An Artificial Intelligence-Based Power Assist Control Method for Exoskeleton Robots

Abstract

The present invention discloses a power-assisted control method for an exoskeleton robot based on artificial intelligence, which relates to the field of intelligent control technology. The method includes: determining the power device of the exoskeleton robot; obtaining multiple indicators for assisting the power output of the power device; obtaining a first motion scenario, and performing parameter learning on the order of the multiple indicators according to the correlation between the first motion scenario and the multiple indicators to obtain auxiliary learning parameters; the power device controls the power motor to adjust the power parameters according to the auxiliary learning parameters. It solves the technical problems in the prior art that the power-assisted control of the exoskeleton robot is not intelligent and accurate enough, and it is difficult to dynamically adjust the power output according to the specific motion scenario, resulting in low energy efficiency and poor user experience. By intelligently identifying the limb motion scenario and combining multiple power output indicators for parameter learning, the power parameters of the power device are accurately adjusted, thereby achieving the technical effects of improving energy efficiency and enhancing user experience.

Description

Technical Field

[0001] The present invention relates to the field of intelligent control technology, and particularly to a power-assisted control method for an exoskeleton robot based on artificial intelligence. Background Art

[0002] By simulating the human bone structure, an exoskeleton robot provides additional power support or motion assistance functions for users, greatly improving the human's motor ability and endurance. However, the power-assisted control of exoskeleton robots has always been one of the key factors restricting the improvement of their performance. Traditional exoskeleton robots have problems such as insufficient intelligence, slow dynamic response, and high energy consumption in power-assisted control. These defects limit their application effects in complex motion scenarios. Traditional methods often can only provide a fixed power output mode, and it is difficult to effectively adjust according to the user's real-time motion state, thus affecting the overall energy efficiency and user experience. Summary of the Invention

[0003] This application provides a power-assisted control method for an exoskeleton robot based on artificial intelligence, which solves the technical problems in the prior art that the power-assisted control of exoskeleton robots is not intelligent and accurate enough, and it is difficult to dynamically adjust the power output according to specific motion scenarios, resulting in low energy efficiency and poor user experience.

[0004] This application provides a power-assisted control method for an exoskeleton robot based on artificial intelligence, and the method includes:

[0005] Determine the power device of the exoskeleton robot, where the power device includes a power input end and a power output end; obtain multiple indicators for assisting the power output of the power device, and the multiple indicators include a power output stability indicator, a power output continuity indicator, a power output step indicator, and a power output energy-saving indicator; receive the sensing data set of the sensing module, identify the limb motion scenario according to the sensing data set to obtain the first motion scenario, perform parameter learning on the order of the multiple indicators according to the correlation between the first motion scenario and the multiple indicators to obtain auxiliary learning parameters; the power device controls the power motor to adjust the power parameters according to the auxiliary learning parameters.

[0006] Further, after determining the power device of the exoskeleton robot, judge whether the power device is a single power output; if the power device is a single power output, input the auxiliary learning parameters into the power input end of the power device, identify the initial power parameters according to the auxiliary learning parameters, and output updated power parameters from the power output end; the power device controls the power motor to adjust the power parameters with the updated power parameters.

[0007] Further, if the power device does not have a single power output, identify multiple power output lines of the power device; input the auxiliary learning parameters into multiple power input ends of the power device, identify multiple initial power parameters according to the auxiliary learning parameters, and output multiple updated power parameters from the power output end; the power device controls the power motor to adjust the power parameters with the multiple updated power parameters.

[0008] Further, establish a collaborative line group of the multiple power output lines; group the multiple updated power parameters according to the collaborative line group, output an updated power parameter group, and perform collaborative update on the updated power parameters within the updated power parameter group to output multiple collaboratively updated power parameters; the power device controls the power motor to adjust the power parameters with the multiple collaboratively updated power parameters.

[0009] Further, the sensing module includes an environmental sensing module, and the environmental sensing module includes a CMOS sensor, a radar sensor, a GPS, and an RTK sensor, which are used for collecting sensing data of the environment where the wearable user of the exoskeleton robot is located; obtain the environmental sensing data set collected by the environmental sensing module, and add the environmental sensing data set to the sensing data set.

[0010] Further, the sensing module further includes a limb sensing module, and the limb sensing module includes a first-order inertial sensor and a second-order inertial sensor; the first-order inertial sensor is used for collecting upper limb inertial motion sensing data of the wearable user of the exoskeleton robot, and the second-order inertial sensor is used for collecting lower limb inertial motion sensing data of the wearable user of the exoskeleton robot; obtain the limb sensing data set collected by the limb sensing module, store the limb sensing data set in the sensing data set, and synchronize and align the environmental sensing data set and the limb sensing data set in time series using the dynamic time warping (DTW) algorithm for storage.

[0011] Further, determine whether the exoskeleton robot is worn on the leg. If the exoskeleton robot is worn on the leg, set a foot pressure sensor, collect foot pressure sensing data of the wearable user of the exoskeleton robot according to the foot pressure sensor, and obtain a foot pressure sensing data set; add the foot pressure sensing data set to the limb sensing data set for storage update, and the updated limb sensing data set includes a first-order inertial sensing data set, a second-order inertial sensing data set, and a foot pressure sensing data set.

[0012] Further, a motion scene recognition model is constructed. The motion scene recognition model includes a motion scene recognition basic layer and a motion scene recognition refinement layer. The motion scene recognition basic layer is obtained by training basic scene labels for sensor data samples based on a support vector machine. The motion scene recognition refinement layer uses a convolutional neural network to train refined scene labels for the sensor data samples and the basic scene labels. Among them, the basic scene labels at least include walking on flat ground, going up and down stairs, and walking on a ramp, and the refined scene labels include gait phase and slope. The motion scene recognition model is called to recognize the limb motion scene of the sensor data set to obtain the first motion scene.

[0013] Further, the correlation between the first motion scene and the multiple indicators is recognized to obtain multiple correlation coefficients. A parameter learning layer is constructed according to the multiple correlation coefficients to sort the multiple indicators, and multiple queue indicators are output. A parameter learning layer is constructed according to the multiple queue indicators, and the parameter learning layer is connected to the power output end of the power device for queue feedback learning to output auxiliary learning parameters.

[0014] Further, when the correlation coefficient regarding the power output step index in the multiple correlation coefficients is the largest, the weight ratio of the parameter learning layer of the power output step index is set to be greater than 0.5.

[0015] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0016] First, the power device of the exoskeleton robot is determined. Among them, the power device includes a power input end and a power output end. Then, multiple indicators for assisting the power output of the power device are obtained. The multiple indicators include a power output stability indicator, a power output continuity indicator, a power output step indicator, and a power output energy-saving indicator. Then, the sensor data set of the sensor module is received, and the limb motion scene is recognized according to the sensor data set to obtain the first motion scene. The parameter learning of the indicator order of the multiple indicators is performed according to the correlation between the first motion scene and the multiple indicators to obtain auxiliary learning parameters. Finally, the power device controls the power motor to adjust the power parameters according to the auxiliary learning parameters. This solves the technical problems in the prior art that the power assistance control of the exoskeleton robot is not intelligent and accurate enough, and it is difficult to dynamically adjust the power output according to the specific motion scene, resulting in low energy efficiency and poor user experience. By intelligently recognizing the limb motion scene and combining multiple power output indicators for parameter learning, the power parameters of the power device are accurately adjusted, thereby achieving the technical effects of improving energy efficiency and enhancing user experience. Description of the Drawings

[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention, the following will briefly introduce the accompanying drawings required for the description of the embodiments. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other accompanying drawings can be obtained based on these drawings.

[0018] Figure 1 Schematic flow diagram of a power-assisted control method for an exoskeleton robot based on artificial intelligence provided by an embodiment of the present application;

[0019] Figure 2 Schematic flow diagram of obtaining a first motion scenario in a power-assisted control method for an exoskeleton robot based on artificial intelligence provided by an embodiment of the present application. Detailed implementation manners

[0020] By providing a power-assisted control method for an exoskeleton robot based on artificial intelligence, the present application solves the technical problems in the prior art that the power-assisted control of exoskeleton robots is not intelligent and accurate enough, and it is difficult to dynamically adjust the power output according to specific motion scenarios, resulting in low energy efficiency and poor user experience.

[0021] The following will clearly and completely describe the technical solutions in the embodiments of the present application with reference to the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only some embodiments of the present application, rather than all embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present application without creative efforts fall within the scope of protection of the present application.

[0022] It should be noted that the terms "including" and "having" are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or server that includes a series of steps or units does not necessarily have to be limited to those steps or units clearly listed, but may include other steps or modules that are not clearly listed or are inherent to these processes, methods, products, or devices.

[0023] Embodiment, as Figure 1 shown, an embodiment of the present application provides a power-assisted control method for an exoskeleton robot based on artificial intelligence, wherein the method includes:

[0024] Determine the power device of the exoskeleton robot, wherein the power device includes a power input end and a power output end.

[0025] The power device of the exoskeleton robot is a device used to achieve energy transmission and drive control. The power device mainly includes a power input end and a power output end. The power input end is used to receive the energy provided by an external power source, such as a battery, a fuel cell or other power sources, and this part needs to have a stable and reliable energy access ability; while the power output end undertakes the function of converting the input energy into mechanical energy and outputting it to each actuator of the exoskeleton, and its design needs to ensure that the output energy meets the accuracy, continuity and stability requirements under different motion modes.

[0026] Furthermore, after determining the power device of the exoskeleton robot, judge whether the power device is a single power output; if the power device is a single power output, input the auxiliary learning parameters into the power input end of the power device, identify the initial power parameters according to the auxiliary learning parameters, and output the updated power parameters from the power output end; the power device controls the power motor to adjust the power parameters with the updated power parameters.

[0027] After determining the basic structure of the power device of the exoskeleton robot, next, judge the output mode of the power device to determine whether it is a single power output mode. If it is determined that the power device is a single power output mode, further input the auxiliary learning parameters obtained through intelligent recognition and multi-index parameter learning into the power input end of the power device. At this time, these auxiliary learning parameters are used to re-identify and analyze the initial power parameters originally set in the power system in order to accurately reflect the energy requirements in the current motion scenario. Then, after the parameter identification process, the power output end of the power device will output the updated power parameters according to the adjustment result, and these updated parameters are used as the basis for subsequent power adjustment. Finally, the power device controls the internal power motor and uses the updated power parameters to precisely adjust the output, so as to realize the real-time optimization of the power output of the exoskeleton robot during the auxiliary movement process.

[0028] Furthermore, if the power device is not a single power output, identify the multiple power output lines of the power device; input the auxiliary learning parameters into the multiple power input ends of the power device, identify the multiple initial power parameters according to the auxiliary learning parameters, and output the multiple updated power parameters from the power output end; the power device controls the power motor to adjust the power parameters with the multiple updated power parameters.

[0029] When the power device has non-single power output, by identifying multiple power output lines in the power device, it is ensured that the system can accurately adjust each line separately. Specifically, first, through detection and identification technologies, each independent power output line in the power device is determined, and the corresponding functions and scope of action of each line are clarified; subsequently, the auxiliary learning parameters obtained through motion scenario recognition and multi-index parameter learning are respectively input into each power input end, so as to identify and correct the initial power parameters of each output line. Next, based on the input auxiliary learning parameters, the system analyzes and processes multiple initial power parameters, generates multiple updated power parameters corresponding to each output line, and the power output end outputs these updated parameters. Finally, the power device precisely controls the internal power motor and adjusts the power according to the updated power parameters of each line, so as to achieve more detailed and dynamic energy distribution in the multi-output mode, and ensure that the exoskeleton robot can achieve efficient, accurate and energy-saving power assistance in complex motion scenarios.

[0030] Furthermore, after the multiple updated power parameters are output by the power output end, the method further includes:

[0031] Establish a collaborative line group for the multiple power output lines; group the multiple updated power parameters according to the collaborative line group, output an updated power parameter group, and perform collaborative update on the updated power parameters within the updated power parameter group, and output multiple collaboratively updated power parameters; the power device controls the power motor to adjust the power parameters with the multiple collaboratively updated power parameters.

[0032] After the multiple updated power parameters are output by the power output end, a collaborative line group among the multiple power output lines is further established. Specifically, the system first identifies each independent power output line, and divides them into one or more collaborative line groups according to the functions, working states and mutual correlations of each line in the overall power transmission. Subsequently, the system groups the multiple updated power parameters generated by each power output line according to the established collaborative line group to form corresponding updated power parameter groups. Within each updated power parameter group, the system uses a preset collaborative update algorithm to integrate and collaboratively update the updated power parameters within the group, so as to output multiple collaboratively updated power parameters, ensuring the unity and coordination of the parameters among the output lines. Finally, based on these collaboratively updated power parameters, the power device precisely adjusts the power parameters by controlling the internal power motor, so as to achieve the purpose of optimizing the efficient, accurate and energy-saving power output of the exoskeleton robot in different motion scenarios.

[0033] Obtain multiple indicators for assisting the power output of the power device, where the multiple indicators include power output stability indicator, power output continuity indicator, power output step indicator, and power output energy-saving indicator.

[0034] In order to accurately adjust the power output, the system needs to preset multiple indicators related to the power output performance, including power output stability indicator, power output continuity indicator, power output step indicator, and power output energy-saving indicator. Specifically, by analyzing and processing the real-time data collected by the sensor, the system first calculates the power output stability indicator to evaluate the fluctuation and stability of the power output during long-term operation; subsequently, by measuring the continuity and smoothness of the output signal, the power output continuity indicator is extracted to reflect the continuity and consistency in the energy transfer process; furthermore, when the system detects a sudden change in the load or control instruction, the power output step indicator is obtained by measuring the response rate and amplitude to evaluate the response performance of the system under sudden change conditions; finally, by comparing the actual output power with the input energy, the power output energy-saving indicator is calculated to reflect the efficiency and energy-saving effect of the overall energy conversion.

[0035] By analyzing and processing the real-time data collected by the sensor, the system first preprocesses the original data, including operations such as noise filtering and data smoothing, to ensure the accuracy and reliability of the data. The preprocessed data reflects the actual operating conditions of the exoskeleton robot in different working states. Subsequently, based on these data, the system uses a dynamic analysis algorithm to calculate the indicator reflecting the power output stability. Specifically, the power output stability indicator determines whether there are significant fluctuations or abnormal changes in the output signal by evaluating the fluctuation amplitude, change rate, and long-term trend of the power output parameters (such as current, voltage, rotation speed, etc.), thereby quantifying the stability of the system's power output.

[0036] Receive the sensing data set of the sensing module, identify the limb movement scenario according to the sensing data set, obtain the first movement scenario, and perform parameter learning on the order of the multiple indicators according to the correlation between the first movement scenario and the multiple indicators to obtain the auxiliary learning parameters.

[0037] The system forms a sensing data set (such as joint angles, angular velocities, ground reaction forces, etc.) by receiving multi-dimensional data collected by the sensing module during the user's movement. Then, the sensing data set is preprocessed to remove noise and abnormal data; subsequently, the preprocessed data is analyzed using a trained motion scenario recognition model (such as a classification network based on machine learning or deep learning) to accurately determine the current limb motion scenario of the user, and then the first motion scenario (such as walking, running, or going up and down stairs, etc.) is obtained; after identifying the first motion scenario, the system sorts or assigns weights to these indicators according to the correlation between this scenario and the foregoing multiple power output indicators (such as power output stability indicator, power output continuity indicator, power output step indicator, power output energy-saving indicator), and on this basis, parameter learning is carried out, that is, the contribution degree of each indicator to the final decision is dynamically adjusted to obtain the optimal configuration, thereby obtaining the auxiliary learning parameters; the auxiliary learning parameters include a comprehensive consideration of achieving high stability, continuity, step response ability, and energy-saving effect in the current motion scenario, providing targeted guidance and optimization basis for the subsequent power device during the power output adjustment process.

[0038] Furthermore, the sensing module includes an environmental sensing module, and the environmental sensing module includes a CMOS sensor, a radar sensor, a GPS, and an RTK sensor, which are used to collect sensing data on the environment where the wearable user of the exoskeleton robot is located; the environmental sensing data set collected by the environmental sensing module is obtained, and the environmental sensing data set is added to the sensing data set.

[0039] In order to more comprehensively obtain information about the user's environment and thus more accurately adjust the auxiliary control strategy of the exoskeleton robot, the system is configured with an environmental sensing module, which includes various sensing devices such as a CMOS sensor, a radar sensor, a GPS, and an RTK sensor, and is used to sense visual information, obstacle distribution, and precise position information of the user's surrounding environment, etc. Specifically, the CMOS sensor can capture environmental image data in real time, the radar sensor is responsible for detecting the distance and azimuth of obstacles or target objects, and the GPS and RTK sensors provide higher-precision position information and positioning capabilities. By preprocessing the environmental sensing data set obtained by the above environmental sensing module and merging it with the user's body sensation sensing data into the same sensing data set, the system can more accurately evaluate the current environmental conditions and motion requirements of the user during the subsequent motion scenario recognition and power parameter learning stages, thereby improving the intelligence and adaptability of the exoskeleton robot's auxiliary output in different scenarios.

[0040] Furthermore, the sensing module further includes a limb sensing module, and the limb sensing module includes a first-order inertial sensor and a second-order inertial sensor; the first-order inertial sensor is used for collecting upper limb inertial motion sensing data of the user wearing the exoskeleton robot, and the second-order inertial sensor is used for collecting lower limb inertial motion sensing data of the user wearing the exoskeleton robot; obtaining the limb sensing data set collected by the limb sensing module, storing the limb sensing data set into the sensing data set, and synchronizing and aligning the environmental sensing data set and the limb sensing data set in time series by using the dynamic time warping (DTW) algorithm for storage.

[0041] To more comprehensively capture the motion characteristics of the user wearing the exoskeleton robot at different limb parts, the system is also configured with a limb sensing module, which includes a first-order inertial sensor and a second-order inertial sensor. Among them, the first-order inertial sensor is used for real-time collection of the upper limb inertial motion parameters of the user, and the second-order inertial sensor is used for real-time collection of the lower limb inertial motion parameters of the user; after obtaining the limb sensing data set collected by the above-mentioned limb sensing module, the system will store the data set into a unified sensing data set, and at the same time, it will also perform time series synchronization and alignment on the environmental sensing data set obtained by the previous environmental sensing module and the limb sensing data set through the dynamic time warping (DTW) algorithm, so that the two are exactly consistent in the time dimension; this not only ensures that in the subsequent identification of the user's motion scenario and the adjustment of power output, the environmental information and the limb motion information can be effectively fused and analyzed on the same time basis, but also lays a data foundation for the exoskeleton robot to achieve more accurate motion assistance in a complex environment.

[0042] To ensure that the recognition of the user's motion scenario and the subsequent power assistance decision-making can be comprehensively analyzed based on the environmental information and the limb motion information on the same time basis, after obtaining the environmental sensing data set and the limb sensing data set, the system uses the dynamic time warping (DTW) algorithm to achieve time series synchronization and alignment of these two parts of data. Specifically, the DTW algorithm calculates the similarity between different data sequences and performs a non-linear mapping on the time axis, so that even when the sampling rate or sampling duration is different, the environmental data and the limb data can be accurately matched and aligned in the time dimension. Through this time series synchronization and alignment process, it can be ensured that in the subsequent motion scenario recognition and auxiliary learning parameter calculation processes, various sensing information is on the same time basis, avoiding analysis errors caused by inconsistent data timestamps, thereby improving the perception accuracy and response efficiency of the exoskeleton robot to the user's motion state and the surrounding environment.

[0043] Furthermore, it is determined whether the exoskeleton robot is worn on the leg. If the exoskeleton robot is worn on the leg, a foot pressure sensor is set, and the foot pressure sensing data of the user wearing the exoskeleton robot is collected according to the foot pressure sensor to obtain a foot pressure sensing data set; the foot pressure sensing data set is added to the limb sensing data set for storage and update, and the updated limb sensing data set includes a first-order inertial sensing data set, a second-order inertial sensing data set, and a foot pressure sensing data set.

[0044] To accurately identify and collect the sole force of the user when wearing the exoskeleton robot on the leg, the system first determines whether the exoskeleton robot is in the leg-wearing state; if it is confirmed that it is in the leg-wearing state, the foot pressure sensor set at the corresponding foot position of the exoskeleton robot is activated to collect the sole pressure sensing data of the user in real time and form a foot pressure sensing data set. Subsequently, the system adds the foot pressure sensing data set to the obtained first-order inertial sensing data set and second-order inertial sensing data set for storage and update, so that the updated limb sensing data set contains upper limb inertial data, lower limb inertial data, and sole pressure data at the same time.

[0045] Furthermore, as Figure 2 shown, according to the sensing data set, limb movement scenario recognition is performed to obtain a first movement scenario. The method includes:

[0046] A movement scenario recognition model is constructed. The movement scenario recognition model includes a movement scenario recognition basic layer and a movement scenario recognition refinement layer. The movement scenario recognition basic layer is obtained by training basic scenario labels for sensing data samples based on a support vector machine, and the movement scenario recognition refinement layer uses a convolutional neural network to train refined scenario labels for the sensing data samples and the basic scenario labels; wherein, the basic scenario labels at least include flat ground walking, going up and down stairs, and ramp walking, and the refined scenario labels include gait phase and slope; the movement scenario recognition model is called to perform limb movement scenario recognition on the sensing data set to obtain a first movement scenario.

[0047] Preferably, in order to more accurately identify the limb movement scenario and obtain the first movement scenario, a movement scenario recognition model composed of a movement scenario recognition basic layer and a movement scenario recognition refinement layer is first constructed: Among them, the movement scenario recognition basic layer trains the sensing data samples based on the support vector machine (SVM) to obtain basic scenario labels such as flat ground walking, going up and down stairs, and ramp walking; On this basis, the movement scenario recognition refinement layer uses the convolutional neural network (CNN) to further refine the training of the same batch of sensing data samples and basic scenario labels to obtain more refined scenario labels such as gait phase and slope. Through this hierarchical recognition method, the category of the overall movement scenario can be quickly determined first; Next, use CNN to extract features and perform deep learning on the classified basic scenario data to identify specific movement details (such as the gait phase corresponding to different walking stages or the specific walking slope). The system processes the time-series aligned sensing data set by calling the movement scenario recognition model, first uses SVM for rough scenario classification, and then uses CNN for refinement classification, so as to finally obtain the first movement scenario including basic scenario information and scenario details, providing a more accurate and multi-dimensional reference for the subsequent calculation of the auxiliary learning parameters of the power device and the adjustment of the power output.

[0048] Furthermore, parameter learning of the order of the multiple indicators is performed according to the correlation between the first movement scenario and the multiple indicators, and the method includes:

[0049] Identify the correlation between the first movement scenario and the multiple indicators, and obtain multiple correlation coefficients; Construct a parameter learning layer according to the multiple correlation coefficients to sort the multiple indicators, output multiple queue indicators, construct a parameter learning layer according to the multiple queue indicators, connect the parameter learning layer to the power output end of the power device for queue feedback learning, and output auxiliary learning parameters.

[0050] After the system recognizes the first movement scenario, it will first calculate the correlation between the movement scenario and multiple indicators such as the power output stability index, the power output continuity index, the power output step index, and the power output energy saving index, and obtain the corresponding multiple correlation coefficients; Based on these correlation coefficients, the system performs priority or weight sorting on the above indicators in the parameter learning layer to form multiple queue indicators. Subsequently, the parameter learning layer will connect it to the power output end of the power device according to the obtained multiple queue indicators, and perform queue-based feedback learning on the priority order of each indicator to continuously correct and optimize the required weights and algorithm parameters. Through this mechanism, the system can adaptively adjust the attention to different indicators and finally output auxiliary learning parameters, providing a targeted configuration plan for the power output of the exoskeleton robot, so that the power device can output and adjust in a more reasonable and accurate manner under the identified first movement scenario.

[0051] To quantify the correlation between the first motion scenario and the power output stability index, power output continuity index, power output step index, and power output energy-saving index, the system first compares the power output characteristics collected in this motion scenario with the reference values of each index according to historical sample data or real-time sensing data, and conducts statistical analysis in combination with the definitions of each index (such as fluctuation measurement to measure stability, output smoothness to measure continuity, step response to measure mutation adaptability, and energy utilization rate to measure energy-saving effect); specifically in implementation, methods such as Pearson correlation and Spearman correlation can be used to quantitatively calculate the deviation degree, fluctuation trend, and response rate between different scenario data and the indexes, so as to obtain one or more correlation coefficients, respectively reflecting the sensitivity and importance of the first motion scenario to each index. Through the measurement of the above-mentioned correlation, it can provide a reliable basis for sorting and weight allocation of multiple indexes in the subsequent parameter learning layer.

[0052] Furthermore, when the correlation coefficient regarding the power output step index is the largest among the multiple correlation coefficients, the weight ratio of the parameter learning layer of the power output step index is made greater than 0.5.

[0053] When the calculated multiple correlation coefficients show that the correlation coefficient of the power output step index is the maximum value, the system will specially increase the weight of this step index in the parameter learning layer. The specific implementation method is: make the weight ratio of the parameter learning layer of the power output step index exceed 0.5, so that this index can obtain a higher priority in the subsequent queue feedback learning and auxiliary learning parameter generation process. In this way, the system can ensure a significant increase in the attention to the power output step characteristics in a motion situation where the scene changes rapidly or high-sensitivity response is required, and achieve a more rapid adaptation to the mutation load or action switching scene.

[0054] The power device controls the power motor to adjust the power parameters according to the auxiliary learning parameters.

[0055] After obtaining the auxiliary learning parameters, the power device will input the parameters into the internal control module. The control module will adjust the operating mode of the power motor (such as output torque, speed or current, etc.) in real time according to the optimization strategies for different power output indicators (such as stability, continuity, step response and energy saving, etc.) in the auxiliary learning parameters. Specifically, the system will compare the auxiliary learning parameters with the currently collected sensing data (including the force on the exoskeleton joint, the movement amplitude, the user's physical state, etc.) by monitoring the working state of the power motor in real time. If a deviation is detected, the motor will be corrected in terms of speed or torque according to the preset threshold and adjustment range in the auxiliary learning parameters, so that the power output tends to the optimal state, and ultimately realizes the efficient energy utilization and precise power support for the exoskeleton robot.

[0056] In summary, the embodiments of the present application at least have the following technical effects:

[0057] First, determine the power device of the exoskeleton robot, where the power device includes a power input end and a power output end. Then, obtain multiple indicators for assisting the power output of the power device, and the multiple indicators include a power output stability indicator, a power output continuity indicator, a power output step indicator, and a power output energy saving indicator. Next, receive the sensing data set of the sensing module, identify the limb movement scenario according to the sensing data set, obtain the first movement scenario, and perform parameter learning on the order of the multiple indicators according to the correlation between the first movement scenario and the multiple indicators to obtain the auxiliary learning parameters. Finally, the power device controls the power motor to adjust the power parameters according to the auxiliary learning parameters. This solves the technical problems in the prior art that the power assistance control of the exoskeleton robot is not intelligent and accurate enough, and it is difficult to dynamically adjust the power output according to the specific movement scenario, resulting in low energy efficiency and poor user experience. By intelligently identifying the limb movement scenario and combining multiple power output indicators for parameter learning, the power parameters of the power device are accurately adjusted, thus achieving the technical effects of improving energy efficiency and enhancing the user experience.

[0058] It should be noted that the above sequence of the embodiments of the present application is only for description and does not represent the advantages and disadvantages of the embodiments. And the above describes specific embodiments of this specification. The processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0059] The above are only the preferred embodiments of the present application and are not intended to limit the present application. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

[0060] This specification and the accompanying drawings are merely illustrative of the present application and are considered to cover any and all modifications, variations, combinations, or equivalents within the scope of the present application. Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the present application and its equivalent technologies, the present application is intended to include these changes and modifications therein.

Claims

1. A power-assisted control method for an exoskeleton robot based on artificial intelligence, characterized in that: The method comprises: Determine a power device of the exoskeleton robot, wherein the power device includes a power input end and a power output end; Acquiring multiple indicators for assisting the power output of the power device, the multiple indicators including a power output stability indicator, a power output continuity indicator, a power output step indicator, and a power output energy-saving indicator; Receiving a sensor data set from a sensor module, performing limb movement scene recognition according to the sensor data set, acquiring a first movement scene, and performing parameter learning of the indicator sequence for the multiple indicators according to the correlation between the first movement scene and the multiple indicators to obtain auxiliary learning parameters; The power device controls the power motor to adjust the power parameters according to the auxiliary learning parameters; The sensor module includes an environmental sensor module, which includes a CMOS sensor, a radar sensor, a GPS and an RTK sensor, and is used to collect sensor data of the environment in which the user wearing the exoskeleton robot is located; Acquire the environmental sensing data set collected by the environmental sensing module, and add the environmental sensing data set to the sensing data set; The sensor module also includes a limb sensor module, and the limb sensor module includes a first-order inertial sensor and a second-order inertial sensor; The first-order inertial sensor is used to collect upper limb inertial motion sensing data of the user wearing the exoskeleton robot, and the second-order inertial sensor is used to collect lower limb inertial motion sensing data of the user wearing the exoskeleton robot; Acquire a limb sensor data set collected by the limb sensor module, store the limb sensor data set in a sensor data set, and perform time-series synchronization and alignment storage on the environment sensor data set and the limb sensor data set using a dynamic time warping (DTW) algorithm; Performing limb movement scene recognition according to the sensor data set to obtain a first movement scene, the method comprising: Constructing a motion scene recognition model, the motion scene recognition model comprising a motion scene recognition base layer and a motion scene recognition refinement layer, the motion scene recognition base layer performs basic scene label training on sensor data samples based on a support vector machine to obtain basic scene labels, and the motion scene recognition refinement layer performs refined scene label training on the sensor data samples and the basic scene labels using a convolutional neural network to obtain; The basic scene labels at least include walking on flat ground, going up and down stairs, and walking on a ramp, and the detailed scene labels include gait phase and slope; Calling the motion scene recognition model to perform limb motion scene recognition on the sensor data set to obtain a first motion scene; According to the correlation between the first motion scene and the multiple indicators, parameter learning of the indicator sequence is performed on the multiple indicators, and the method includes: Identifying the correlation between the first motion scene and the multiple indicators, and obtaining multiple correlation coefficients; A parameter learning layer is constructed according to the multiple correlation coefficients to sort the multiple indicators, and multiple queue indicators are output. A parameter learning layer is constructed according to the multiple queue indicators, and the parameter learning layer is connected to the power output end of the power device to perform queue feedback learning, and output auxiliary learning parameters.

2. The artificial intelligence-based exoskeleton robot power-assisted control method according to claim 1, characterized in that: After determining the power device of the exoskeleton robot, determining whether the power device is a single power output; If the power device has a single power output, the auxiliary learning parameter is input into the power input end of the power device, the initial power parameter is identified according to the auxiliary learning parameter, and the updated power parameter is output from the power output end; The power device controls the power motor to adjust the power parameters according to the updated power parameters.

3. The artificial intelligence-based exoskeleton robot power-assisted control method according to claim 2, characterized in that: If the power device does not have a single power output, identifying multiple power output lines of the power device; Inputting the auxiliary learning parameters into a plurality of power input terminals of the power device, identifying a plurality of initial power parameters according to the auxiliary learning parameters, and outputting a plurality of updated power parameters from the power output terminal; The power device controls the power motor to adjust the power parameters using the multiple updated power parameters.

4. The artificial intelligence-based exoskeleton robot power-assisted control method according to claim 3, characterized in that: After the power output terminal outputs a plurality of updated power parameters, the method further includes: Establishing a coordinated line group of the plurality of power output lines; Grouping the multiple updated power parameters according to the cooperative line group, outputting an updated power parameter group, and cooperatively updating the updated power parameters in the updated power parameter group, outputting multiple cooperatively updated power parameters; The power device controls the power motor to adjust the power parameters using the multiple coordinated updated power parameters.

5. The artificial intelligence-based exoskeleton robot power-assisted control method according to claim 1, characterized in that: Determine whether the exoskeleton robot is worn on the legs. If the exoskeleton robot is worn on the legs, set a foot pressure sensor, collect plantar pressure sensing data of the user wearing the exoskeleton robot according to the foot pressure sensor, and obtain a foot pressure sensing data set; The foot pressure sensing data set is added to the limb sensing data set for storage and update, and the updated limb sensing data set includes a first-order inertial sensing data set, a second-order inertial sensing data set and a foot pressure sensing data set.

6. The artificial intelligence-based exoskeleton robot power-assisted control method according to claim 1, characterized in that: If the correlation coefficient of the power output step indicator among the multiple correlation coefficients is the largest, the parameter learning layer weight ratio of the power output step indicator is set to be greater than 0.5.

Images