Personalized trajectory planning method based on coring dynamic motion primitives
By using a method based on nucleated dynamic motion primitives, real-time information on the healthy leg joints is collected to generate a reference trajectory for the affected leg. This solves the problems of high training consumption and poor dynamic adaptability in existing technologies, and achieves low consumption and high adaptability in personalized trajectory planning.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-12-22
- Publication Date
- 2026-04-03
AI Technical Summary
In existing technologies, personalized gait trajectory planning methods rely on mathematical models or neural networks, which suffer from high training costs and lack of dynamic adaptability, and cannot effectively adapt to the differences in movement characteristics among different patients.
A method based on kernelized dynamic motion primitives is adopted. By collecting joint information of the healthy leg in real time, extracting joint feature points, generating a reference trajectory of the affected leg, and combining Gaussian mixture model and kernel function, the trajectory is dynamically updated to adapt to individual differences.
It achieves personalized trajectory planning with low training consumption, and can generate personalized trajectories online to adapt to different wearers, thereby improving the motion adaptability of the lower limb rehabilitation exoskeleton robot.
Smart Images
Figure CN121774499A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of trajectory, and more specifically to a personalized trajectory planning method based on kernelized dynamic motion primitives. Background Technology
[0002] Lower limb exoskeleton robots, as advanced assistive rehabilitation devices, have significant application value in improving the walking function of patients with gait abnormalities caused by stroke, spinal cord injury, and aging. However, the significant differences in gait patterns among this patient group pose a major challenge to achieving personalized gait assistance with exoskeleton robots. Therefore, generating highly personalized motion trajectories based on each wearer's motion characteristics has become key to improving the adaptability and assistive effectiveness of exoskeleton systems. Based on this understanding, current research is increasingly focused on developing advanced personalized trajectory planning technologies to achieve a high degree of matching between the exoskeleton and the wearer's motion characteristics, thereby providing more natural and efficient walking assistance for patients with mobility impairments.
[0003] Currently, research on personalized gait trajectory planning can be broadly categorized into model-based methods and machine learning methods. Model-based personalized trajectory planning uses mathematical models to build trajectory generators. The control parameters of the trajectory generator are adjusted in real-time based on the wearer's actual movement to generate personalized trajectories for different movement states. However, this method relies on the accuracy of the established mathematical model of the trajectory generator and requires adjustment of the trajectory generation control parameters. It also simplifies the lower limbs into a two-dimensional rod model, failing to consider the gait characteristics of lower limb movement. Machine learning-based personalized trajectory planning, on the other hand, builds neural networks to establish a mapping relationship between human characteristic attributes (such as age, height, weight, and leg length) and movement attributes (such as stride length and gait speed) and lower limb joint movement angles, thereby obtaining personalized trajectories for different individuals. However, this method relies on databases; as the data dimensionality of the input model increases, training becomes costly. Furthermore, this method plans the entire gait cycle and lacks dynamic adaptability when gait changes continuously.
[0004] Therefore, to address the problem of poor adaptability in trajectory planning, a personalized trajectory planning method based on kernelized dynamic motion primitives is needed. This method has the advantage of requiring less training and can take into account both trajectory flexibility and human motion characteristics. Summary of the Invention
[0005] In view of this, the purpose of this invention is to overcome the shortcomings of the prior art and provide a personalized trajectory planning method based on kernelized dynamic motion primitives, which has the advantage of low training consumption and can take into account both trajectory flexibility and human motion characteristics.
[0006] The personalized trajectory planning method based on kernelized dynamic motion primitives of the present invention includes:
[0007] Collect joint movement information of the healthy leg and the affected leg during the patient's walking process;
[0008] Extract joint feature point information corresponding to joint motion information;
[0009] The joint feature point information is processed to obtain the processed joint feature point information;
[0010] Based on nucleated motion primitives, the reference trajectory of the affected leg is generated using the processed joint feature point information.
[0011] Furthermore, the joint motion information includes knee joint characteristic events and hip joint characteristic events;
[0012] The knee joint characteristic events include heel strike, maximum joint angle during the support phase, minimum joint angle during the support phase, maximum shutdown angular velocity during the swing phase, maximum joint angle during the swing phase, and minimum joint angle during the swing phase.
[0013] The hip joint characteristic events include heel strike, minimum joint angular velocity during the stance phase, minimum joint angle during the stance phase, maximum angular velocity during the swing phase, and maximum joint angle during the swing phase.
[0014] Furthermore, the joint feature point information includes joint angles and joint angular velocities.
[0015] Furthermore, the joint feature point information is processed, specifically including:
[0016] Will The joint angles of the healthy leg at different time points are stored as follows: Joint angular velocity stored as The joint angle corresponding to the affected leg is stored as... ;
[0017] The initial angular asymmetry factor is determined according to the following formula: The amplitude asymmetry factor is :
[0018] ;
[0019] ;
[0020] in, , These are the initial joint angles at the start of the gait phase for the affected leg and the unaffected leg, respectively. , These are the joint angles at the largest and smallest joint characteristic points of the affected leg, respectively. , These are the joint angles at the largest and smallest joint feature points of the healthy leg, respectively.
[0021] The joint angle corresponding to the healthy leg and joint angular velocity Processing:
[0022] ;
[0023] ;
[0024] in, and These represent the processed joint angle and joint angular velocity, respectively.
[0025] Furthermore, based on the nucleated motion primitives, the processed joint feature point information is used to generate a reference trajectory for the affected leg, specifically including:
[0026] Constructing the parametric trajectory model of the joint :
[0027] ;
[0028] in, For joint angle; Joint angular velocity; for vidiquities; for The first differential; This is the weight vector; ; The dimension of the output vector;
[0029] Assuming weight vector , where the mean With variance Unknown, the parametric trajectory model satisfies: ;
[0030] To match the parametric trajectory model with the reference trajectory model, the unknown parameters are solved by minimizing the KL divergence between the parametric trajectory model and the reference trajectory model. The objective function is:
[0031] ;
[0032] in, , ; This represents the mean; Indicates variance;
[0033] By minimizing the objective function, the unknown parameters in the parametric trajectory model are obtained. and Corresponding solution and for:
[0034] ; ;
[0035] In the formula, ; , ;
[0036] It is a penalty coefficient introduced to prevent overfitting; Indicates diagonalization; The length of the joint trajectory;
[0037] Given a new time input Then the mean value of the joint parameter trajectory is:
[0038] ;
[0039] By defining a kernel function, we avoid calculating the inner product of high-dimensional vectors. The kernel function is:
[0040] ;
[0041] in, Representing the parametric trajectory model basis functions in Indicates will Substitute the value into the basis function, This represents the i-th value in the normalized time series. This represents the j-th value in the normalized time series. It is a constant;
[0042] The kernel matrix is:
[0043] ;
[0044] In the formula, ;
[0045] ;
[0046] ;
[0047] ;
[0048] Therefore, we have: ;
[0049] In the formula, ;
[0050] ;
[0051] To make the trajectory pass through the desired point , Let M be the m-th time point in the time series, where M is the maximum number of time points. ; For the reference trajectory model To modulate, let:
[0052] ; ;in, It is a constant;
[0053] A new time series Input into the reference trajectory model to generate joint trajectories. The trajectory can pass through the desired point.
[0054] Furthermore, a reference trajectory model for the joint is constructed, specifically including:
[0055] Historical data of multiple subjects at different speeds and different time states were obtained, and the original trajectories formed by the historical data were mapped to time. The interval is used to obtain the teaching database. for:
[0056] ;
[0057] in, This is the normalized time series; It is a two-dimensional vector representing the joint flexion / extension angle and angular velocity; These are the trajectory points of the joint trajectory; The length of the joint trajectory; Number the joint trajectory; The maximum number of the joint trajectory;
[0058] Based on teaching database The following joint probability distribution is established using the Gaussian mixture model algorithm:
[0059] ;
[0060] in, belong ; belong ; Let represent the prior probability of the c-th Gaussian component; Let be the mean of the c-th Gaussian component; Let be the variance of the c-th Gaussian component;
[0061] The reference trajectory model is obtained through Gaussian mixture regression. ;
[0062] in, ; Represent a conditional probability distribution. This represents the nth time point in the normalized time series of the reference trajectory model. This indicates its corresponding general distribution; Let this be a Gaussian distribution with a mean of 1 / 2. The variance is .
[0063] Furthermore, it also includes: dynamically updating the reference trajectory of the affected leg.
[0064] By utilizing joint feature points extracted from the healthy side joint during real-time movement, the reference trajectory model is adjusted in real time, so that the planned trajectory passes through the desired point while the trajectory shape conforms to the movement pattern of the lower limbs of a healthy human.
[0065] The beneficial effects of this invention are as follows: This invention discloses a personalized trajectory planning method based on nucleated dynamic motion primitives. It collects real-time information on the hip and knee joint angles, angular velocities, and gait phases of the healthy and affected legs during a patient's walking process. Then, it extracts feature points of the healthy leg's joint trajectory, considering the asymmetry of the joint trajectories between the left and right legs, and adjusts the extracted feature points based on the historical gait data of the affected leg. A feature point storage area is set up to store feature points; each time a new feature point is extracted, the feature point cache is updated. The feature point storage area is used as model input to generate a reference trajectory for the affected leg. During the patient's walking process, as the feature points of the healthy leg are dynamically extracted and updated, the nucleated dynamic motion primitives continuously generate the reference trajectory for the affected leg. This invention can generate personalized trajectories adapted to different wearers, thus providing technical support for improving the motion adaptability of lower limb rehabilitation exoskeleton robots to different wearers. Attached Figure Description
[0066] The present invention will be further described below with reference to the accompanying drawings and embodiments:
[0067] Figure 1 This is a schematic diagram of the personalized trajectory planning method of the present invention;
[0068] Figure 2 This is a schematic diagram illustrating the selection of lower limb joint feature events according to the present invention;
[0069] Figure 3 This is a schematic diagram of the joint reference trajectory of the present invention. Detailed Implementation
[0070] The present invention will be further described below with reference to the accompanying drawings, as shown in the figures:
[0071] This embodiment discloses a personalized trajectory planning method based on kernelized dynamic motion primitives, including the following steps:
[0072] Collect joint movement information of the healthy leg and the affected leg during the patient's walking process;
[0073] Extract joint feature point information corresponding to joint motion information;
[0074] The joint feature point information is processed to obtain the processed joint feature point information;
[0075] Based on nucleated motion primitives, the reference trajectory of the affected leg is generated using the processed joint feature point information.
[0076] In this embodiment, the joint motion information includes knee joint characteristic events and hip joint characteristic events;
[0077] The knee joint characteristic events include heel strike, reaching the maximum joint angle during the stance phase, reaching the minimum joint angle during the stance phase, reaching the maximum shut-off angular velocity during the swing phase, reaching the maximum joint angle during the swing phase, and reaching the minimum joint angle during the swing phase; respectively corresponding to Figure 2 FK1-FK5 in the middle;
[0078] The hip joint characteristic events include heel strike, minimum joint angular velocity during the stance phase, minimum joint angle during the stance phase, maximum angular velocity during the swing phase, and maximum joint angle during the swing phase; respectively corresponding to Figure 2 FH1-FH5 in the series.
[0079] The joint feature point information includes joint angles and joint angular velocities.
[0080] Since hemiplegic patients primarily express their motor intentions through their unaffected leg (e.g., controlling walking speed, stride length, and cadence), their gait information can reflect their true motor intentions. This invention maps the gait information of the unaffected joint to the affected joint, using the movement state of the unaffected side to reflect the movement state of the affected leg. Specifically, by real-time detection of key feature events in the movement of the unaffected joint, the corresponding angle and angular velocity data are extracted and used as input for generating the trajectory of the affected joint.
[0081] During the patient's walking process, a joint feature point storage area is established to dynamically record the angle and angular velocity data of the healthy joint corresponding to the occurrence of joint feature events. A moving window can be used to dynamically update the storage area. The window width is 1. Whenever a joint feature point is detected, it is updated to the feature point storage area corresponding to the moving window, and the window is moved to the next position. When the moving window is at the end of the feature point storage area, the moving window returns to the initial position of the feature point storage area.
[0082] In this embodiment, considering the asymmetrical gait of hemiplegic patients during natural walking, which causes the joint angle amplitude of the healthy leg and the affected leg to differ from the initial angle during the gait cycle, this invention combines the movement trajectory of the affected leg in the previous gait cycle and processes the feature points from the healthy joint feature point storage area to make the healthy joint feature points conform to the movement law of the affected joint.
[0083] The joint feature point information is processed. Taking the processing of knee joint feature points as an example, the processing method for hip joint feature points is the same, specifically including:
[0084] Will The joint angles of the healthy leg at different time points are stored as follows: Joint angular velocity stored as The joint angle corresponding to the affected leg is stored as... ;in, It can take the value 6;
[0085] To compensate for bilateral asymmetry, the initial angular asymmetry factor is determined according to the following formula: The amplitude asymmetry factor is :
[0086] ;
[0087] ;
[0088] in, , These are the initial joint angles at the start of the gait phase for the affected leg and the unaffected leg, respectively. , These are the joint angles at the largest and smallest joint characteristic points of the affected leg, respectively. , These are the joint angles of the largest and smallest joint feature points of the healthy leg, respectively; the largest or smallest value is the largest or smallest among several values in the storage area.
[0089] The joint angle corresponding to the healthy leg and joint angular velocity Processing:
[0090] ;
[0091] ;
[0092] in, and These represent the processed joint angle and joint angular velocity, respectively.
[0093] The processed feature points can then be used to update the feature point storage area. .
[0094] In this embodiment, a reference trajectory model of the joint can be pre-constructed, specifically including:
[0095] Based on publicly available data and literature, historical data of multiple subjects at different speeds and different time states were obtained, and the original trajectories formed by the historical data were mapped to time. The teaching database is obtained from the interval (in seconds). for:
[0096] ;
[0097] in, This is the normalized time series; It is a two-dimensional vector representing the joint flexion / extension angle and angular velocity; These are the trajectory points of the joint trajectory, where n is an integer from 1 to 100; This is the length of the joint trajectory, with a value of 100. Number the joint trajectory; Let n be the maximum number of the joint trajectory, with a value of 10; then for n = 2 and h = 3... This is represented as 0.02 seconds when the trajectory number is 3.
[0098] Based on teaching database The following joint probability distribution is established using the Gaussian mixture model algorithm:
[0099] ;
[0100] in, belong ; belong ; Let represent the prior probability of the c-th Gaussian component; Let be the mean of the c-th Gaussian component; Let be the variance of the c-th Gaussian component;
[0101] for example, , ; This represents the component related to t in the mean of the c-th Gaussian component, where t is a 1-dimensional column vector. It is a 2-dimensional column vector. It's a 1+2=3 dimensional column vector. Corresponding to The first dimension, Corresponding to 2-3 dimensions; This represents the covariance between t and t'. Indicate t and covariance, express covariance with t, express and The covariance.
[0102] The reference trajectory model is obtained through Gaussian mixture regression. ;
[0103] in, ; Represent a conditional probability distribution. This represents the nth time point in the normalized time series of the reference trajectory model. This indicates its corresponding general distribution; Let this be a Gaussian distribution with a mean of 1 / 2. The variance is .
[0104] like Figure 3 (a) and Figure 3 (b) shows the Gaussian distribution mean (solid line) and standard deviation (shaded) of the reference trajectories of the knee and hip joints, respectively.
[0105] In this embodiment, generating the reference trajectory of the affected leg specifically includes:
[0106] Constructing the parametric trajectory model of the joint :
[0107] ;
[0108] in, For joint angle; Joint angular velocity; for vidiquities; for The first differential; This is the weight vector; ; This is the dimension of the output vector, which is 1.
[0109] Assuming weight vector , where the mean With variance Unknown, the parametric trajectory model satisfies: ;
[0110] To match the parametric trajectory model with the reference trajectory model, the KL divergence between the parametric trajectory model and the reference trajectory model is minimized for the unknown parameters (mean). With variance The unknown is solved, and its objective function is: ;
[0111] in, , ; This represents the mean; Indicates variance;
[0112] By minimizing the objective function, the unknown parameters in the parametric trajectory model are obtained. and Corresponding solution and for:
[0113] ; ;
[0114] In the formula, ; ,
[0115] ; ;
[0116] It is a penalty coefficient introduced to prevent overfitting; Indicates diagonalization; The length of the joint trajectory;
[0117] Given a new time input This time also belongs to An interval, distinct from the time values mentioned above, The value can be 0.15s.
[0118] time The mean value of the corresponding joint parameter trajectory is:
[0119] ;
[0120] By defining a kernel function, we avoid calculating the inner product of high-dimensional vectors. The kernel function is:
[0121] ;
[0122] in, Representing the parametric trajectory model basis functions in; Indicates will Substitute the value into the basis function, This represents the i-th value in the normalized time series, where i can take values from 1 to 100; This represents the j-th value in the normalized time series, where j can take values from 1 to 100; It is a constant, and can be taken as 24;
[0123] The kernel matrix is:
[0124] ;
[0125] In the formula, ;
[0126] ;
[0127] ;
[0128] ;
[0129] Therefore, we have: ;
[0130] In the formula, ;
[0131] ;
[0132] To make the trajectory pass through the desired point , This represents the m-th time point in the time series; M is the maximum number of time points, which can be selected based on the actual operating conditions. ; For the reference trajectory model To modulate, let:
[0133] ; ;in, It is a constant and can take the value 10. -3 ;
[0134] A new time series Input into the reference trajectory model to generate joint trajectories. The trajectory can pass through the desired point.
[0135] In this embodiment, online trajectory adjustment can dynamically adjust the gait trajectory based on the wearer's real-time movement status, further adapting to individual differences and changes during movement. The invention also includes: dynamically updating the reference trajectory of the affected leg.
[0136] By utilizing joint feature points extracted from the healthy side joint during real-time movement, the reference trajectory model is adjusted in real time, so that the planned trajectory passes through the desired point while the trajectory shape conforms to the movement pattern of the lower limbs of a healthy human.
[0137] Specifically: Assuming feature points are extracted from the healthy joint at the current moment. After processing, the feature point storage area is Taking the knee joint as an example, the knee joint reference trajectory model is as follows: The average knee joint reference trajectory is The mean feature points of the knee joint reference trajectory were extracted to obtain the indexes of 6 feature points of the knee joint reference trajectory. Update the knee joint reference trajectory;
[0138] because The mean is set to the feature point angle and angular velocity values, and the variance is a small value: ; ;
[0139] In the formula, To find a minimum value, take For the updated reference trajectory, the knee joint trajectory passing through the feature point can be obtained as follows: During human walking, the feature point storage area is dynamically updated, and the planned trajectory is continuously adjusted online in combination with the joint coordinated movement characteristics.
[0140] Since the new time series is 0~1s, To normalize the time of the trajectory, it needs to be scaled to meet the time requirements of the lower limb joint trajectories at different speeds. Therefore, the scaling factor is determined by the time of the feature points extracted from the healthy side joint, and the trajectory is scaled accordingly.
[0141] Assuming the appearance time of the feature points extracted from the healthy joint is The time series of knee joint trajectory is as follows: Set the time scaling factor , Then we have:
[0142] ; .
[0143] Finally, the lower limb exoskeleton robot can track the reference trajectory planned by this invention through an impedance controller, assisting the affected leg in walking.
[0144] In a personalized motion trajectory generation test experiment, the personalized trajectory planning method of this invention showed that, under different subjects and different speed conditions, the mean root mean square error of hip and knee joint angle estimation was small, and the mean root mean square errors of hip and knee joint planned trajectories were respectively... and In exoskeleton wearing experiments, the personalized trajectory planning method of this invention exhibited less human-computer interaction force on the posterior thigh and posterior calf compared to the Direct Replication (DR) gait mapping algorithm. The experimental results verified that the trajectory planning method of this invention can generate personalized trajectories adapted to different wearers online.
[0145] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.
Claims
1. A personalized trajectory planning method based on kernelized dynamic motion primitives, characterized in that: include: Collect joint movement information of the healthy leg and the affected leg during the patient's walking process; Extract joint feature point information corresponding to joint motion information; The joint feature point information is processed to obtain the processed joint feature point information; Based on nucleated motion primitives, the reference trajectory of the affected leg is generated using the processed joint feature point information.
2. The personalized trajectory planning method based on kernelized dynamic motion primitives according to claim 1, characterized in that: The joint motion information includes knee joint characteristic events and hip joint characteristic events; The knee joint characteristic events include heel strike, maximum joint angle during the support phase, minimum joint angle during the support phase, maximum shutdown angular velocity during the swing phase, maximum joint angle during the swing phase, and minimum joint angle during the swing phase. The hip joint characteristic events include heel strike, minimum joint angular velocity during the stance phase, minimum joint angle during the stance phase, maximum angular velocity during the swing phase, and maximum joint angle during the swing phase.
3. The personalized trajectory planning method based on kernelized dynamic motion primitives according to claim 1, characterized in that: The joint feature point information includes joint angles and joint angular velocities.
4. The personalized trajectory planning method based on kernelized dynamic motion primitives according to claim 1, characterized in that: The joint feature point information is processed, specifically including: Will The joint angles of the healthy leg at different time points are stored as follows: Joint angular velocity stored as The joint angle corresponding to the affected leg is stored as... ; The initial angular asymmetry factor is determined according to the following formula: The amplitude asymmetry factor is : ； ； in, , These are the initial joint angles at the start of the gait phase for the affected leg and the unaffected leg, respectively. , These are the joint angles at the largest and smallest joint characteristic points of the affected leg, respectively. , These are the joint angles at the largest and smallest joint feature points of the healthy leg, respectively. The joint angle corresponding to the healthy leg and joint angular velocity Processing: ； ； in, and These represent the processed joint angle and joint angular velocity, respectively.
5. The personalized trajectory planning method based on kernelized dynamic motion primitives according to claim 1, characterized in that: Based on nucleated motion primitives, and utilizing processed joint feature point information, a reference trajectory for the affected leg is generated, specifically including: Constructing the parametric trajectory model of the joint : ； in, For joint angle; Joint angular velocity; for vidiquities; for The first derivative; This is the weight vector; ; The dimension of the output vector; Assuming weight vector , where the mean With variance Unknown, the parametric trajectory model satisfies: ; To match the parametric trajectory model with the reference trajectory model, the unknown parameters are solved by minimizing the KL divergence between the parametric trajectory model and the reference trajectory model. The objective function is: ； in, , ; This represents the mean; Indicates variance; By minimizing the objective function, the unknown parameters in the parametric trajectory model are obtained. and Corresponding solution and for: ； ； In the formula, ; , ; It is a penalty coefficient introduced to prevent overfitting; Indicates diagonalization; The length of the joint trajectory; Given a new time input Then the mean value of the joint parameter trajectory is: ； By defining a kernel function, we avoid calculating the inner product of high-dimensional vectors. The kernel function is: ； in, Representing the parametric trajectory model basis functions in Indicates will Substitute the value into the basis function, This represents the i-th value in the normalized time series. This represents the j-th value in the normalized time series. It is a constant; The kernel matrix is: ； In the formula, ; ； ； ； Therefore, we have: ; In the formula, ; ； To make the trajectory pass through the desired point , Let M be the m-th time point in the time series, where M is the maximum number of time points. ; For the reference trajectory model To modulate, let: ; ;in, It is a constant; A new time series Input into the reference trajectory model to generate joint trajectories. The trajectory can pass through the desired point.
6. The personalized trajectory planning method based on kernelized dynamic motion primitives according to claim 5, characterized in that: Constructing a reference trajectory model for the joints, specifically including: Historical data of multiple subjects at different speeds and different time states were obtained, and the original trajectories formed by the historical data were mapped to time. The interval is used to obtain the teaching database. for: ； in, This is the normalized time series; It is a two-dimensional vector representing the joint flexion / extension angle and angular velocity; These are the trajectory points of the joint trajectory; The length of the joint trajectory; Number the joint trajectory; The maximum number of the joint trajectory; Based on teaching database The following joint probability distribution is established using the Gaussian mixture model algorithm: ； in, belong ; belong ; Let represent the prior probability of the c-th Gaussian component; Let be the mean of the c-th Gaussian component; Let be the variance of the c-th Gaussian component; The reference trajectory model is obtained through Gaussian mixture regression. ; in, ; Represent a conditional probability distribution. This represents the nth time point in the normalized time series of the reference trajectory model. This indicates its corresponding general distribution; Let this be a Gaussian distribution with a mean of 1 / 2. The variance is .
7. The personalized trajectory planning method based on kernelized dynamic motion primitives according to claim 6, characterized in that: Also includes: The reference trajectory of the affected leg is dynamically updated: By using joint feature points extracted from the healthy side joint during real-time movement, the reference trajectory model is adjusted in real time, so that the planned trajectory passes through the desired point while the trajectory shape conforms to the movement pattern of the lower limbs of a healthy human.