matrix is ​​obtained by taking the partial derivative of the joint angle with respect to the position vector, which can correlate joint movement with virtual body velocity in real time, and then based on the force-torque mapping relationship:

[0069]

[0070] in, This is a vector representation of torque. This is the transpose of the matrix. For vector representation of force, f x This is a virtual force in the horizontal direction. f z This is a virtual force in the vertical direction. f x and f z Used for height maintenance and disturbance resistance f θ This is a virtual force for posture, used to control upper body posture stability.

[0071] The final output of knee and hip joint torque drives the actuator to move, achieving stable control in the squatting state. The entire process does not rely on complex robot dynamics models, but relies entirely on the defined VMC calculation method to complete the closed-loop control from position mapping to torque output.

[0072] Example 2 In the application scenario of this embodiment, the wearer needs to move objects after putting on the exoskeleton. The actions required during the entire moving process include: Walk upright to the vicinity of the object's location, and change from an upright standing position to an upper body tilted position; this process involves bending over. When the wearer holds an object and lifts it, the angle of the upper body tilt decreases; while walking while holding an object, the upper body will generally still tilt slightly. When the object is moved to its destination, the angle of the upper body is increased again to put the object down; this process involves bending over again. Finally, it undergoes another lifting motion to restore it to an upright standing position.

[0073] Considering that holding an object may obstruct the wearer's vision and cause inconvenience, this embodiment not only provides lumbar support but also additionally predicts gait parameters for the next moment based on the wearer's walking speed pattern.

[0074] Then, in the next moment, the exoskeleton can perform corresponding walking movements based on the predicted gait parameters. During walking, the wearer can feel the exoskeleton's movements and thus make corresponding walking movements. This achieves walking guidance in situations where the field of vision is obstructed. This embodiment only relates to walking guidance on flat ground when the field of vision is obstructed by carrying goods.

[0075] This invention also provides an exoskeleton transport and collaborative assistance control method. During walking, the exoskeleton sensing system detects the current movement of the exoskeleton in real time and obtains a movement signal; the movement signal includes multiple gait parameters that satisfy a temporal relationship; the gait parameters include: gait cycle duration, average angular velocity and amplitude of the hip joint angle, and average angular velocity and amplitude of the knee joint angle; When the exoskeleton wearer's current movement type is identified as walking based on motion signals, it includes: Q1: Calculate the lumbar torque during walking according to the exoskeleton lifting assistance control method described in Example 1; Q2: Construct a gait template library; the gait template library contains time-series curves of various standard gait parameters corresponding to different walking speeds; the time-series curves of standard gait parameters are used to calculate gait parameters at corresponding walking speeds; Furthermore, a gait template library is constructed, including: Q2-1: Collect multiple initial gait parameters from the motion signal; the multiple initial gait parameters satisfy a temporal relationship; Gait parameters were collected in accordance with the method described in Example 1. During the collection process, the parameters were detected at equal time intervals, and the resulting initial gait parameters satisfied the temporal relationship. The sampling frequency during the collection process was set to 5 to 30 Hz.

[0076] Q2-2: Extract standard gait parameters based on the initial gait parameters; Since gait parameters include the gait cycle duration, the average angular velocity and amplitude of the hip joint angle, and the average angular velocity and amplitude of the knee joint angle, this embodiment uses positive zero-crossing detection to segment single gait cycle samples (initial gait parameters within a gait cycle), including: extracting three-dimensional features of cycle duration, average angular velocity, and amplitude, and eliminating the influence of dimensions through z-score normalization, finally obtaining standard gait parameters.

[0077] The z-score normalization process calculates the standard gait parameters using the following formula:

[0078] in, For standard gait parameters, These are the initial gait parameters. μ The mean of multiple initial gait parameters. represents the standard deviation of multiple initial gait parameters.

[0079] Q2-3: Using clustering, the multiple standard gait parameters are divided into various gait modes according to different paces; the gait modes include: slow mode, medium speed mode, and fast mode; Q2-4: Use a high-order Fourier series to fit the standard gait parameters belonging to the same step speed mode to obtain the time series curve of the standard gait parameters; In this embodiment, a fifth-order Fourier series is specifically used for gait fitting to obtain the time-series curve of standard gait parameters; the fitting formula is:

[0080] in, The time series curves for standard gait parameters are shown. k For order, t For time, a For constant terms, a k and b k The coefficients are the k-th order Fourier coefficients. T This represents the average gait cycle duration for this gait pattern. π Pi is the mathematical constant of a circle.

[0081] The specific fitting process includes: using standard gait parameters as fitting data, and based on the least squares method, aiming to minimize the sum of squared errors between the fitted values ​​and the actual parameter values, obtaining... a , a k , b k The optimal solution.

[0082] Then a, a k , b k Substitute the data into the fitting formula, sample the gait cycle time at equal intervals and calculate the corresponding fitted values, then connect the fitted values ​​to generate a time series curve of standard gait parameters.

[0083] Q2-5: Compile the time-series curves of standard gait parameters for each gait mode into a gait template library.

[0084] Q3: During the walking process, multiple real-time gait parameters that satisfy the temporal relationship are extracted from the action signal using a sliding window method; In some implementations, real-time gait parameters are captured using a 30-frame sliding window.

[0085] Q4: Based on a two-layer LSTM network (two-layer long short-term memory network) and the DTW algorithm (dynamic time warping algorithm), determine the gait parameters of the time series curves of standard gait parameters corresponding to multiple real-time gait parameters in the gait template library at the next time step, including: Q4-1: Input multiple real-time gait parameters into a two-layer LSTM network to obtain the current gait mode; the two-layer LSTM network is used to predict and output the corresponding gait mode based on the input gait parameters. In this embodiment, a two-layer LSTM network for gait pattern recognition has the following structure: The input layer receives the time sequence of standardized gait parameters of the hip and knee joints extracted by a sliding window; the first LSTM layer has 16 hidden units, outputs the sequence and connects to a Dropout layer with a dropout probability of 0.1 to extract primary spatiotemporal features and suppress overfitting; the second LSTM layer has 8 hidden units, takes the output of the last time step, performs deep fusion to reduce dimensionality and mine deep temporal patterns; the output layer consists of a fully connected layer and a Softmax layer, outputs the probability distribution of each gait pattern, and takes the maximum probability as the current gait pattern.

[0086] The dual-layer LSTM network is trained and optimized with multiple sets of gait samples at different speeds, which can quickly capture the temporal features of gait parameters, realize the recognition of gait patterns, and provide accurate positioning for subsequent DTW algorithm template matching.

[0087] Q4-2: Using the DTW algorithm, the reference time on the time series curve of the standard gait parameter corresponding to the current gait mode is matched for multiple real-time gait parameters; Specifically, the DTW algorithm's process includes: The standard gait parameter time series curve corresponding to the walking speed mode is used as the template sequence, and the real-time gait parameter time series extracted by the sliding window is used as the sequence to be matched. Calculate the Euclidean distance between all pairs of points in the two sequences, construct a distance matrix, and the elements in the distance matrix represent the feature similarity of the corresponding pairs of points; Based on the local path constraint rules of right, down, and down-right, a cumulative cost matrix is ​​constructed using the distance matrix, where each element is the minimum cumulative distance from the starting point to the corresponding point. By backtracking from the end of the cumulative cost matrix, the optimal regularization path with the minimum cumulative cost is searched to achieve non-equidistant time alignment of the two sequences; Extract the matching position of the last sampling point of the sequence to be matched on the template sequence, and the corresponding time point is the reference time on the time series curve of the standard gait parameters.

[0088] Q4-3: Based on the reference time and the time series curve of the standard gait parameters, calculate the gait parameters at the next time step after the reference time.

[0089] Q5: Provide assistance to the wearer based on the waist torque during walking and the gait parameters at the next moment.

[0090] In one implementation, the duration between the next moment and the current moment is one-third of a single gait cycle.

[0091] Since there may be differences between the predicted gait parameters and the actual gait parameters, this embodiment uses a more recent future timeframe in another implementation to minimize prediction errors; for example, predicting gait parameters only for the next 0.05 seconds. This is because the gait frequency of human walking is between 0.5 Hz and 2 Hz, and the duration of a single gait cycle is between 0.5 seconds and 2 seconds. Within 0.05 seconds, the change in gait parameter values ​​is relatively small, resulting in a smaller error; thus achieving the goal of reducing prediction errors.

[0092] The above embodiments identify three movement types—bending, lifting, and walking—by changing the hip joint angle. They combine plantar pressure and joint angle to calculate load mass and upper body tilt angle in real time. Gain parameters are adaptively configured according to different movement types, and a lumbar torque calculation model including posture closed-loop and gravity compensation is constructed to achieve continuous and dynamic lumbar assistance output. For walking movements, a gait template library is further constructed by combining positive zero-crossing detection, clustering, and fifth-order Fourier fitting. A two-layer LSTM and DTW algorithm is used to complete gait pattern recognition and precise gait timing alignment, predicting gait parameters for the next moment, and achieving coordinated control of lumbar assistance and lower limb walking assistance.

[0093] Compared with existing constant assist solutions, the solution in the above embodiment can adaptively adjust the amount of assistance according to the type of action, load quality and upper body posture, avoiding insufficient or excessive assistance, significantly improving the smoothness of human-computer interaction and work safety; at the same time, by using gait prediction to achieve walking assistance to follow ahead, in high-intensity load scenarios such as logistics handling and industrial assembly, it can effectively reduce the load on the waist muscles and physical consumption of workers, and reduce the risk of lumbar muscle strain and joint damage.

[0094] refer to Figure 3 In this embodiment, a sensor system on the exoskeleton is used to collect human motion data, and then absolute encoders at the hip and knee joints are used to identify the type of movement. The types of movement include: standing, walking, bending, squatting, and lifting.

[0095] The standing state employs a multi-parameter coupled compensation zero-force dragging algorithm. The core of this algorithm is to construct a multi-rigid-body dynamic model of the exoskeleton linkage and a motor friction model to achieve precise decoupling of gravity and friction.

[0096] The core formula of the zero-force drag algorithm is:

[0097] in, τ 0 represents the joint compensation torque. τ G For gravitational torque, in exoskeleton robots it can be simplified to the torque relative to joint angles. θ trigonometric functions sin(θ) The function, i.e. τ G =Ksin(θ) ; K These are function coefficients, determined based on the distance from the center of gravity of the rotating part to the joint; τ f This is the frictional torque.

[0098] The frictional torque at the joint satisfies:

[0099] Coulomb friction K 1 and viscous friction K 2 components, This represents the joint angular velocity.

[0100] Real-time force feedback from the hip and knee joints is collected, and the joint compensation torque is dynamically adjusted to counteract friction and gravity at the joint, ensuring a dynamic balance between assist torque and resistance torque.

[0101] Walking status: Gait parameters are calculated based on a two-layer LSTM network, DTW algorithm, and gait template library; Bending and lifting states: The lumbar torque is calculated using an exoskeleton lifting assistance control method; The VMC method was used to calculate the knee and hip torques during the squatting position.

[0102] The final output control parameters provide assistance; the control parameters include: joint compensation torque, gait parameters, lumbar torque, knee joint torque and hip joint torque.

[0103] The above description is merely a preferred embodiment of the present invention and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of the invention is not limited to the specific combination of the above-described technical features, but also includes other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the inventive concept. For example, technical solutions formed by substituting the above-described features with (but not limited to) technical features with similar functions disclosed in this invention.

Claims

1. A method for controlling the lifting assistance of an exoskeleton, characterized in that, include: S1: Obtain the current movement type of the exoskeleton wearer, the mass of the object being carried, and the upper body tilt angle of the exoskeleton wearer; S2: Calculate the gain correction coefficient based on the action type, the mass of the object being moved, and the upper body tilt angle of the exoskeleton wearer; S3: Use the gain correction coefficient to correct the waist assist gain when standing upright, and obtain the corrected waist assist gain; S4: Calculate the lumbar torque based on the action type and the corrected lumbar assist gain; S5: Provide assistance to the wearer based on the waist torque.

2. The exoskeleton lifting assistance control method according to claim 1, characterized in that, The exoskeleton consists of a lower limb portion and a back plate portion, which are rotatably connected to the back plate portion via a hip joint; the angle of the hip joint is the angle between the back plate portion and the vertical direction. The types of movements include: bending over, lifting, and walking. Obtain the current movement type of the exoskeleton wearer, including: If the sum of the hip joint angle at the previous moment and the angle error threshold is less than the hip joint angle at the current moment, it is determined to be a bending motion. If the difference between the hip joint angle at the previous moment and the angle error threshold is greater than the hip joint angle at the current moment, it is determined to be a lifting motion. If the absolute value of the difference between the hip joint angle at the previous moment and the hip joint angle at the current moment is less than the angle error threshold, it is determined to be a walking motion.

3. The exoskeleton lifting assistance control method according to claim 2, characterized in that, The exoskeleton also includes a foot device connected to the lower limbs; the foot device is equipped with pressure sensors for detecting the total mass of the wearer and the object. The mass of the object being transported is the difference between the total mass and the mass of the wearer; the upper body tilt angle is equal to the angle of the hip joint.

4. The exoskeleton lifting assistance control method according to claim 1, characterized in that, The following relationships are satisfied between the action type, the mass of the object being moved, the upper body tilt angle of the exoskeleton wearer, and the gain correction coefficient: in, Y This is the gain correction factor. a 0、 a 1. a 2 represents the gain parameter. θ The angle of the upper body tilt. m For the mass of the object being transported, m 0 represents the ratio of the exoskeleton's rated load weight to gravitational acceleration; the gain parameter is set according to the action type; the gain parameter is set according to the action type; The waist support gain is the product of the waist support gain when standing upright and the gain correction coefficient.

5. The exoskeleton lifting assistance control method according to claim 4, characterized in that, The gain parameter is set according to the action type as follows: When the action type is a bending motion, the gain parameter satisfies: a 0、 a 1 and a The sum of 2 is less than 1 and greater than 0, so that the gain function can be used to characterize the functional relationship between the mass of the object being carried and the upper body tilt angle and the gain of the exoskeleton on the waist during the bending motion; When the action type is a lifting action, the gain parameter satisfies: a 0 is greater than 1 and a 1 and a 2 are all greater than 0, so that the gain function can be used to characterize the functional relationship between the mass of the object being carried and the upper body tilt angle and the gain of the exoskeleton on the waist during the lifting action; When the action type is walking, the gain parameter satisfies: a 0 equals 1 and a 1 and a 2 are all greater than 0, so that the gain function is used to characterize the functional relationship between the mass of the object being carried and the upper body tilt angle and the gain of the exoskeleton on the waist during walking.

6. The exoskeleton lifting assistance control method according to claim 1, characterized in that, The lumbar torque is calculated based on the described action type and the corrected lumbar assist gain, including: If the action type is a bending or walking action, the base torque is calculated based on the waist assist gain in the current state; the base torque is used as the waist torque.

7. The exoskeleton lifting assistance control method according to claim 1, characterized in that, The lumbar torque is calculated based on the described action type and the corrected lumbar assist gain, including: If the action type is a lifting action, the basic torque is calculated based on the waist assist gain in the current state; the load gravity compensation torque is calculated based on the mass of the object being moved, and the sum of the basic torque and the load gravity compensation torque is used as the waist torque.

8. A method for coordinated assistive control of exoskeleton transport, characterized in that, During walking, the exoskeleton sensing system detects the current movements of the exoskeleton in real time and obtains movement signals; the movement signals include multiple gait parameters that satisfy temporal relationships; the gait parameters include: gait cycle duration, average angular velocity and amplitude of hip joint angle, and average angular velocity and amplitude of knee joint angle; When the exoskeleton wearer's current movement type is identified as walking based on motion signals, it includes: Q1: Calculate the lumbar torque during walking motion using an exoskeleton lifting assistance control method according to any one of claims 1-7; Q2: Construct a gait template library; the gait template library contains time-series curves of various standard gait parameters corresponding to different walking speeds; the time-series curves of standard gait parameters are used to calculate gait parameters at corresponding walking speeds; Q3: During the walking process, multiple real-time gait parameters that satisfy the temporal relationship are extracted from the action signal using a sliding window method; Q4: Based on a two-layer LSTM network and the DTW algorithm, determine the gait parameters of the time series curves of the standard gait parameters corresponding to the multiple real-time gait parameters in the gait template library at the next time step; Q5: Provide assistance to the wearer based on the waist torque during walking and the gait parameters at the next moment.

9. The exoskeleton transport coordination and assistance control method according to claim 8, characterized in that, Build a gait template library, including: Q2-1: Collect multiple initial gait parameters from the motion signal; the multiple initial gait parameters satisfy a temporal relationship; Q2-2: Extract standard gait parameters based on the initial gait parameters; Q2-3: Using clustering, the multiple standard gait parameters are divided into various gait modes according to different paces; the gait modes include: slow mode, medium speed mode, and fast mode; Q2-4: Use a high-order Fourier series to fit the standard gait parameters belonging to the same step speed mode to obtain the time series curve of the standard gait parameters; Q2-5: Compile the time-series curves of standard gait parameters for each gait mode into a gait template library.

10. The exoskeleton transport coordination and assistance control method according to claim 8, characterized in that, Based on a two-layer LSTM network and the DTW algorithm, the gait parameters of the standard gait parameter time series curves corresponding to multiple real-time gait parameters in the gait template library are determined at the next time step, including: Q4-1: Input multiple real-time gait parameters into a two-layer LSTM network to obtain the current gait mode; the two-layer LSTM network is used to predict and output the corresponding gait mode based on the input gait parameters. Q4-2: Using the DTW algorithm, the reference time on the time series curve of the standard gait parameter corresponding to the current gait mode is matched for multiple real-time gait parameters; Q4-3: Based on the reference time and the time series curve of the standard gait parameters, calculate the gait parameters at the next time step after the reference time.

Images