Human body tumble protection method, device, equipment and medium

Abstract

The invention relates to the technical field of robots, and discloses a human body tumble protection method and device, an exoskeleton robot and a medium, and the method comprises the steps: obtaining multi-modal motion data, collected by a sensor in real time, of a user; inputting the multi-modal motion data into a tumble identification model for identification and classification to obtain a tumble type probability of the user; if the tumble probability is greater than a preset safety threshold, determining that the user is in a tumble trend, and generating a variable impedance control instruction matched with the target tumble type according to a preset strategy library; and determining a rigidity parameter and a damping parameter of a target joint motor according to the variable impedance control instruction, and determining a target output torque based on the joint motion data, so that the target joint motor performs fall protection on the user based on the target output torque. The fall type can be accurately recognized, flexible buffering can be achieved, and then the injury risk of a user is remarkably reduced.

Description

Technical Field

[0001] This invention relates to the field of robotics, and in particular to a method, device, equipment, and medium for protecting against human falls. Background Technology

[0002] In practical applications of exoskeleton robots, slippery surfaces, obstacles, or user imbalance can pose a risk of falls for users wearing lower limb exoskeletons, potentially leading to fractures, joint injuries, and other serious consequences, severely impacting the safety and practicality of the exoskeleton robot. Therefore, it is necessary to implement fall prevention measures based on exoskeleton robots.

[0003] Related technologies typically only trigger a response after the body has already tilted significantly or even hit the ground, missing the critical pre-collision time window from instability to impact, resulting in a delayed protective activation. Furthermore, they generally only determine whether a fall has occurred, not the specific type of fall. Moreover, because the impact points and injury mechanisms on the human body differ significantly depending on the direction of the fall, current single-mode protection strategies cannot provide differentiated buffering, thus limiting their protective effectiveness.

[0004] Therefore, there is an urgent need for a fall protection method that can predict and accurately identify fall types in advance and provide flexible cushioning, thereby effectively absorbing impact energy during a fall and significantly reducing the risk of injury to users, thus improving the safety and practicality of exoskeleton robots. Summary of the Invention

[0005] This invention provides a method, device, equipment, and medium for protecting the human body from falls, in order to solve the technical problem that the fall prevention strategies applied to exoskeleton robots in related technologies cannot achieve differentiated cushioning and have limited protective effects.

[0006] Firstly, a method for preventing human falls is provided, applied to exoskeleton robots, the method comprising: The system acquires multimodal motion data of the user collected in real time by sensors, wherein the multimodal motion data includes joint motion data, trunk posture data, and leg motion data. The multimodal motion data is input into a fall detection model for identification and classification to obtain the user's fall type probability; wherein, the fall detection model is constructed based on a deep learning model combining convolutional neural networks and long short-term memory networks, and the fall type probability distribution includes the fall probability and the corresponding target fall type; If the probability of falling is greater than a preset safety threshold, the user is determined to be in a falling trend, and a variable impedance control command matching the target fall type is generated according to a preset strategy library. The stiffness and damping parameters of the target joint motor are determined according to the variable impedance control command, and the target output torque is determined based on the joint motion data, so that the target joint motor can protect the user from falls based on the target output torque.

[0007] Secondly, a fall protection device is provided, comprising: The acquisition module is used to acquire multimodal motion data of the user collected in real time by sensors, wherein the multimodal motion data includes joint motion data, trunk posture data and leg motion data; The classification module is used to input the multimodal motion data into the fall recognition model for recognition and classification, and to obtain the fall type probability of the user; wherein, the fall recognition model is constructed based on a deep learning model combining convolutional neural network and long short-term memory network, and the fall type probability distribution includes fall probability and corresponding target fall type; The instruction generation module is used to determine that the user is in a fall trend if the fall probability is greater than a preset safety threshold, and to generate a variable impedance control instruction that matches the target fall type according to a preset strategy library. The control module is used to determine the stiffness and damping parameters of the target joint motor according to the variable impedance control command, and to determine the target output torque based on the joint motion data, so that the target joint motor can protect the user from falls based on the target output torque.

[0008] Thirdly, an exoskeleton robot is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps of the aforementioned method for preventing human falls.

[0009] Fourthly, a computer-readable storage medium is provided, which stores a computer program that, when executed by a processor, implements the steps of the aforementioned method for preventing human falls.

[0010] In the aforementioned methods, devices, exoskeleton robots, and storage media for human fall protection, the method, applied to an exoskeleton robot, includes: acquiring multimodal motion data of the user collected in real time by sensors, wherein the multimodal motion data includes joint motion data, trunk posture data, and leg motion data; inputting the multimodal motion data into a fall recognition model for identification and classification to obtain the user's fall type probability; wherein the fall recognition model is constructed based on a deep learning model combining convolutional neural networks and long short-term memory networks, and the fall type probability distribution includes the fall probability and the corresponding target fall type; if the fall probability is greater than a preset safety threshold, it is determined that the user is in a fall tendency, and a variable impedance control command matching the target fall type is generated according to a preset strategy library; the stiffness parameters and damping parameters of the target joint motor are determined according to the variable impedance control command, and the target output torque is determined based on the joint motion data, so that the target joint motor provides fall protection for the user based on the target output torque. This application, by collecting the user's multimodal motion data in real time, can accurately reflect the user's joint motion, trunk posture, and leg motion state, thereby monitoring the user's motion changes in real time. Based on a fall detection model, it can efficiently identify a user's fall trend and determine whether the user is at risk of falling by analyzing the probability distribution of fall types (including fall probability and its corresponding target fall type). Once the fall probability exceeds a preset safety threshold, it immediately reacts by automatically triggering a variable impedance control mechanism to precisely adjust the stiffness and damping parameters of the target joint motor, thereby effectively preventing or mitigating the impact of falls on the user. Attached Figure Description

[0011] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments of the present invention will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0012] Figure 1 This is a schematic diagram of an application environment for a method for preventing human falls according to an embodiment of the present invention; Figure 2 This is a schematic flowchart of a method for preventing human falls according to an embodiment of the present invention; Figure 3 yes Figure 2 A schematic diagram of a specific implementation method for step S20; Figure 4 This is a schematic diagram of a fall protection device according to an embodiment of the present invention; Figure 5 This is a schematic diagram of the structure of an exoskeleton robot according to one embodiment of the present invention; Figure 6 This is another structural schematic diagram of the exoskeleton robot in one embodiment of the present invention. Detailed Implementation

[0013] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0014] The fall protection method provided in this invention can be applied to, for example... Figure 1 In this application environment, the client communicates with the server via a network. The server can acquire multimodal motion data of the user collected in real time by sensors through the client. This multimodal motion data includes joint motion data, trunk posture data, and leg motion data. The multimodal motion data is input into a fall detection model for identification and classification to obtain the user's fall type probability. The fall detection model is constructed based on a deep learning model combining convolutional neural networks and long short-term memory networks. The fall type probability distribution includes the fall probability and the corresponding target fall type. If the fall probability is greater than a preset safety threshold, the user is determined to be in a fall tendency, and a variable impedance control command matching the target fall type is generated according to a preset strategy library. The stiffness and damping parameters of the target joint motor are determined based on the variable impedance control command, and the target output torque is determined based on the joint motion data, so that the target joint motor provides fall protection for the user based on the target output torque. In this invention, by collecting the user's multimodal motion data in real time, the user's joint motion, trunk posture, and leg motion status can be accurately reflected, thereby enabling real-time monitoring of the user's motion changes. Based on a fall detection model, it can efficiently identify a user's fall trend and determine whether the user is at risk of falling by analyzing the probability distribution of fall types (including fall probability and its corresponding target fall type). Once the fall probability exceeds a preset safety threshold, it immediately reacts by automatically triggering a variable impedance control mechanism to precisely adjust the stiffness and damping parameters of the target joint motor, thereby effectively preventing or mitigating the impact of falls on the user.

[0015] The client can be, but is not limited to, various personal computers, laptops, smartphones, tablets, and portable wearable devices. The server can be implemented using a standalone server or a server cluster consisting of multiple servers. The invention will now be described in detail through specific embodiments.

[0016] Please see Figure 2 As shown, Figure 2 A schematic flowchart of a method for preventing human falls provided in an embodiment of the present invention, the method being applied to an exoskeleton robot, includes the following steps: S10: Acquire multimodal motion data of the user collected in real time by sensors. The multimodal motion data includes joint motion data, trunk posture data, and leg motion data.

[0017] For example, sensors integrated throughout the device can be used to collect the user's real-time motion data in a comprehensive and multi-dimensional manner, resulting in multimodal motion data. These sensors include, but are not limited to, inertial measurement units (IMUs) and joint encoders located at the hip and knee joints of the exoskeleton, as well as pressure sensors placed on the soles of the feet.

[0018] For example, multimodal motion data includes joint motion data, trunk posture data, and leg motion data. Joint motion data is acquired in real-time by joint encoders installed at the hip and knee joints, capturing the absolute angles, rotational angular velocities, and instantaneous accelerations of each joint to reflect the motion phase of the user's lower limbs. Posture data is acquired by an inertial measurement unit (IMU) installed at the center of the exoskeleton's backplate, collecting the trunk's three-axis accelerations, angular velocities, and Euler angles (pitch, roll, and yaw) in three-dimensional space to determine whether the body's overall center of gravity deviates from safety boundaries. Leg motion data is obtained in real-time by auxiliary inertial measurement units distributed symmetrically at the thigh and calf bars, monitoring the swing posture of both legs and their spatial geometric relationship with the ground.

[0019] By collecting multimodal data through the above embodiments, the exoskeleton robot no longer relies on a single physical threshold (such as simple acceleration exceeding the limit) for judgment. Instead, by fusing the motion information of the joints, torso, and legs, the exoskeleton robot greatly enhances its perception depth of the user's complex movement patterns, laying a physical data foundation for fall prediction.

[0020] S20: Input multimodal motion data into the fall recognition model for identification and classification to obtain the probability of the user's fall type.

[0021] The fall recognition model is constructed based on a deep learning model that combines convolutional neural networks and long short-term memory networks. The fall type probability distribution includes the fall probability and the corresponding target fall type.

[0022] For example, fall detection models can utilize their built-in convolutional neural network layers to extract spatial correlation features from multimodal motion data, capturing spatial dimension features by imprinting the coupling relationships between different sensor nodes (such as the coordination between trunk tilt and limb movements). Simultaneously, fall detection models can utilize long short-term memory network layers to extract dynamic trend features from multimodal motion data, leveraging their temporal memory capabilities to capture the temporal evolution from human instability to the final fall, thus providing crucial temporal feature support for fall detection.

[0023] Furthermore, the classification units of the fall detection model can be calculated based on the aforementioned fused spatial correlation features and dynamic trend features, ultimately outputting a fall type probability distribution. This probability distribution clearly defines the corresponding target fall type and its probability of occurrence. The target fall types include, but are not limited to, normal walking, standing, sitting, and fine-grained classifications such as forward falls, backward falls, and side falls. By monitoring the output fall probabilities and combining them with preset threshold logic, specific fall actions can be accurately identified or predicted, providing a basis for subsequent differentiated adaptive flexible protection control strategies.

[0024] In some embodiments, multimodal motion data is input into a fall detection model for identification and classification to obtain the user's fall type probability, including: extracting spatial correlation features from the multimodal motion data through the convolutional neural network of the fall detection model; and extracting dynamic trend features from the multimodal motion data through the long short-term memory network of the fall detection model; and outputting the fall type probability through the classification unit of the fall detection model based on the spatial correlation features and the dynamic trend features.

[0025] For example, fall detection models can utilize their constructed convolutional neural network (CNN) layers to perform convolution operations on the input multimodal motion data to extract spatial correlations between different data points. For instance, CNNs can capture the coupling relationships between motion data from the back, thighs, and calves, identifying spatial correlation features such as the relationship between trunk tilt angle and lower limb elevation. Furthermore, fall detection models can also utilize their constructed long short-term memory (LSTM) network layers to perform time-series analysis on multimodal motion data. Since falling is a process with temporal evolution characteristics, extracting dynamic trend features of movements through LSTM networks can capture the evolutionary patterns from loss of balance to the final fall, providing temporal dimension feature support for fall prediction.

[0026] Furthermore, after acquiring spatial correlation features and dynamic trend features, these features can be fused and input into the classification unit (output layer) of the fall detection model. The classification unit can calculate based on the fused features and finally output the fall type probability of the current user's movement state. The fall type probability covers multiple preset fall categories, including but not limited to normal walking, standing, sitting, and forward, backward, and lateral falls. Logical judgment can be achieved by monitoring the output fall probability. For example, when the fall probability of a certain fall type exceeds a preset threshold (e.g., the threshold is set to 0.85, 0.9, etc., without limitation here), it can be determined that a fall of that type is about to occur, and an corresponding adaptive flexible protection control strategy can be triggered based on the identified specific fall type (e.g., forward, backward, or lateral).

[0027] Among them, such as Figure 3 As shown, step S20, which involves inputting multimodal motion data into the fall recognition model for identification and classification to obtain the user's fall type probability, also includes the following steps: S21: Perform noise reduction filtering on the multimodal sensor data to obtain the processed data.

[0028] S22: Use a sliding time window to extract the processed data to obtain a time series data sequence.

[0029] S23: Input the time series data sequence into the fall recognition model for recognition and classification to obtain the probability of the user's fall type.

[0030] For example, the acquired multimodal sensor data can first undergo denoising filtering. To address signal interference in dynamic environments, algorithms such as Kalman filtering can be used to eliminate random noise and baseline drift, thereby extracting physical variables that accurately characterize human motion. This provides structured data support for subsequent deep learning models, ensuring the robustness of the input features.

[0031] Furthermore, to capture the highly transient temporal evolution of a fall, a sliding time window technique can be used to refine the processed data. For example, by presetting a specific window length (e.g., 500ms) and step frequency (e.g., 50ms), the continuous motion state stream can be reconstructed into a temporally coherent time-series data sequence. This serialization encapsulation method allows the model input to simultaneously cover the dynamic evolution trajectory of human balance from instability to fall, providing the necessary temporal feature foundation for fall prediction.

[0032] Furthermore, the generated time-series data can be synchronously input into the built-in fall detection model. The fall detection model is based on a hybrid architecture combining convolutional neural networks (CNNs) and long short-term memory networks (LSTMs): CNNs extract spatial correlation features between multimodal motion data, while LSTMs extract dynamic trend features of the action sequences. Thus, based on deep fusion and feature mapping of spatiotemporal features, the fall detection model can ultimately output the probability of the fall type in the user's current motion state.

[0033] S30: If the probability of falling is greater than the preset safety threshold, the user is determined to be in a falling trend, and a variable impedance control command matching the target fall type is generated according to the preset strategy library.

[0034] For example, the probability of a target fall type output by the fall detection model can be compared in real time with a preset safety threshold (such as 0.85 or 0.90, which can be set according to needs). If the fall probability value of a certain target fall type exceeds the preset safety threshold, it is determined that the user is currently in a fall trend, and according to the built-in preset strategy library, a variable impedance control command matching the target fall type is retrieved and generated.

[0035] It should be noted that the preset strategy library contains pre-stored impedance parameter mapping tables optimized for different fall types, including dynamic adjustment schemes for the stiffness and damping coefficients of each moving joint. Through this process, a rapid mapping from the identified fall type probability to the underlying hardware execution instructions is achieved, ensuring that the protective actions have extremely high directional specificity.

[0036] Furthermore, the generated variable impedance control commands are designed to simulate the yielding and cushioning behavior of biological muscles under impact loads by adjusting the motor's output characteristics. For example, the command generated for a forward fall will instruct the knee joint to enter a high-damping state to control the orderly flexion of the lower limbs and absorb the impact force upon landing; while the command for a backward fall will limit the backward tilting speed of the torso by increasing the stiffness of the hip joint. Through this precise command matching, flexible support with high biocompatibility and energy dissipation efficiency can be provided to users under complex fall types, thereby minimizing the risk of secondary injury.

[0037] In some embodiments, the target fall type includes forward fall, backward fall, and lateral fall; the variable impedance control command includes a first control command, a second control command, and a third control command; the target joint motors include a knee joint motor, a hip joint motor, and a lower limb joint motor on the fall side; and the variable impedance control command matching the target fall type is generated according to a preset strategy library, including: If the target fall type is a forward fall, a first control command is generated, which instructs the damping parameter of the knee joint motor to be increased to a first preset threshold and the stiffness parameter of the hip joint motor to be increased to a second preset threshold. If the target fall type is a backward fall, a second control command is generated, which instructs the stiffness parameter of the hip joint motor to be increased to a third preset threshold and the damping parameter of the knee joint motor to be increased to a fourth preset threshold. If the target fall type is a lateral fall, a third control command is generated, which instructs the damping parameter of the lower limb joint motor on the fall side to be increased to a fifth preset threshold. The third preset threshold is greater than the second preset threshold, and the fourth preset threshold is greater than or equal to the first preset threshold. The fifth preset threshold is greater than the first preset threshold and / or the fourth preset threshold.

[0038] For example, if the identification result indicates that the target fall type is a forward fall, a first control command is generated to instruct the damping parameter of the knee joint motor to increase to a first preset threshold, thereby allowing the knee joint to flex in a controlled state to absorb the kinetic energy of the ground impact; at the same time, it instructs to increase the stiffness parameter of the hip joint motor to a second preset threshold, maintaining the basic mechanical shape of the torso and preventing excessive phase folding, thereby achieving a protective effect that simulates knee flexion cushioning; if the target fall type is determined to be a backward fall, a second control command is generated to deal with the extremely high risk of the tailbone and back of the head landing. The second control command instructs to further increase the stiffness parameter of the hip joint motor to a third preset threshold (where the third preset threshold is greater than the second preset threshold), using high-gain stiffness to forcibly limit the backward leaning speed of the torso; at the same time, it instructs the damping parameter of the knee joint motor to increase to a fourth preset threshold (where the fourth preset threshold is greater than or equal to the first preset threshold), maximizing the consumption of gravitational potential energy through motor anti-torque, supporting the formation of a triangular protective geometry; if the target fall type is determined to be a lateral fall, a third control command is generated to maintain the stability of the lower limb support. This instruction significantly increases the damping parameter of the lower limb joint motor on the fall side to a fifth preset threshold (where the fifth preset threshold is greater than the first and fourth preset thresholds). By applying an extremely high level of damping gain to the fall side, it is possible to effectively delay the lateral momentum and maintain the geometry of the lower limb support as much as possible, preventing structural damage caused by leg crossing or instantaneous loss of control.

[0039] It should be noted that the damping parameters are set to dissipate the gravitational potential energy of the body during the fall through the counter-torque of the motor. In a forward fall, the knee joint needs to maintain controlled flexion under resistance (first preset threshold) to prolong the impact time window. In a backward fall, to prevent the center of gravity from shifting too quickly backward and causing damage to the back of the head, a fourth preset threshold is set to be greater than or equal to the first preset threshold. Through a higher level of damping gain, the absorption of backward kinetic energy is maximized, ensuring the dynamic stability of the exoskeleton support structure. For lateral falls, since the anatomical stability of the human body is weakest in the lateral plane, a fifth preset threshold is set to be significantly greater than the first and fourth preset thresholds. This aims to instantly generate a large amount of motion resistance through the motor, forcibly maintaining the geometry of the lower limb on the falling side and preventing structural sprains caused by lateral joint instability.

[0040] Furthermore, the hip joint stiffness parameter determines the exoskeleton's ability to maintain trunk position. In a forward fall, the second preset threshold maintains moderate stiffness, allowing the body to fold moderately to cooperate with the knee joint for cushioning; while setting the third preset threshold (backward fall) to be greater than the second preset threshold means that in a backward fall, a stronger rigid support torque must be provided to counteract the trunk's tendency to lean backward, thereby preventing accelerated impact after loss of center of gravity.

[0041] The above embodiments establish a rigorous parameter hierarchy system through a preset strategy library: by setting the third preset threshold to be greater than the second preset threshold, and the fifth preset threshold to be greater than the first and / or fourth preset threshold, it is ensured that the exoskeleton robot can output the optimal biomechanical torque expectation value under different fall types, which greatly improves the safety and environmental robustness of the exoskeleton system under complex fall conditions.

[0042] Based on the above embodiments, the target joint motor also includes a lower limb joint motor on the opposite side of the fall. If the target fall type is a lateral fall, the third control command is also used to instruct the stiffness parameter of the lower limb joint motor on the opposite side of the fall to be increased to a sixth preset threshold or to instruct the lower limb joint motor on the opposite side of the fall to be locked; wherein, the sixth preset threshold is greater than or equal to the third preset threshold.

[0043] For example, if the identification result indicates that the target fall type is a lateral fall, the third control command will simultaneously instruct the stiffness parameter of the lower limb joint motor on the opposite side of the fall to be increased to the sixth preset threshold, or directly instruct the control of that side joint motor to enter a locked state. This instantaneous stiffening process aims to form a stable rigid support for the non-falling limb, preventing momentum disturbances caused by disordered swinging of the lower limbs or entanglement of the legs during a lateral fall.

[0044] It should be understood that, since the lateral support stability of the human body is much lower than that of the longitudinal plane when falling sideways, the sixth preset threshold is set to be greater than or equal to the third preset threshold. This forces the exoskeleton robot to output extreme rigid torque to forcefully maintain the geometric configuration of the contralateral lower limb, ensuring that it can act as a support fulcrum to help mitigate lateral impact.

[0045] This embodiment achieves a synergistic effect of cushioning on one side and supporting on the other by differentially controlling the lower limb on the side of the fall. This allows for maximum control over the lateral shift speed of the center of gravity trajectory, avoiding overload on the pelvis or femoral neck due to joint collapse, thus providing highly biocompatible and structurally robust safety protection in the extremely dangerous scenario of a lateral fall.

[0046] S40: Determine the stiffness and damping parameters of the target joint motor according to the variable impedance control command, and determine the target output torque based on the joint motion data, so that the target joint motor can protect the user from falls based on the target output torque.

[0047] For example, the mechanical parameter characteristics of the corresponding target joint motor can be parsed in real time based on the received variable impedance control commands (such as the first, second, or third control commands). Specifically, this involves mapping abstract control commands to the real-time stiffness and damping parameters of the target joint (such as the hip or knee joint) in the current fall scenario, and determining the target output torque based on joint motion data. This process realizes a quantitative conversion from identification category to physical stiffness and flexibility, ensuring that the mechanical impedance characteristics of the exoskeleton can match the user's fall momentum.

[0048] For example, after obtaining the stiffness and damping parameters, the target output torque can be deconstructed by combining joint motion data (such as angles and angular velocities) in various ways: for example, by using a variable impedance dynamics model, the spring-damping physical yield characteristics can be constructed through analytical calculations to calculate the torque benchmark that meets the basic flexible protection requirements; or, data-driven algorithms such as teaching learning or reinforcement learning can be introduced to output a globally optimized nonlinear compensation torque by retrieving biomechanical expert databases or policy networks; in addition, physiological torque estimation can be performed by combining surface electromyography signals extracted by physiological sensing units, or the center of gravity compensation torque can be calculated based on the offset rate of the plantar pressure center. Finally, by decoupling and weighting the multi-source torques, a target output torque that conforms to the constraints of physical laws and has human physiological adaptability can be generated to achieve efficient dissipation of impact energy and precise maintenance of human steady state.

[0049] It should be noted that the aforementioned target joint motors refer to the power execution units dynamically selected based on the identified target fall type. Specifically, these include hip and knee joint motors distributed on both sides of the exoskeleton, as well as the fall-side lower limb joint motor and the fall-opposite lower limb joint motor specifically defined in lateral fall scenarios. During protective execution, by differentially scheduling the impedance parameters of these motors, the knee joint motors achieve knee flexion and buffering to dissipate kinetic energy during forward falls through high damping, while the hip joint motors limit the backward leaning speed of the torso to maintain postural stability during backward falls through high stiffness. In lateral falls, they work together to drive the fall-side motor to implement high damping support and the opposite-side motor to implement high stiffness locking (or locking), thereby constructing a multi-joint flexible protection matrix with direction specificity to ensure precise protection of the user's key biomechanical nodes under various fall types.

[0050] In some embodiments, the joint motion data includes joint angle and joint angular velocity. The stiffness parameters and damping parameters of the target joint motor are determined according to the variable impedance control command, and the target output torque is determined based on the joint motion data, including: determining the stiffness parameters and damping parameters of the target joint motor according to the variable impedance control command. The target output torque is obtained using the following formula: τ = Kp*(θ d -θ)+Kd*(ω d -ω); Where θ is the joint angle, ω is the joint angular velocity, and θ d To preset the desired joint angle, ω d τ is the preset desired angular velocity of the joint; Kp is the target output torque; Kd is the stiffness parameter; and Kd is the damping parameter.

[0051] For example, the desired angle and angular velocity of the preset joint can be used as a reference benchmark. The stiffness term can be used to achieve elastic correction of the human body posture deviation, and the damping term can be used to simulate the viscous yielding characteristics of biological muscles to absorb impact kinetic energy. In this way, the abstract control strategy is transformed into a dynamic torque output with physical impedance characteristics, so that the target joint motor can implement flexible fall protection with high biocompatibility for the user based on real-time motion feedback.

[0052] In some embodiments, the method further includes: continuously acquiring the fall type probability obtained by analyzing multimodal motion data using a fall recognition model; and determining that the target output torque of the target joint motor is zero in response to a change in the fall type probability from being greater than a preset safety threshold to being less than or equal to the preset safety threshold.

[0053] For example, to ensure system safety and smooth human-computer interaction after protective actions, the current risk level can be dynamically assessed by continuously acquiring fall type probabilities from real-time multimodal motion data obtained from the fall recognition model. Once the fall type probability decreases from above a preset safety threshold to below or equal to the preset safety threshold (e.g., determining that the human body is in a static landing state or the danger has passed), the target output torque of the target joint motor is determined to be zero. By promptly cutting off the variable impedance compensation torque, the exoskeleton quickly enters a zero-torque or resistance-free following state, thereby eliminating the risk of mechanical locking or malfunction of the motor after the protective action ends. This not only facilitates users to adjust their posture autonomously or receive external rescue after a fall, but also achieves a seamless and safe switch from emergency flexible protection to normal motion baseline.

[0054] As can be seen, the above scheme, by collecting users' multimodal motion data in real time, can accurately reflect the user's joint movements, trunk posture, and leg movement status, thereby monitoring the user's motion changes in real time. Based on the fall recognition model, it can efficiently identify the user's fall trend and determine whether the user is at risk of falling by analyzing the probability distribution of fall types (including fall probability and its corresponding target fall type). Once the fall probability exceeds a preset safety threshold, it immediately reacts by automatically triggering a variable impedance control mechanism to precisely adjust the stiffness and damping parameters of the target joint motor, thereby effectively preventing or mitigating the impact of falls on the user.

[0055] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0056] In one embodiment, a fall protection device is provided, which corresponds one-to-one with the fall protection methods described in the above embodiments. For example... Figure 4 As shown, the fall protection device includes an acquisition module 101, a classification module 102, an instruction generation module 103, and a control module 104. Detailed descriptions of each functional module are as follows: The acquisition module 101 is used to acquire multimodal motion data of the user collected in real time by sensors, wherein the multimodal motion data includes joint motion data, trunk posture data and leg motion data; The classification module 102 is used to input the multimodal motion data into the fall recognition model for recognition and classification to obtain the fall type probability of the user; wherein, the fall recognition model is constructed based on a deep learning model combining convolutional neural network and long short-term memory network, and the fall type probability distribution includes fall probability and corresponding target fall type; The instruction generation module 103 is used to determine that the user is in a fall trend if the fall probability is greater than a preset safety threshold, and to generate a variable impedance control instruction that matches the target fall type according to a preset strategy library. The control module 104 is used to determine the stiffness parameters and damping parameters of the target joint motor according to the variable impedance control command, and to determine the target output torque based on the joint motion data, so that the target joint motor can protect the user from falls based on the target output torque.

[0057] The acquisition module 101 is used to extract spatial correlation features from the multimodal motion data through the convolutional neural network of the fall recognition model; and to extract dynamic trend features from the multimodal motion data through the long short-term memory network of the fall recognition model; and to output the fall type probability through the classification unit of the fall recognition model based on the spatial correlation features and the dynamic trend features.

[0058] The classification module 102 is used to perform noise reduction filtering on the multimodal sensor data to obtain processed data; to use a sliding time window to truncate the processed data to obtain a time-series data sequence; and to input the time-series data sequence into the fall recognition model for recognition and classification to obtain the probability of the user's fall type.

[0059] The instruction generation module 103 is configured to: generate a first control instruction if the target fall type is a forward fall, wherein the first control instruction instructs the damping parameter of the knee joint motor to be increased to a first preset threshold and the stiffness parameter of the hip joint motor to be increased to a second preset threshold; generate a second control instruction if the target fall type is a backward fall, wherein the second control instruction instructs the stiffness parameter of the hip joint motor to be increased to a third preset threshold and the damping parameter of the knee joint motor to be increased to a fourth preset threshold; generate a third control instruction if the target fall type is a lateral fall, wherein the third control instruction instructs the damping parameter of the lower limb joint motor on the fall side to be increased to a fifth preset threshold; wherein the third preset threshold is greater than the second preset threshold, and the fourth preset threshold is greater than or equal to the first preset threshold; and the fifth preset threshold is greater than the first preset threshold and / or the fourth preset threshold.

[0060] Control module 104 is used to determine the stiffness and damping parameters of the target joint motor according to the variable impedance control command; and to obtain the target output torque using the following formula: τ = Kp*(θ d -θ)+Kd*(ω d -ω); Where θ is the joint angle, ω is the joint angular velocity, and θ d To preset the desired joint angle, ω d τ is the preset desired angular velocity of the joint; Kp is the target output torque; Kd is the stiffness parameter; and Kd is the damping parameter.

[0061] Control module 104 is used to continuously acquire the fall type probability obtained by the fall recognition model from the analysis of the multimodal motion data; in response to the fall type probability changing from greater than the preset safety threshold to less than or equal to the preset safety threshold, determine that the target output torque of the target joint motor is zero.

[0062] This invention provides a fall protection device that accurately reflects the user's joint movements, trunk posture, and leg movements by collecting the user's multimodal motion data in real time, thereby monitoring changes in the user's movement. Based on a fall recognition model, it can efficiently identify the user's fall trend and determine whether the user is at risk of falling by analyzing the probability distribution of fall types (including fall probability and its corresponding target fall type). Once the fall probability exceeds a preset safety threshold, it immediately reacts by automatically triggering a variable impedance control mechanism to precisely adjust the stiffness and damping parameters of the target joint motor, thereby effectively preventing or mitigating the impact of falls on the user.

[0063] Specific limitations regarding fall protection devices can be found in the description of fall protection methods above, and will not be repeated here. Each module in the aforementioned fall protection device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in the processor of the exoskeleton robot in hardware form or independent of it, or stored in the exoskeleton robot's memory in software form, so that the processor can call and execute the corresponding operations of each module.

[0064] In one embodiment, an exoskeleton robot is provided, which can be a server-side component, and its internal structure diagram can be as follows: Figure 5 As shown, the exoskeleton robot includes a processor, memory, network interface, and database connected via a system bus. The processor provides computational and control capabilities. The memory includes non-volatile and / or volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The network interface allows communication with external clients via a network connection. When executed by the processor, the computer program implements the functions or steps of a fall protection method on the server side.

[0065] In one embodiment, an exoskeleton robot is provided, which can be a client, and its internal structure diagram can be as follows: Figure 6 As shown, the exoskeleton robot includes a processor, memory, network interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The network interface is used to communicate with an external server via a network connection. When executed by the processor, the computer program implements the client-side functions or steps of a fall protection method.

[0066] In one embodiment, an exoskeleton robot is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it performs the following steps: The system acquires multimodal motion data of the user collected in real time by sensors, wherein the multimodal motion data includes joint motion data, trunk posture data, and leg motion data. The multimodal motion data is input into a fall detection model for identification and classification to obtain the user's fall type probability; wherein, the fall detection model is constructed based on a deep learning model combining convolutional neural networks and long short-term memory networks, and the fall type probability distribution includes the fall probability and the corresponding target fall type; If the probability of falling is greater than a preset safety threshold, the user is determined to be in a falling trend, and a variable impedance control command matching the target fall type is generated according to a preset strategy library. The stiffness and damping parameters of the target joint motor are determined according to the variable impedance control command, and the target output torque is determined based on the joint motion data, so that the target joint motor can protect the user from falls based on the target output torque.

[0067] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, the computer program performing the following steps when executed by a processor: The system acquires multimodal motion data of the user collected in real time by sensors, wherein the multimodal motion data includes joint motion data, trunk posture data, and leg motion data. The multimodal motion data is input into a fall detection model for identification and classification to obtain the user's fall type probability; wherein, the fall detection model is constructed based on a deep learning model combining convolutional neural networks and long short-term memory networks, and the fall type probability distribution includes the fall probability and the corresponding target fall type; If the probability of falling is greater than a preset safety threshold, the user is determined to be in a falling trend, and a variable impedance control command matching the target fall type is generated according to a preset strategy library. The stiffness and damping parameters of the target joint motor are determined according to the variable impedance control command, and the target output torque is determined based on the joint motion data, so that the target joint motor can protect the user from falls based on the target output torque.

[0068] It should be noted that the functions or steps that the computer-readable storage medium or exoskeleton robot can achieve are described in the relevant descriptions of the server side and client side in the aforementioned method embodiments. To avoid repetition, they will not be described one by one here.

[0069] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0070] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is used as an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.

[0071] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A method for preventing human falls, characterized in that, Applied to exoskeleton robots, the method includes: The system acquires multimodal motion data of the user collected in real time by sensors, wherein the multimodal motion data includes joint motion data, trunk posture data, and leg motion data. The multimodal motion data is input into a fall detection model for identification and classification to obtain the user's fall type probability; wherein, the fall detection model is constructed based on a deep learning model combining convolutional neural networks and long short-term memory networks, and the fall type probability distribution includes the fall probability and the corresponding target fall type; If the probability of falling is greater than a preset safety threshold, the user is determined to be in a falling trend, and a variable impedance control command matching the target fall type is generated according to a preset strategy library. The stiffness and damping parameters of the target joint motor are determined according to the variable impedance control command, and the target output torque is determined based on the joint motion data, so that the target joint motor can protect the user from falls based on the target output torque.

2. The method according to claim 1, characterized in that, The step of inputting the multimodal motion data into a fall detection model for identification and classification to obtain the probability of the user's fall type includes: Spatial correlation features are extracted from the multimodal motion data using the convolutional neural network of the fall recognition model; and, Dynamic trend features are extracted from the multimodal motion data through the long short-term memory network of the fall recognition model; Based on the spatial correlation features and the dynamic trend features, the fall type probability is output by the classification unit of the fall recognition model.

3. The method according to claim 1, characterized in that, The step of inputting the multimodal motion data into a fall detection model for identification and classification to obtain the probability of the user's fall type further includes: The multimodal sensor data is subjected to noise reduction filtering to obtain the processed data; A sliding time window is used to extract the processed data to obtain a time-series data sequence; The time-series data sequence is input into the fall recognition model for identification and classification to obtain the probability of the user's fall type.

4. The method according to claim 1, characterized in that, The target fall types include forward falls, backward falls, and lateral falls. The variable impedance control commands include a first control command, a second control command, and a third control command. The target joint motors include a knee joint motor, a hip joint motor, and a lower limb joint motor on the fall side. Generating variable impedance control commands matching the target fall types according to a preset strategy library includes: If the target fall type is a forward fall, the first control command is generated. The first control command is used to instruct the damping parameter of the knee joint motor to be increased to a first preset threshold and the stiffness parameter of the hip joint motor to be increased to a second preset threshold. If the target fall type is a backward fall, then the second control command is generated. The second control command is used to instruct the stiffness parameter of the hip joint motor to be increased to a third preset threshold, and to instruct the damping parameter of the knee joint motor to be increased to a fourth preset threshold. If the target fall type is the lateral fall, then the third control command is generated, which is used to instruct the damping parameter of the lower limb joint motor on the fall side to be increased to the fifth preset threshold. Wherein, the third preset threshold is greater than the second preset threshold, and the fourth preset threshold is greater than or equal to the first preset threshold; the fifth preset threshold is greater than the first preset threshold and / or the fourth preset threshold.

5. The method according to claim 4, characterized in that, The target joint motor also includes a lower limb joint motor on the opposite side of the fall. If the target fall type is the lateral fall, the third control command is also used to instruct the stiffness parameter of the lower limb joint motor on the opposite side of the fall to be increased to a sixth preset threshold or to instruct the lower limb joint motor on the opposite side of the fall to be locked. Wherein, the sixth preset threshold is greater than or equal to the third preset threshold.

6. The method according to claim 1, characterized in that, The joint motion data includes joint angles and joint angular velocities. The process of determining the stiffness and damping parameters of the target joint motor based on the variable impedance control command, and determining the target output torque based on the joint motion data, includes: Based on the variable impedance control command, determine the stiffness parameters and damping parameters of the target joint motor; The target output torque is obtained using the following formula: τ = Kp*(θ d -θ)+Kd*(ω d -ω); Where θ is the joint angle, ω is the joint angular velocity, and θ d To preset the desired joint angle, ω d τ is the preset desired angular velocity of the joint; Kp is the target output torque; Kd is the stiffness parameter; and Kd is the damping parameter.

7. The method according to claim 1, characterized in that, The method further includes: The probability of the fall type is continuously obtained by analyzing the multimodal motion data using the fall recognition model. In response to the change in the probability of the fall type from being greater than the preset safety threshold to being less than or equal to the preset safety threshold, the target output torque of the target joint motor is determined to be zero.

8. A fall protection device, characterized in that, include: The acquisition module is used to acquire multimodal motion data of the user collected in real time by sensors, wherein the multimodal motion data includes joint motion data, trunk posture data and leg motion data; The classification module is used to input the multimodal motion data into the fall recognition model for recognition and classification, and to obtain the fall type probability of the user; wherein, the fall recognition model is constructed based on a deep learning model combining convolutional neural network and long short-term memory network, and the fall type probability distribution includes fall probability and corresponding target fall type; The instruction generation module is used to determine that the user is in a fall trend if the fall probability is greater than a preset safety threshold, and to generate a variable impedance control instruction that matches the target fall type according to a preset strategy library. The control module is used to determine the stiffness and damping parameters of the target joint motor according to the variable impedance control command, and to determine the target output torque based on the joint motion data, so that the target joint motor can protect the user from falls based on the target output torque.

9. An exoskeleton robot, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method for preventing human falls as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of the method for preventing human falls as described in any one of claims 1 to 7.

Images