Moxibustion robot trajectory planning method and device based on tactile guidance, and moxibustion robot

Abstract

The present invention discloses a moxibustion robot trajectory planning method, device, and moxibustion robot based on tactile guidance. During the offline training phase, the moxibustion robot is controlled to perform tactile interactive operations on a predetermined moxibustion technique on a moxibustion area. The robot uses the control and trajectory data during the interactive operation to train a tactile guidance model for the predetermined moxibustion technique, learning and memorizing the acupuncture point locations and tactile observation signals corresponding to different moxibustion technique trajectories. During automatic moxibustion trajectory planning, the robot uses the tactile guidance model corresponding to the technique, based on the moxibustion acupuncture points and tactile observation signals at a given initial moment, to predict and fine-tune the moxibustion trajectory of the selected moxibustion technique in real time. This invention can accurately simulate moxibustion techniques, adjust moxibustion trajectories in real time, and ensure the safety of the moxibustion process.

Description

Technical Field

[0001] The present invention relates to the field of medical and health equipment, and in particular to a moxibustion robot trajectory planning method, system and equipment based on tactile guidance. Background Art

[0002] In the field of automated medical equipment, precise positioning and trajectory planning are crucial for performing complex medical operations. Previous vision-based technologies allowed robots to navigate in unknown environments, building environmental maps and positioning themselves in real time through sensor data. However, these methods mostly rely on visual sensors, which are easily affected by changes in ambient lighting and occlusions, and are not suitable for close-range tasks such as moxibustion. Moxibustion is a traditional Chinese medicine treatment method. In automated moxibustion equipment, precise trajectory planning is crucial for replicating the moxibustion practitioner's techniques and achieving therapeutic effects. However, existing moxibustion robot systems mostly rely on preset mechanical trajectories and lack adaptability to individual patient differences. Summary of the Invention

[0003] In response to the problems that the trajectory planning of existing moxibustion robots during automated moxibustion is not accurate enough, the moxibustion practitioner's techniques cannot be reproduced, and there is a lack of real-time safety monitoring and efficacy assurance, the present invention provides a moxibustion robot trajectory planning method, system and equipment based on tactile guidance, which can accurately simulate moxibustion techniques, adjust the moxibustion trajectory in real time, and ensure the safety of the moxibustion process.

[0004] In order to achieve the above technical objectives, the present invention adopts the following technical solutions:

[0005] A moxibustion robot trajectory planning method based on tactile guidance, comprising:

[0006] Step 1: The moxibustion robot is controlled to perform a tactile interactive operation of a predetermined moxibustion technique on the moxibustion area, and the following data during the interactive operation are recorded: the moxibustion robot's control input u, operating state x, moxibustion acupoint l, and its tactile observation signal h;

[0007] Step 2: Using the data recorded in step 1, training a tactile guidance model for a predetermined moxibustion technique; the tactile guidance model is used to predict the moxibustion acupoints and the state of the moxibustion robot at the current moment based on the control input and tactile observation signals at all previous moments;

[0008] Step 3: When a moxibustion technique is selected, the corresponding tactile guidance model is used to predict and fine-tune the moxibustion trajectory of the selected moxibustion technique in real time based on the moxibustion acupoints and tactile observation signals at a given initial moment.

[0009] Furthermore, the tactile observation signal includes the pressure, hardness and texture of the moxibustion acupoints.

[0010] Furthermore, the tactile observation signal also includes the temperature of the moxibustion acupoint;

[0011] During the moxibustion operation, when the temperature exceeds the safety threshold, reduce the moxibustion intensity or temporarily stop the moxibustion operation; when the temperature is lower than the effective threshold of the technique, increase the moxibustion intensity.

[0012] Furthermore, the operating state of the moxibustion robot includes the position and posture of the moxibustion head in the robot coordinate system.

[0013] Furthermore, the tactile guidance model is represented based on Bayesian, deep learning or factor graph, and trained using corresponding optimization methods.

[0014] Furthermore, the tactile guidance model is represented based on Bayesian representation; including:

[0015] First establish the tactile guidance problem:

[0016] p(x t ,l t |h 1:t ,u 1:t )(1)

[0017] Where, subscript t represents the corresponding time, subscript 1:t represents the set of all time periods from 1 to t; x t represents the operating state of the moxibustion robot at time t, l t represents the moxibustion point at time t, h 1:t represents the tactile observation signal at all times from 1 to t, u 1:t represents the control input of the moxibustion robot from time 1 to t; p(·|·) represents the conditional probability;

[0018] Then, according to the Bayesian formula, the tactile guidance model shown in formula (1) is converted into:

[0019] p(x t ,l t |h 1:t ,u 1:t )∝p(h t |x t ,l t ,h 1:t-1 ,u 1:t )p(x t ,l t |h 1:t-1 ,u 1:t )(2)

[0020] In the formula, ∝ represents a proportional relationship, h 1:t-1 represents the tactile observation signal at all times from 1 to t-1;

[0021] Based on the fact that tactile observation is only related to the moxibustion robot state and moxibustion acupoints at the same moment, formula (2) is re-expressed as:

[0022] p(x t ,l t |h 1:t ,u 1:t )∝p(h t |x t ,l t )p(x t ,l t |h 1:t-1 ,u 1:t )(3)

[0023] Among them, based on the fact that the state of the moxibustion robot at a certain moment is only related to the previous moment, that is, the state probability model of the moxibustion robot is a Markov process, the probability p(x t ,l t |h 1:t-1 ,u 1:t ) is expressed as:

[0024] p(x t ,l t |h 1:t-1 ,u 1:t )=∫p(x t |x t-1 ,u t )p(x t-1 ,l t-1 |h 1:t-1 ,u 1:t-1 )dx t-1 (4)

[0025] Combining equations (3) and (4), we can get the posterior probability distribution of the tactile guidance model:

[0026] p(x t ,l t |h 1:t ,u 1:t )=p(h t |x t ,l t )∫p(x t |x t-1 ,u t )p(x t-1 ,l t-1 |h 1:t-1 ,u 1:t-1 )dx t-1 (5)

[0027] Finally, the data recorded in step 1 is used to optimize the posterior probability distribution of the tactile guidance model shown in formula (5).

[0028] Furthermore, the Kalman filter is used to optimize the posterior probability distribution of the tactile guidance model shown in formula (5). The solution process is:

[0029] Initialization: Define the moxibustion robot state x1, moxibustion acupoint l1 and covariance matrix ∑1 at the initial time t=1; each moment corresponds to a moxibustion acupoint;

[0030] Prediction: Update time t = t + 1; use the state transition model x of the moxibustion robot t =f(x t-1 ,u t ), predict the moxibustion robot state x t The prior distribution p(x t |x t-1 ,u t ); Calculate the Jacobian matrix J of the state transition model t , then update the state covariance matrix ∑ t :

[0031]

[0032] Where f represents the state transition model, ε t is the noise covariance of the prediction process;

[0033] Update: Using tactile observation model Get the current state x t Tactile observation signals under And calculate the Jacobian matrix H of the tactile observation model t , get the Kalman gain K t , and then use the current tactile observation h t Update the state and covariance matrix of the moxibustion robot:

[0034]

[0035] ∑ t =(IK t H t )∑ t (11)

[0036] Where h represents the tactile observation model, ζ t is the noise covariance of tactile observation; x t represents the moxibustion robot state predicted by the state transition model, h t represents the actual collected tactile observation signal, represents the tactile observation signal obtained by the tactile observation model, represents the updated state of the moxibustion robot, and I represents the identity matrix;

[0037] Repeat the above prediction and update steps until the tactile interaction operation of the predetermined moxibustion technique on all acupoints is completed.

[0038] Furthermore, step 3 performs real-time prediction and fine-tuning of the moxibustion trajectory, specifically:

[0039] Step 3-1, given the initial state x1 and moxibustion point l1 of the moxibustion robot at time t=1;

[0040] Step 3-2, update time t=t+1; use the tactile guidance model according to the control input u from time 1 to time t 1:t and tactile observation signal h 1:t , predict the state of the moxibustion robot at time t and moxibustion points

[0041] Step 3-3, real-time acquisition of the moxibustion robot's prediction of moxibustion acupoints using the tactile guidance model Tactile observation signals collected from multiple points around the body are used to predict moxibustion acupoints using a tactile guidance model. The tactile observation signal is Select the tactile observation signals collected from all points The tactile observation signal with the greatest similarity is recorded as the current best tactile observation

[0042] Step 3-4: Control the moxibustion robot to maintain the current optimal tactile observation Corresponding status

[0043]

[0044] Where, Keeping the moxibustion robot in shape Control input;

[0045] Step 3-5: When the moxibustion of the current moxibustion point is completed, return to step 3-2 and continue the moxibustion operation on the next moxibustion point.

[0046] A moxibustion robot trajectory planning device based on tactile guidance, comprising:

[0047] The data recording module is used to record the following data during the tactile interaction operation of the moxibustion robot receiving control to perform a predetermined moxibustion technique on the moxibustion area: the control input u, the operating state x, the moxibustion acupoint l and the tactile observation signal h of the moxibustion robot;

[0048] The model training module is used to train a tactile guidance model for a predetermined moxibustion technique using the data recorded by the data recording module; the tactile guidance model is used to predict the moxibustion acupoints and the state of the moxibustion robot at the current moment based on the control input and tactile observation signals at all previous moments;

[0049] The tactile guidance module is composed of a trained tactile guidance model and is used to: when the corresponding moxibustion technique is selected, based on the moxibustion points and tactile observation signals at the given initial moment, perform real-time prediction and fine-tuning of the moxibustion trajectory of the selected moxibustion technique.

[0050] A moxibustion robot comprises a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor implements any one of the above-mentioned moxibustion robot trajectory planning methods based on tactile guidance.

[0051] Compared with the prior art, the present invention has the following technical effects:

[0052] The moxibustion robot trajectory planning method based on tactile guidance provided by the present invention is divided into two parts: offline training and online automatic moxibustion trajectory planning. In the offline training stage, the moxibustion robot is controlled to perform tactile interactive operations of predetermined moxibustion techniques on the moxibustion area, and uses the control and trajectory data during the interactive operation process to train the tactile guidance model of the predetermined moxibustion technique, learn and memorize the acupuncture point positions and tactile observation signals corresponding to different moxibustion technique trajectories; in the automatic moxibustion trajectory planning, the moxibustion robot predicts and fine-tunes the moxibustion trajectory of the selected moxibustion technique in real time based on the moxibustion acupuncture points and tactile observation signals at a given initial moment according to the tactile guidance model corresponding to the technique, and tracks the moxibustion trajectory at a certain operating speed; at the same time, the temperature feedback collected by the tactile sensor can be used to control the distance between the moxibustion and the human body surface, ensuring that the moxibustion process is both safe and effective. Therefore, the present invention can accurately simulate moxibustion techniques, adjust the moxibustion trajectory in real time, and ensure the safety of the moxibustion process. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 This is a flow chart of the trajectory planning method of the moxibustion robot based on tactile guidance in the present invention;

[0054] Figure 2 Schematic diagram of the moxibustion head with a tactile sensor of the present invention; reference numerals: 1-moxibustion placement area, 2-sensor, 3-communication port, 4-sensor protective shell, 5-cover, 6-pressing plate, 7-box body, 8-ventilation hole;

[0055] Figure 3 This is a diagram showing the trajectory predicted using the tactile guidance model in the present invention;

[0056] Figure 4This is a schematic diagram of temperature feedback in the present invention to ensure the safety and effectiveness of moxibustion. DETAILED DESCRIPTION

[0057] The following is a detailed description of an embodiment of the present invention. This embodiment is based on the technical solution of the present invention, provides a detailed implementation method and a specific operation process, and further explains the technical solution of the present invention.

[0058] Example 1

[0059] This embodiment provides a moxibustion robot trajectory planning method based on tactile guidance, referring to Figure 1 As shown, the following steps are included:

[0060] Step 1: The moxibustion robot is controlled to perform tactile interactive operations of predetermined moxibustion techniques on the moxibustion area, and records control and trajectory data during the interactive operations.

[0061] Step 1-1, start the moxibustion robot equipped with a tactile sensor. Under the guidance and control of the moxibustion therapist, the moxibustion robot performs a series of tactile interactive operations such as pressing and sliding on the moxibustion points of the human body according to the predetermined moxibustion techniques. The moxibustion head of the moxibustion robot performs tactile interactive operations, such as Figure 2 shown.

[0062] Step 1-2, record the following data during the tactile interaction operation: the control input u, operating state x, moxibustion acupoints and their tactile signals h of the moxibustion robot.

[0063] Among them, the control input of the moxibustion robot refers to the control input quantity of the driving motors of each joint of the moxibustion robot.

[0064] The operating state x of the moxibustion robot refers to the state variables such as the position and posture of the moxibustion head of the moxibustion robot in the robot coordinate system.

[0065] Different moxibustion techniques are formed by performing moxibustion operations on multiple moxibustion points in sequence (i.e., tactile interactive operations).

[0066] Tactile signals include tactile data such as pressure and tactile images. These are used as observation data and, after data preprocessing, are subsequently used to train the tactile guidance model. Tactile signals also include temperature data, which is used to control the intensity of moxibustion.

[0067] Step 2: Using the data recorded in step 1, train a tactile guidance model for a predetermined moxibustion technique; the tactile guidance model is used to predict the moxibustion points and the state of the moxibustion robot at the current moment based on the control input and tactile observation signals at all previous moments.

[0068] For a fixed moxibustion subject, during the tactile interaction process, there is a certain correspondence between the control input u, the operating state x, the moxibustion acupoints, and the tactile signal h. For example, when a new control input is input, the moxibustion robot will enter a new state, causing the moxibustion robot to interact with different acupoints, thereby generating tactile observation data for the corresponding acupoints. Training the tactile guidance model aims to understand the relationship between these data.

[0069] This embodiment represents the tactile guidance model based on Bayesian representation, including the following process:

[0070] Step 2-1: Express the tactile guidance model of moxibustion manipulation in a Bayesian form:

[0071] p(x t ,l t |h 1:t ,u 1:t )(1)

[0072] Where, subscript t represents the corresponding time, subscript 1:t represents the set of all time periods from 1 to t; x t represents the operating state of the moxibustion robot at time t, l t represents the moxibustion point at time t, h 1:t represents the tactile observation signal at all times from 1 to t, u 1:t represents the control input of the moxibustion robot from time 1 to t; p(·|·) represents the conditional probability.

[0073] Therefore, the purpose of formula (1) is to calculate the tactile observation h at all times from 1 to t. 1:t and control input u 1:t , calculate the current state x of the moxibustion robot t and acupuncture point locations t The joint posterior probability density function of .

[0074] Step 2-2, according to the Bayesian formula, the tactile guidance model shown in formula (1) is converted into:

[0075] p(x t ,l t |h 1:t ,u 1:t )∝p(h t |x t ,l t ,h 1:t-1 ,u 1:t )p(x t ,l t |h 1:t-1 ,u 1:t )(2)

[0076] In the formula, ∝ represents a proportional relationship, h 1:t-1 Represents the tactile observation signal at all times from 1 to t-1.

[0077] Step 2-3, since tactile observations are independent, the probability model is p(h t |x t ,l t ), that is, the tactile observation at the current moment is only related to the moxibustion robot state and moxibustion acupoints at the current moment. Formula (2) can be re-expressed as:

[0078] p(x t ,l t |h 1:t ,u 1:t )∝p(h t |x t ,l t )p(x t ,l t |h 1:t-1 ,u 1:t )(3)

[0079] Assume that the motion probability model of the moxibustion robot is a Markov process p(x t |x t-1 ,u t ), that is, the current state of the moxibustion robot is only related to the state at the previous moment, so the probability p(x t ,l t |h 1:t-1 ,u 1:t ) is expressed as:

[0080] p(x t ,l t |h 1:t-1 ,u 1:t )=∫p(x t |x t-1 ,u t )p(x t-1 ,l t-1 |h 1:t-1 ,u 1:t-1 )dx t-1 (4)

[0081] In steps 2-4, the posterior probability distribution of the tactile guidance model is obtained by combining equations (3) and (4):

[0082] p(x t ,l t |h 1:t ,u 1:t )=p(h t |x t ,l t)∫p(x t |x t-1 ,u t )p(x t-1 ,l t-1 |h 1:t-1 ,u 1:t-1 )dx t-1 (5)

[0083] The ultimate optimization goal of the above formula is to make the posterior probability p(x t ,l t |h 1:t ,u 1:t ) is maximized, that is, based on past experience p(x t-1 ,l t-1 |h 1:t-1 ,u 1:t-1 ), the robot’s prior knowledge of state transition p(x t |x t-1 ,u t ) and the probability model of tactile observation p(h t |x t ,l t ), find the most likely state of the current moxibustion robot and the most likely position of the corresponding moxibustion point.

[0084] Step 2-5, using the data recorded in step 1, optimize and solve the posterior probability distribution of the tactile guidance model shown in formula (5).

[0085] The optimal solution of the posterior probability in formula (5) can be obtained using a series of classical optimization methods applicable to Bayesian problems, such as Monte Carlo and Kalman filter, thereby simultaneously realizing the construction of human moxibustion area and the positioning of specific human acupuncture points.

[0086] This embodiment uses a Kalman filter to optimize and solve equation (5), and the process is:

[0087] Initialization: Define the moxibustion robot state x1, moxibustion acupoint l1 and covariance matrix ∑1 at the initial time t=1; each moment corresponds to a moxibustion acupoint.

[0088] Prediction: Update time t = t + 1; use the state transition model x of the moxibustion robot t =f(x t-1 ,u t ), predict the moxibustion robot state x t The prior distribution p(x t |x t-1 ,u t ); Calculate the Jacobian matrix J of the state transition model t , then update the state covariance matrix ∑ t :

[0089]

[0090] Where f represents the state transition model, ε t is the noise covariance of the forecast process.

[0091] The state transition model x used in the above prediction t =f(x t-1 ,u t ), is determined by the forward and reverse motion of the moxibustion robot. That is, at the previous moment, the robot’s state x t-1 At this moment, a new input u is given t , the robot will move to the new state x t , that is, the position and posture of the moxibustion robot have changed accordingly based on the new input. In the field of robot control, the state transition model is usually described as x t =Jx t-1 +Bu t +ω t ,ω t is Gaussian noise, which is related to the actual control system and specific structural parameters of the moxibustion robot. The present invention only uses the function to express it as x t =f(x t-1 ,u t ).

[0092] In the Bayesian expression of this embodiment, the state transition model x t =Jx t-1 +Bu t +ω t Considered as a Gaussian distribution, the predicted state is x t The prior distribution p(x t |x t-1 ,u t ).

[0093] Update: Using tactile observation model Update the current state x t Tactile observation signals under And calculate the Jacobian matrix H of the tactile observation model t , get the Kalman gain K t , and then use the current tactile observation h t Update the state and covariance matrix of the moxibustion robot:

[0094]

[0095] ∑ t =(IK t H t )∑t (11)

[0096] Where h represents the tactile observation model, ζ t is the noise covariance of tactile observation; x t represents the moxibustion robot state predicted by the state transition model, h t represents the actual collected tactile observation signal, represents the tactile observation signal obtained by the tactile observation model, represents the updated state of the moxibustion robot, and I represents the identity matrix;

[0097] Repeat the above prediction and update steps until the tactile interaction of the predetermined moxibustion technique on all acupoints is completed. After the training is completed, the probability model of formula (1) is obtained, which is the final problem to be solved, so that step 3 can be continued to collect a tactile observation based on the initial given state of the moxibustion robot and the corresponding moxibustion acupoints, and then start predicting the state that the robot should reach at the next moment and the corresponding moxibustion acupoints. Finally, all the moxibustion acupoints and corresponding robot states that the moxibustion technique passes through during the teaching process can be predicted in sequence.

[0098] Step 3: When a moxibustion technique is selected, the corresponding tactile guidance model is used to predict, fine-tune and execute the moxibustion trajectory of the selected moxibustion technique in real time based on the moxibustion acupoints and tactile observation signals at a given initial moment.

[0099] Step 3-1, select the moxibustion technique and give the moxibustion robot the initial state x1 and moxibustion point l1 at the initial time t=1.

[0100] Step 3-2, update time t=t+1; use the tactile guidance model according to the control input u from time 1 to time t 1:t and tactile observation signal h 1:t , predict the state of the moxibustion robot at time t and moxibustion points This will enable the planning of a complete set of moxibustion trajectories.

[0101] Step 3-3, since the distance distribution of acupuncture points in the moxibustion area of ​​different patients is generally slightly different, it is impossible to be exactly the same. These acupuncture points generally show similarities in tactile observation, so the current state can be fine-tuned with the help of the robot's real-time tactile observation. Therefore, this embodiment obtains the tactile observation collected during the current operation of the moxibustion robot in real time, and is committed to achieving consistency between the tactile observation and the tactile observation corresponding to the acupuncture point position predicted by the tactile guidance model. Assume that the acupuncture point position predicted by the tactile guidance at time t is The corresponding tactile observation is The tactile observations obtained by the moxibustion robot during its current operation are The similarity evaluation function is Sim, and the most consistent current tactile observation is:

[0102]

[0103] To find the tactile observation more quickly, the moxibustion robot will only search at multiple points near the current acupuncture point. The similarity evaluation function for tactile observations can use a variety of similarity evaluation methods for tactile signals, including distance metrics for time series tactile signals, bag-of-words metrics, and rotational invariance metrics for tactile image signals. Finally, a weighted sum of these signal metrics is taken.

[0104] Step 3-4: When the most consistent tactile observation is found, it is considered that the tactile observation is the most consistent with the tactile data generated by the corresponding moxibustion acupoint. Therefore, it can be considered that the best moxibustion position has been found, and the moxibustion operation is limited to the state of the moxibustion robot corresponding to the tactile observation:

[0105]

[0106] Where, Keeping the moxibustion robot in shape control input.

[0107] like Figure 3 As shown, trajectory A represents the trajectory of the selected moxibustion technique predicted by the tactile guidance model, and trajectory A' is the moxibustion trajectory after fine-tuning based on real tactile observation.

[0108] Step 3-5: When the moxibustion of the current moxibustion point is completed, return to step 3-2 and continue the moxibustion operation on the next moxibustion point.

[0109] During moxibustion, if the temperature exceeds the safety threshold, the moxibustion intensity is reduced or temporarily stopped to ensure the safety of the moxibustion process; if the temperature falls below the effective threshold, the moxibustion intensity is increased to ensure the effectiveness of the moxibustion technique. The moxibustion intensity is achieved by dynamically adjusting the distance between the moxibustion tip and the human body surface. Specifically, classic PID methods can be used to adjust the distance between the moxibustion tip and the human body surface based on the feedback temperature.

[0110] like Figure 4 As shown, B1 represents when the safety threshold is exceeded, move away to reduce the intensity of moxibustion or stop moxibustion; B2 represents the abnormal temperature position; B3 represents the starting position; B4 represents the running direction; B5 represents when the threshold is below the effective threshold, move closer to increase the intensity of moxibustion.

[0111] Example 2

[0112] This embodiment provides a moxibustion robot trajectory control device based on tactile guidance, comprising:

[0113] The data recording module is used to record the following data during the tactile interaction operation of the moxibustion robot receiving control to perform a predetermined moxibustion technique on the moxibustion area: the control input u, the operating state x, the moxibustion acupoint l and the tactile observation signal h of the moxibustion robot;

[0114] The model training module is used to train a tactile guidance model for a predetermined moxibustion technique using the data recorded by the data recording module; the tactile guidance model is used to predict the moxibustion acupoints and the state of the moxibustion robot at the current moment based on the control input and tactile observation signals at all previous moments;

[0115] The tactile guidance module is composed of a trained tactile guidance model and is used to: when the corresponding moxibustion technique is selected, based on the moxibustion points and tactile observation signals at the given initial moment, perform real-time prediction and fine-tuning of the moxibustion trajectory of the selected moxibustion technique.

[0116] The implementation methods of the above modules are the same as those described in Example 1, and will not be repeated in this embodiment.

[0117] Example 3

[0118] This embodiment provides a moxibustion robot, including a memory and a processor, wherein a computer program is stored in the memory. When the computer program is executed by the processor, the processor implements the moxibustion robot trajectory planning method based on tactile guidance described in Example 1.

[0119] The above embodiments are preferred embodiments of the present application. Ordinary technicians in this field can also make various changes or improvements on this basis. Without departing from the overall concept of the present application, these changes or improvements should fall within the scope of protection required by the present application.

Claims

1. A moxibustion robot trajectory planning method based on tactile guidance, characterized in that: include: Step 1: The moxibustion robot receives control to perform tactile interactive operations on the moxibustion area using predetermined moxibustion techniques, and records the following data during the interactive operation: control input of the moxibustion robot , operating status , moxibustion acupoints and its tactile observation signal ; Step 2: Using the data recorded in step 1, training a tactile guidance model for a predetermined moxibustion technique; the tactile guidance model is used to predict the moxibustion acupoints and the state of the moxibustion robot at the current moment based on the control input and tactile observation signals at all previous moments; Step 3: When a moxibustion technique is selected, the corresponding tactile guidance model is used to predict and fine-tune the moxibustion trajectory of the selected moxibustion technique in real time based on the given moxibustion acupoints and tactile observation signals at the initial moment; The tactile guidance model is represented based on Bayesian representation; including: First establish the tactile guidance problem: (1) In the formula, the subscript Indicates the corresponding time, subscript Indicates from 1 to The collection of all moments of represents the operating status of the moxibustion robot at time t, Indicates the moxibustion point at time t, represents the tactile observation signal at all times from 1 to t, represents the control input of the moxibustion robot from 1 to t at all times; represents conditional probability; Then, according to the Bayesian formula, the tactile guidance model shown in formula (1) is converted into: (2) Where, Indicates a proportional relationship, Indicates from 1 to Tactile observation signals at all times; Based on the fact that tactile observation is only related to the moxibustion robot state and moxibustion acupoints at the same moment, formula (2) is re-expressed as: (3) Among them, the state of the moxibustion robot at a certain moment is only related to the previous moment, that is, the state probability model of the moxibustion robot is a Markov process, and the probability in formula (3) is Expressed as: (4) Combining equations (3) and (4), we can get the posterior probability distribution of the tactile guidance model: (5) Finally, the data recorded in step 1 is used to optimize the posterior probability distribution of the tactile guidance model shown in equation (5); The Kalman filter is used to optimize the posterior probability distribution of the tactile guidance model shown in formula (5). The solution process is: Initialization: Define the state of the moxibustion robot at the initial time t=1 , moxibustion acupoints and the covariance matrix ; Each moment corresponds to a moxibustion point; Prediction: Update time ; Using the state transition model of the moxibustion robot , predict the moxibustion robot status Prior distribution of ; Calculate the Jacobian matrix of the state transition model , then update the state covariance matrix : (6) (7) Where, represents the state transition model, is the noise covariance of the prediction process; Update: Using tactile observation model Get the current state Tactile observation signals under , and calculate the Jacobian matrix of the tactile observation model , obtain the Kalman gain , and then use the current tactile observation Update the state and covariance matrix of the moxibustion robot: (8) (9) (10) (11) Where, represents the tactile observation model, is the noise covariance of tactile observations; represents the state of the moxibustion robot predicted by the state transition model, represents the actual collected tactile observation signal, represents the tactile observation signal obtained by the tactile observation model, Indicates the updated status of the moxibustion robot. represents the identity matrix; Repeat the above prediction and update steps until the tactile interaction operation of the predetermined moxibustion technique on all acupoints is completed.

2. The moxibustion robot trajectory planning method based on tactile guidance according to claim 1, characterized in that, The tactile observation signals include pressure, hardness, and texture of the moxibustion acupoints.

3. The moxibustion robot trajectory planning method based on tactile guidance according to claim 2, characterized in that, The tactile observation signal also includes the temperature of the moxibustion acupoint; During the moxibustion operation, when the temperature exceeds the safety threshold, reduce the moxibustion intensity or temporarily stop the moxibustion operation; when the temperature is lower than the effective threshold of the technique, increase the moxibustion intensity.

4. The moxibustion robot trajectory planning method based on tactile guidance according to claim 1, characterized in that: The operating status of the moxibustion robot, including the position and posture of the moxibustion head in the robot coordinate system.

5. The moxibustion robot trajectory planning method based on tactile guidance according to claim 1, characterized in that: Step 3: Real-time prediction and fine-tuning of the moxibustion trajectory, specifically: Step 3-1, give the moxibustion robot the initial time Status and moxibustion points ; Step 3-2, update time ; Using the tactile guidance model, according to the control input from time 1 to t and tactile observation signals , predict the state of the moxibustion robot at time t and moxibustion points ; Step 3-3, real-time acquisition of the moxibustion robot's prediction of moxibustion acupoints using the tactile guidance model Tactile observation signals collected from multiple points around the body are used to predict moxibustion acupoints using a tactile guidance model. The tactile observation signal is , select the tactile observation signals collected from all points The tactile observation signal with the greatest similarity is recorded as the current best tactile observation ; Step 3-4: Control the moxibustion robot to maintain the current optimal tactile observation Corresponding status : (12) Where, Keeping the moxibustion robot in shape Control input; Step 3-5: When the moxibustion of the current moxibustion point is completed, return to step 3-2 and continue the moxibustion operation on the next moxibustion point.

6. A moxibustion robot trajectory planning device based on tactile guidance, characterized in that: include: The data recording module is used to record the following data during the tactile interaction operation of the moxibustion robot receiving control to perform a predetermined moxibustion technique on the moxibustion area: the control input of the moxibustion robot , operating status , moxibustion acupoints and its tactile observation signal ; The model training module is used to train a tactile guidance model for a predetermined moxibustion technique using the data recorded by the data recording module; the tactile guidance model is used to predict the moxibustion acupoints and the state of the moxibustion robot at the current moment based on the control input and tactile observation signals at all previous moments; The tactile guidance module is composed of a trained tactile guidance model and is used to: when a corresponding moxibustion technique is selected, based on the given moxibustion acupoints and tactile observation signals at the initial moment, perform real-time prediction and fine-tune the moxibustion trajectory of the selected moxibustion technique; The tactile guidance model is represented based on Bayesian representation; including: First establish the tactile guidance problem: (1) In the formula, the subscript Indicates the corresponding time, subscript Indicates from 1 to The collection of all moments of represents the operating status of the moxibustion robot at time t, Indicates the moxibustion point at time t, represents the tactile observation signal at all times from 1 to t, represents the control input of the moxibustion robot from 1 to t at all times; represents conditional probability; Then, according to the Bayesian formula, the tactile guidance model shown in formula (1) is converted into: (2) Where, Indicates a proportional relationship, Indicates from 1 to Tactile observation signals at all times; Based on the fact that tactile observation is only related to the moxibustion robot state and moxibustion acupoints at the same moment, formula (2) is re-expressed as: (3) Among them, the state of the moxibustion robot at a certain moment is only related to the previous moment, that is, the state probability model of the moxibustion robot is a Markov process, and the probability in formula (3) is Expressed as: (4) Combining equations (3) and (4), we can get the posterior probability distribution of the tactile guidance model: (5) Finally, the data recorded in step 1 is used to optimize the posterior probability distribution of the tactile guidance model shown in equation (5); Specifically, the Kalman filter is used to optimize the posterior probability distribution of the tactile guidance model shown in formula (5). The solution process is: Initialization: Define the state of the moxibustion robot at the initial time t=1 , moxibustion acupoints and the covariance matrix ; Each moment corresponds to a moxibustion point; Prediction: Update time ; Using the state transition model of the moxibustion robot , predict the moxibustion robot status Prior distribution of ; Calculate the Jacobian matrix of the state transition model , then update the state covariance matrix : (6) (7) Where, represents the state transition model, is the noise covariance of the prediction process; Update: Using tactile observation model Get the current state Tactile observation signals under , and calculate the Jacobian matrix of the tactile observation model , obtain the Kalman gain , and then use the current tactile observation Update the state and covariance matrix of the moxibustion robot: (8) (9) (10) (11) Where, represents the tactile observation model, is the noise covariance of tactile observations; represents the state of the moxibustion robot predicted by the state transition model, represents the actual collected tactile observation signal, represents the tactile observation signal obtained by the tactile observation model, Indicates the updated status of the moxibustion robot. represents the identity matrix; Repeat the above prediction and update steps until the tactile interaction operation of the predetermined moxibustion technique on all acupoints is completed.

7. A moxibustion robot, comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the computer program is executed by the processor, the processor is caused to implement the method according to any one of claims 1 to 5.

Images