Multi-module switching system based on AI model and application of multi-module switching system in health care robot

Abstract

The invention relates to the field of health care robots, in particular to a multi-module switching system based on an AI model and application of the multi-module switching system in a health care robot, and the multi-module switching system comprises a posture sensing module, a signal cleaning module, a model approaching module, a reaction output module and a memory training module, the signal cleaning module is used for generating a random Gaussian noise training AI model, the model approaching module is used for calculating the matching degree of robot sensing signals and all the modules, the reaction output module is used for activating robot actions, and the memory training module is used for constructing a reaction memory library and adjusting subsequent matching weights. According to the method, the signals collected by the sensor can be clearer and more stable, the signal quality is improved, the input error of the sensor is reduced, the resource utilization efficiency of the system is optimized, the overall performance is improved, meanwhile, the flexibility and adaptability of the AI model are improved, and the stability and reliability of the health care robot system are improved.

Description

Technical Field

[0001] The present invention relates to the field of health care robots, and specifically to a multi-module switching system based on an AI model and its application in health care robots. Background Art

[0002] A health care robot is an intelligent robot that provides rehabilitation and maintenance services. This robot can monitor the user's physiological state and activities through sensors and cameras, provide customized rehabilitation plans and maintenance services for the elderly or injured people, and realize functions such as rehabilitation training and health status monitoring.

[0003] The health care robot senses the external state by identifying the potential point in the sensor, analyzes the electrical signal, and then drives the robot to output action. Since the physical state and behavioral characteristics of the service object are constantly changing, the robot needs to be equipped with different modules, but most health care robots currently have no unified standards for the conditions for switching modules, and cannot provide users with personalized module switching solutions.

[0004] In addition, since the service environment of the health care robot is relatively complex, and the complex environment easily produces varying degrees of interference in the sensor's sensor signal, the robot cannot accurately obtain external sensor information, affecting the robot's module switching judgment and action output efficiency, which greatly reduces the robot's adaptability and service quality. Summary of the invention

[0005] The purpose of the present invention is to provide a multi-module switching system based on an AI model and its application in a health care robot to solve the problems raised in the above-mentioned background technology.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions: a multi-module switching system based on an AI model, comprising: a posture sensing module, a signal cleaning module, a model approaching module, a reaction output module and a memory training module; The posture sensing module is used to pre-transmit standard signal sites in the robot's posture sensor, each type of electrical signal site corresponds to the specific physical parameters, behavioral parameters and posture parameters of the health care object, obtain historical classification results, generate a historical site set of each type of electrical signal, and calculate the aggregation degree of the set, which reflects the average intensity of environmental noise; The signal cleaning module is used to add random Gaussian noise with a distribution divergence equal to the degree of aggregation to the standard electrical signal sites according to the set aggregation degree, to obtain a random noise set, to use the random noise set as a training library and the standard electrical signal as an output, to train the AI ​​model, and to use the trained AI model to parse the electrical signal sites of the posture sensor and convert them into corresponding standard electrical signal outputs; The model approach module is used to collect all sensor data within one action cycle of the robot, simulate the multidimensional distribution of the electrical signal sites corresponding to all sensor data in the virtual space, perform the same operation with the multidimensional distribution model between the health care modules, obtain the degree of proximity between the multidimensional distribution and each health care module, and arrange all modules in the order of proximity; The reaction output module is used to generate a robot reaction library according to the module proximity and the corresponding output of each module, using each identified electrical signal site as a key value, matching the corresponding output of each health care module, calculating the output value of the matching result in each module according to the module proximity and matching consistency, and activating the robot reaction in sequence according to the output value; The memory training module is used to perform convergence training on the robot reactions triggered by adjacent electrical signal sites to form a reaction memory bank, and adjust the convergence weight of the health care module and the diffusion divergence of the signal recognition model according to the matching degree of the activated electrical signal sites in the reaction memory bank.

[0007] Further, the posture sensing module includes: a signal location unit and an environment aggregation unit; The signal location unit is used to format the sensors of the health care robot, determine the standard signal location according to the sensor hardware parameters, and is managed uniformly by the main control chip; The environment aggregation unit is used to record the actual electrical signal position of the sensor each time the sensor returns a signal, generate a set linked list in the database, and calculate the aggregation degree of the set.

[0008] Further, the signal cleaning module includes: a random noise unit and an impurity removal training unit; The random noise unit is used to generate random Gaussian noise with a divergence equal to the degree of aggregation according to the degree of aggregation of elements in the historical site set; The impurity removal training unit is used to add random Gaussian noise into the standard signal site to generate training noise, and use the training noise to train the AI ​​classification model, so that the AI ​​classification model has data cleaning capabilities.

[0009] Furthermore, the model approach module includes: a spatial distribution unit, a multi-simulation combination unit and a model switching unit; The spatial distribution unit is used to spread the electrical signal sites of all sensors in the virtual space to form an actual signal model; The multi-simulation unit is used to simulate the service interval of each module of the robot in the virtual space, and calculate the proportion of the actual signal model located in the service interval as the approach output; The model switching unit arranges all modules in order of proximity and switches the service module of the robot to the first module.

[0010] Further, the reaction output module includes: a key-value matching unit and a behavior activation unit; The key value matching unit is used to identify the subsequent electrical signal site, use the subsequent electrical signal site as the key value, accumulate the output actions of each module according to the mapping amount and weight, and match the corresponding output in the current service module; The behavior activation unit is used to activate the output action of the robot according to the matching result, and end the service cycle after the robot performs the output action for a fixed time.

[0011] Furthermore, the memory training module includes: a convergent classification unit, a memory library unit and a memory weighting unit; The convergence classification unit is used to perform convergence training on the electrical signal data and the robot output using the AI ​​model; The memory unit is used to record the electrical signal locations of each sensor and the robot movements during the service cycle in a database; The memory weighting unit is used to adjust the matching weights of each module using the memory bank according to the matching degree of the activated electrical signal sites in the memory bank in the subsequent action cycle.

[0012] The application of the multi-module switching system based on the AI ​​model in the health care robot includes the following steps: Step S1. Determine the standard signal location according to the hardware parameters of the posture sensor in the health care robot, and record the actual electrical signal location of the sensor each time the sensor returns a signal, and generate a signal set in the database; Step S2. Calculate the aggregation degree of the signal set, generate random Gaussian noise consistent with the discrete degree and aggregation degree, add the random Gaussian noise into the standard signal site, obtain a noise library, use the noise library to train the AI ​​model, and obtain an AI model with data cleaning capabilities; Step S3. Collect all sensor data of the robot during the service cycle, spread each sensor data in the virtual space, obtain the actual signal model, calculate the mapping amount of the actual signal model in each service module space, and arrange each service module in descending order according to the mapping amount; Step S4. Switch the working module of the robot to the service module ranked first, use the subsequent electrical signal sites as key values, add up the output actions of each module according to the mapping amount and weight, obtain the value of each action, activate the action with the highest output value of the robot, and end the current cycle after the robot performs the output action for a fixed time; Step S5. Record the electrical signal sites and actual action outputs of all sensors within the cycle to form a memory bank, and adjust the weight of each action in the subsequent service cycle according to the sensor information uploaded within the cycle.

[0013] Further, step S1 includes: Step S11. Format each sensor of the health care robot, and determine the standard signal site according to the hardware parameters of the sensor, wherein the hardware parameters of the sensor include: sensor piezoelectric efficiency, temperature and humidity sensing efficiency, and human body monitoring data conversion efficiency, so that the standard signal site of each sensor corresponds to the specific body parameters, behavior parameters, and posture parameters of the health care object; Step S12. When the sensor generates a corresponding electrical signal according to external physical changes, the level strength value of each electrical signal site is recorded in the database, and all recorded electrical signal site data are stored in a set form to form a signal set A, wherein A={T1, T2,…, Tn}, where n represents the number of records, and Tn represents the level strength of the electrical signal site in the nth record.

[0014] Further, step S2 includes: Step S21. Calculate the degree of aggregation of the signal set, where the degree of aggregation represents the average of the squares of the differences between the elements in the set and the standard signal sites, satisfying I=1 / n·[(T1-T0) 2 + (T2-T0) 2 +…+(Tn-T0) 2 ], where I is the degree of polymerization and T0 is the level intensity of the standard signal site; Step S22. Generate random Gaussian noise according to the degree of aggregation, so that the generated noise meets the following conditions: Where F(t) represents random Gaussian noise, t represents time, t0 represents the time when the sensor signal occurs, and e is the base of the natural logarithm; Step S23. Randomly change the value of t to obtain a Gaussian noise set, add each element in the random Gaussian noise set to the standard signal site, generate a noise library, and use the noise library to train the AI ​​model so that when the AI ​​model inputs the signal site, it automatically outputs the standard signal site with the closest level intensity, thereby realizing the cleaning process of the signal site. The AI ​​models include: Mask R-CNN model, DeepLab model, U-Net model and Watershed model.

[0015] Further, step S3 includes: Step S31. Acquire the electrical signals of each sensor of the robot during the service cycle and clean the electrical signals. There is only one electrical signal for the same sensor during the service cycle. When the same sensor transmits multiple electrical signals during the service cycle, end the current service cycle and allocate the excess electrical signals to the next service cycle. Step S32. Generate an m-dimensional virtual space in the control chip, and annotate the collected electrical signals in the form of vectors, where m represents the number of electrical signals in the service cycle, and obtain an actual signal model V, where V=(d1, d2, ..., dm), where dm represents the level strength of the electrical signal transmitted by the mth sensor; Step S33. Expand the robot module in the form of intervals and represent the robot module as {[D1+, D1-], [D2+, D2-], …, [Dn+, Dn-]}, where D1+, D2+ and Dn+ represent the upper bounds of the intervals of the electrical signal level strengths of the sensors numbered 1, 2 and n, respectively, and D1-, D2- and Dn- represent the lower bounds of the intervals of the electrical signal level strengths of the sensors numbered 1, 2 and n, respectively. Calculate the mapping vector of the actual signal model V in the n-dimensional space, record the module length of the portion of the interval where the mapping vector falls in each robot module as the mapping amount of the actual signal model in each module space, and arrange the modules in descending order of the mapping amount.

[0016] Further, step S4 includes: Step S41. Switch the robot's working module to the first module with the highest mapping amount, and after receiving the subsequent signal, calculate the value of each action according to the following formula: Among them, E represents the value of the action, k represents the number of robot service modules, and u i represents the weight of the i-th service module, p i represents the probability of the action being called in the i-th service module, d i Represents the influence parameter of the subsequent signal on the actual signal model, which is determined by the rules of the AI ​​model, r i Represents the mapping amount of the actual signal model in the i-th service module; Step S42: Select the action with the highest value as the output action of the current service cycle, activate the robot to perform the output action, and end the service cycle after the robot performs the output action for a fixed time.

[0017] Further, step S5 includes: Step S51. Record the electrical signal sites and robot actions of each sensor during the service period, and use the AI ​​model to perform convergence training on the electrical signal data and the robot output. When the convergence training result is positive feedback, store the electrical signal sites of the sensor and the robot action as a memory pattern in the memory bank; Step S52. In the subsequent action cycle, the matching degree of each electrical signal site in the actual signal model in the memory library is calculated, and the weight of the action recorded in the memory unit in each service module is adjusted according to the matching degree of the memory unit.

[0018] Compared with the prior art, the beneficial effects achieved by the present invention are: 1. The present invention can pre-transmit the electrical signal sites in the attitude sensor, and add random noise with a distribution trend equal to the aggregation degree to the pre-transmitted sites in the historical site set of each type of electrical signal according to the set aggregation degree, so as to clean the recognition results of the sensor, so that the signals collected by the sensor can be clearer and more stable, thereby improving the signal quality and reducing the input error of the sensor.

[0019] 2. The present invention can simulate the multidimensional distribution of electrical signal sites in space, fit the proximity of each health care module, use the identified electrical signal sites as key values, match the corresponding outputs built in each health care module, and activate the robot's behavioral responses in turn, so that the health care robot can quickly switch between different scenes or tasks, enhance its ability to adapt to different environments and tasks, flexibly configure the resource use of the health care robot, optimize the resource utilization efficiency of the system, and improve the overall performance.

[0020] 3. The present invention can perform convergence training on robot reactions triggered by adjacent electrical signal sites, and adjust the convergence weight of the health care module according to the classification strength of the activated electrical signal sites in the reaction memory bank, which can improve the flexibility and adaptability of the AI ​​model and improve the stability and reliability of the health care robot system. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings: Figure 1 It is a structural schematic diagram of a multi-module switching system based on an AI model of the present invention; Figure 2 It is a schematic diagram of the steps of applying the multi-module switching system based on the AI ​​model of the present invention in a health care robot. DETAILED DESCRIPTION

[0022] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0023] See also Figure 1 , the present invention provides a technical solution: a multi-module switching system based on an AI model, including: a posture sensing module, a signal cleaning module, a model approaching module, a reaction output module and a memory training module; The posture sensing module is used to pre-transmit standard signal sites in the robot's posture sensor, each type of electrical signal site corresponds to the specific physical parameters, behavioral parameters and posture parameters of the health care object, obtain historical classification results, generate a historical site set of each type of electrical signal, and calculate the aggregation degree of the set, which reflects the average intensity of environmental noise; The posture sensing module includes: a signal location unit and an environment aggregation unit; The signal location unit is used to format the sensors of the health care robot, determine the standard signal location according to the sensor hardware parameters, and is managed uniformly by the main control chip; The environment aggregation unit is used to record the actual electrical signal position of the sensor each time the sensor returns a signal, generate a set linked list in the database, and calculate the aggregation degree of the set.

[0024] The signal cleaning module is used to add random Gaussian noise with a distribution divergence equal to the degree of aggregation to the standard electrical signal sites according to the set aggregation degree, to obtain a random noise set, to use the random noise set as a training library and the standard electrical signal as an output, to train the AI ​​model, and to use the trained AI model to parse the electrical signal sites of the posture sensor and convert them into corresponding standard electrical signal outputs; The signal cleaning module includes: a random noise unit and an impurity removal training unit; The random noise unit is used to generate random Gaussian noise with a divergence equal to the degree of aggregation according to the degree of aggregation of elements in the historical site set; The impurity removal training unit is used to add random Gaussian noise into the standard signal site to generate training noise, and use the training noise to train the AI ​​classification model, so that the AI ​​classification model has data cleaning capabilities.

[0025] The model approach module is used to collect all sensor data within one action cycle of the robot, simulate the multidimensional distribution of the electrical signal sites corresponding to all sensor data in the virtual space, perform the same operation with the multidimensional distribution model between the health care modules, obtain the degree of proximity between the multidimensional distribution and each health care module, and arrange all modules in the order of proximity; The model approach module includes: a spatial distribution unit, a multi-simulation unit and a model switching unit; The spatial distribution unit is used to spread the electrical signal sites of all sensors in the virtual space to form an actual signal model; The multi-simulation unit is used to simulate the service interval of each module of the robot in the virtual space, and calculate the proportion of the actual signal model located in the service interval as the approach output; The model switching unit arranges all modules in order of proximity and switches the service module of the robot to the first module.

[0026] The reaction output module is used to generate a robot reaction library according to the module proximity and the corresponding output of each module, using each identified electrical signal site as a key value, matching the corresponding output of each health care module, calculating the output value of the matching result in each module according to the module proximity and matching consistency, and activating the robot reaction in sequence according to the output value; The reaction output module includes: a key value matching unit and a behavior activation unit; The key value matching unit is used to identify the subsequent electrical signal site, use the subsequent electrical signal site as the key value, accumulate the output actions of each module according to the mapping amount and weight, and match the corresponding output in the current service module; The behavior activation unit is used to activate the output action of the robot according to the matching result, and end the service cycle after the robot performs the output action for a fixed time.

[0027] The memory training module is used to perform convergence training on the robot reactions triggered by adjacent electrical signal sites to form a reaction memory bank, and adjust the convergence weight of the health care module and the diffusion divergence of the signal recognition model according to the matching degree of the activated electrical signal sites in the reaction memory bank.

[0028] The memory training module includes: a convergent classification unit, a memory library unit and a memory weighting unit; The convergence classification unit is used to perform convergence training on the electrical signal data and the robot output using the AI ​​model; The memory unit is used to record the electrical signal locations of each sensor and the robot movements during the service cycle in a database; The memory weighting unit is used to adjust the matching weights of each module using the memory bank according to the matching degree of the activated electrical signal sites in the memory bank in the subsequent action cycle.

[0029] like Figure 2 As shown in the figure, the application of the multi-module switching system based on the AI ​​model in the health care robot includes the following steps: Step S1. Determine the standard signal location according to the hardware parameters of the posture sensor in the health care robot, and record the actual electrical signal location of the sensor each time the sensor returns a signal, and generate a signal set in the database; Step S1 includes: Step S11. Format each sensor of the health care robot, and determine the standard signal site according to the hardware parameters of the sensor, wherein the hardware parameters of the sensor include: sensor piezoelectric efficiency, temperature and humidity sensing efficiency, and human body monitoring data conversion efficiency, so that the standard signal site of each sensor corresponds to the specific body parameters, behavior parameters, and posture parameters of the health care object; Step S12. When the sensor generates a corresponding electrical signal according to external physical changes, the level strength value of each electrical signal site is recorded in the database, and all recorded electrical signal site data are stored in a set form to form a signal set A, wherein A={T1, T2,…, Tn}, where n represents the number of records, and Tn represents the level strength of the electrical signal site in the nth record.

[0030] Step S2. Calculate the aggregation degree of the signal set, generate random Gaussian noise consistent with the discrete degree and aggregation degree, add the random Gaussian noise into the standard signal site, obtain a noise library, use the noise library to train the AI ​​model, and obtain an AI model with data cleaning capabilities; Step S2 includes: Step S21. Calculate the degree of aggregation of the signal set, where the degree of aggregation represents the average of the squares of the differences between the elements in the set and the standard signal sites, satisfying I=1 / n·[(T1-T0) 2 + (T2-T0) 2 +…+(Tn-T0) 2 ], where I is the degree of polymerization and T0 is the level intensity of the standard signal site; Step S22. Generate random Gaussian noise according to the degree of aggregation, so that the generated noise meets the following conditions: Where F(t) represents random Gaussian noise, t represents time, t0 represents the time when the sensor signal occurs, and e is the base of the natural logarithm; Step S23. Randomly change the value of t to obtain a Gaussian noise set, add each element in the random Gaussian noise set to the standard signal site, generate a noise library, and use the noise library to train the AI ​​model so that when the AI ​​model inputs the signal site, it automatically outputs the standard signal site with the closest level intensity, thereby realizing the cleaning process of the signal site. The AI ​​models include: Mask R-CNN model, DeepLab model, U-Net model and Watershed model.

[0031] Step S3. Collect all sensor data of the robot during the service cycle, spread each sensor data in the virtual space, obtain the actual signal model, calculate the mapping amount of the actual signal model in each service module space, and arrange each service module in descending order according to the mapping amount; Step S3 includes: Step S31. Acquire the electrical signals of each sensor of the robot during the service cycle and clean the electrical signals. There is only one electrical signal for the same sensor during the service cycle. When the same sensor transmits multiple electrical signals during the service cycle, end the current service cycle and allocate the excess electrical signals to the next service cycle. Step S32. Generate an m-dimensional virtual space in the control chip, and annotate the collected electrical signals in the form of vectors, where m represents the number of electrical signals in the service cycle, and obtain an actual signal model V, where V=(d1, d2, ..., dm), where dm represents the level strength of the electrical signal transmitted by the mth sensor; Step S33. Expand the robot module in the form of intervals and represent the robot module as {[D1+, D1-], [D2+, D2-], …, [Dn+, Dn-]}, where D1+, D2+ and Dn+ represent the upper bounds of the intervals of the electrical signal level strengths of the sensors numbered 1, 2 and n, respectively, and D1-, D2- and Dn- represent the lower bounds of the intervals of the electrical signal level strengths of the sensors numbered 1, 2 and n, respectively. Calculate the mapping vector of the actual signal model V in the n-dimensional space, record the module length of the portion of the interval where the mapping vector falls in each robot module as the mapping amount of the actual signal model in each module space, and arrange the modules in descending order of the mapping amount.

[0032] Step S4. Switch the working module of the robot to the service module ranked first, use the subsequent electrical signal sites as key values, add up the output actions of each module according to the mapping amount and weight, obtain the value of each action, activate the action with the highest output value of the robot, and end the current cycle after the robot performs the output action for a fixed time; Step S4 includes: Step S41. Switch the robot's working module to the first module with the highest mapping amount, and after receiving the subsequent signal, calculate the value of each action according to the following formula: Among them, E represents the value of the action, k represents the number of robot service modules, and u i represents the weight of the i-th service module, p i represents the probability of the action being called in the i-th service module, d i Represents the influence parameter of the subsequent signal on the actual signal model, which is determined by the rules of the AI ​​model, r i Represents the mapping amount of the actual signal model in the i-th service module; Step S42: Select the action with the highest value as the output action of the current service cycle, activate the robot to perform the output action, and end the service cycle after the robot performs the output action for a fixed time.

[0033] Step S5. Record the electrical signal sites and actual action outputs of all sensors within the cycle to form a memory bank, and adjust the weight of each action in the subsequent service cycle according to the sensor information uploaded within the cycle.

[0034] Step S5 includes: Step S51. Record the electrical signal sites and robot actions of each sensor during the service period, and use the AI ​​model to perform convergence training on the electrical signal data and the robot output. When the convergence training result is positive feedback, store the electrical signal sites of the sensor and the robot action as a memory pattern in the memory bank; Step S52. In the subsequent action cycle, the matching degree of each electrical signal site in the actual signal model in the memory library is calculated, and the weight of the action recorded in the memory unit in each service module is adjusted according to the matching degree of the memory unit.

[0035] Embodiment: There are three sensors in the health care robot, which monitor the object's body temperature, body posture and humidity respectively. The robot also has three service modules, namely nursing, medical treatment and alarm. In one service cycle, the electrical signal sites of the three sensors are obtained, which are 2.1mV, 0.8mV and 1.4mV respectively. The mapping amounts in the three service modules are 0.21, 0.45 and 1.12 respectively. The robot is switched to the alarm state. When a subsequent signal is received in a subsequent service cycle, the robot action is executed according to the logic of the alarm state.

[0036] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0037] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art can still modify the technical solutions described in the aforementioned embodiments or replace some of the technical features therein by equivalents. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A multi-module switching system based on an AI model, characterized in that: The system includes the following modules: a posture sensing module, a signal cleaning module, a model approaching module, a reaction output module and a memory training module; The posture sensing module is used to pre-transmit standard signal sites in the robot's posture sensor, each type of electrical signal site corresponds to the specific physical parameters, behavioral parameters and posture parameters of the health care object, obtain historical classification results, generate a historical site set of each type of electrical signal, and calculate the aggregation degree of the set, which reflects the average intensity of environmental noise; The signal cleaning module is used to add random Gaussian noise with a distribution divergence equal to the degree of aggregation to the standard electrical signal sites according to the set aggregation degree, to obtain a random noise set, to use the random noise set as a noise library, and to train the AI ​​model with the standard electrical signal as the output, and to use the trained AI model to parse the electrical signal sites of the posture sensor and convert them into corresponding standard electrical signal outputs; The model approach module is used to collect all sensor data within one action cycle of the robot, simulate the multidimensional distribution of the electrical signal sites corresponding to all sensor data in the virtual space, perform the same operation with the multidimensional distribution model between the health care modules, obtain the mapping amount of the multidimensional distribution to each health care module, and arrange all modules in the order of approach; The reaction output module is used to generate a robot reaction library according to the module proximity and the corresponding output of each module, using each identified electrical signal site as a key value to match the corresponding output built in each health care module, and weightedly accumulate the output actions of each module according to the mapping amount and weight to obtain the value of each action, and activate the robot actions in sequence according to the value; The memory training module is used to perform convergence training on the robot reactions triggered by adjacent electrical signal sites to form a memory bank, and adjust the weights of actions in each module according to the matching degree of the activated electrical signal sites in the memory bank.

2. The multi-module switching system based on the AI ​​model according to claim 1, characterized in that: The posture sensing module includes: a signal location unit and an environment aggregation unit; The signal location unit is used to format the sensors of the health care robot, determine the standard signal location according to the sensor hardware parameters, and is managed uniformly by the main control chip; The environment aggregation unit is used to record the actual electrical signal position of the sensor each time the sensor returns a signal, generate a set linked list in the database, and calculate the aggregation degree of the set; The signal cleaning module includes: a random noise unit and an impurity removal training unit; The random noise unit is used to generate random Gaussian noise with a divergence equal to the degree of aggregation according to the degree of aggregation of elements in the historical site set; The impurity removal training unit is used to add random Gaussian noise into the standard signal site to generate training noise, and use the training noise to train the AI ​​classification model, so that the AI ​​classification model has data cleaning capabilities.

3. The multi-module switching system based on the AI ​​model according to claim 2, characterized in that: The model approach module includes: a spatial distribution unit, a multi-simulation unit and a model switching unit; The spatial distribution unit is used to spread the electrical signal sites of all sensors in the virtual space to form an actual signal model; The multi-simulation unit is used to simulate the service interval of each module of the robot in the virtual space, and calculate the proportion of the actual signal model located in the service interval as the approach output; The model switching unit arranges all modules in order of proximity and switches the service module of the robot to the first module.

4. The multi-module switching system based on the AI ​​model according to claim 3 is characterized in that: The reaction output module includes: a key value matching unit and a behavior activation unit; The key value matching unit is used to identify the subsequent electrical signal site, use the subsequent electrical signal site as the key value, accumulate the output actions of each module according to the mapping amount and weight, and match the corresponding output in the current service module; The behavior activation unit is used to activate the output action of the robot according to the matching result, and end the service cycle after the robot performs the output action for a fixed time.

5. The multi-module switching system based on the AI ​​model according to claim 4, characterized in that: The memory training module includes: a convergent classification unit, a memory library unit and a memory weighting unit; The convergence classification unit is used to perform convergence training on the electrical signal data and the robot output using the AI ​​model; The memory unit is used to record the electrical signal locations of each sensor and the robot movements during the service cycle in a database; The memory weighting unit is used to adjust the matching weights of each module using the memory bank according to the matching degree of the activated electrical signal sites in the memory bank in the subsequent action cycle.

6. The application of multi-module switching system based on AI model in health care robot is characterized by: The following steps are involved: Step S1. Determine the standard signal location according to the hardware parameters of the posture sensor in the health care robot, and record the actual electrical signal location of the sensor each time the sensor returns a signal, and generate a signal set in the database; Step S2. Calculate the aggregation degree of the signal set, generate random Gaussian noise consistent with the discrete degree and aggregation degree, add the random Gaussian noise into the standard signal site, obtain a noise library, use the noise library to train the AI ​​model, and obtain an AI model with data cleaning capabilities; Step S3. Collect all sensor data of the robot during the service cycle, spread each sensor data in the virtual space, obtain the actual signal model, calculate the mapping amount of the actual signal model in each service module space, and arrange each service module in descending order according to the mapping amount; Step S4. Switch the working module of the robot to the service module ranked first, use the subsequent electrical signal sites as key values, add up the output actions of each module according to the mapping amount and weight, obtain the value of each action, activate the action with the highest output value of the robot, and end the current cycle after the robot performs the output action for a fixed time; Step S5. Record the electrical signal sites and actual action outputs of all sensors within the cycle to form a memory bank, and adjust the weight of each action in the subsequent service cycle according to the sensor information uploaded within the cycle.

7. The application of the multi-module switching system based on the AI ​​model in a health care robot according to claim 6 is characterized in that: Step S1 includes: Step S11. Format each sensor of the health care robot, and determine the standard signal site according to the hardware parameters of the sensor, wherein the hardware parameters of the sensor include: sensor piezoelectric efficiency, temperature and humidity sensing efficiency, and human body monitoring data conversion efficiency, so that the standard signal site of each sensor corresponds to the specific body parameters, behavior parameters, and posture parameters of the health care object; Step S12. When the sensor generates a corresponding electrical signal according to external physical changes, the level strength value of each electrical signal site is recorded in the database, and all recorded electrical signal site data are stored in a set form to form a signal set A, wherein A={T1, T2,…, Tn}, where n represents the number of records, and Tn represents the level strength of the electrical signal site in the nth record.

8. The application of the multi-module switching system based on the AI ​​model in a health care robot according to claim 7 is characterized in that: Step S2 includes: Step S21. Calculate the degree of aggregation of the signal set, where the degree of aggregation represents the average of the squares of the differences between the elements in the set and the standard signal sites, satisfying I=1 / n·[(T1-T0) 2 + (T2-T0) 2 +…+(Tn-T0) 2 ], where I is the degree of polymerization and T0 is the level intensity of the standard signal site; Step S22. Generate random Gaussian noise according to the degree of aggregation, so that the generated noise meets the following conditions: Where F(t) represents random Gaussian noise, t represents time, t0 represents the time when the sensor signal occurs, and e is the base of the natural logarithm; Step S23. Randomly change the value of t to obtain a Gaussian noise set, add each element in the random Gaussian noise set to the standard signal site, generate a noise library, and use the noise library to train the AI ​​model so that when the AI ​​model inputs the signal site, it automatically outputs the standard signal site with the closest level intensity, thereby realizing the cleaning process of the signal site. The AI ​​models include: Mask R-CNN model, DeepLab model, U-Net model and Watershed model.

9. The application of the multi-module switching system based on the AI ​​model in a health care robot according to claim 8, characterized in that: Step S3 includes: Step S31. Acquire the electrical signals of each sensor of the robot during the service cycle and clean the electrical signals. There is only one electrical signal for the same sensor during the service cycle. When the same sensor transmits multiple electrical signals during the service cycle, end the current service cycle and allocate the excess electrical signals to the next service cycle. Step S32. Generate an m-dimensional virtual space in the control chip, and annotate the collected electrical signals in the form of vectors, where m represents the number of electrical signals in the service cycle, and obtain an actual signal model V, where V=(d1, d2, ..., dm), where dm represents the level strength of the electrical signal transmitted by the mth sensor; Step S33. Expand the robot module in the form of intervals and represent the robot module as {[D1+, D1-], [D2+, D2-], …, [Dn+, Dn-]}, where D1+, D2+ and Dn+ represent the upper bounds of the intervals of the electrical signal level strengths of the sensors numbered 1, 2 and n, respectively, and D1-, D2- and Dn- represent the lower bounds of the intervals of the electrical signal level strengths of the sensors numbered 1, 2 and n, respectively. Calculate the mapping vector of the actual signal model V in the n-dimensional space, record the module length of the portion of the interval where the mapping vector falls in each robot module as the mapping amount of the actual signal model in each module space, and arrange the modules in descending order of the mapping amount.

10. The application of the multi-module switching system based on the AI ​​model in a health care robot according to claim 9 is characterized in that: Step S4 includes: Step S41. Switch the robot's working module to the first module with the highest mapping amount, and after receiving the subsequent signal, calculate the value of each action according to the following formula: Among them, E represents the value of the action, k represents the number of robot service modules, and u i represents the weight of the i-th service module, p i represents the probability of the action being called in the i-th service module, d i Represents the influence parameter of the subsequent signal on the actual signal model, which is determined by the rules of the AI ​​model, r i Represents the mapping amount of the actual signal model in the i-th service module; Step S42: Select the action with the highest value as the output action of the current service cycle, activate the robot to perform the output action, and end the service cycle after the robot performs the output action for a fixed time; Step S5 includes: Step S51. Record the electrical signal sites and robot actions of each sensor during the service period, and use the AI ​​model to perform convergence training on the electrical signal data and the robot output. When the convergence training result is positive feedback, store the electrical signal sites of the sensor and the robot action as a memory pattern in the memory bank; Step S52. In the subsequent action cycle, the matching degree of each electrical signal site in the actual signal model in the memory library is calculated, and the weight of the action recorded in the memory unit in each service module is adjusted according to the matching degree of the memory unit.

Images