Optimization method of walking parameter of exoskeleton for lower limb rehabilitation training
The method optimizes gait parameters for lower limb exoskeletons using Gaussian mixture modeling and multi-armed banded algorithms to address individual patient differences, ensuring efficient and robust rehabilitation training.
Patent Information
- Application Number
- JP2023079939
- Authority / Receiving Office
- JP · JP
- Patent Type
- Applications
- Current Assignee / Owner
- Priority Date
- 2023-04-23
- Filing Date
- 2023-05-15
- Publication Date
- 2024-11-05
- Estimated Expiration
- 2043-05-15
AI Technical Summary
Conventional lower limb rehabilitation exoskeletons fail to consider individual patient parameters, leading to suboptimal rehabilitation outcomes due to a lack of personalized gait optimization based on user preferences.
A method involving Gaussian mixture modeling and multi-armed banded algorithm to cluster and optimize gait parameters, predicting posterior probability of user satisfaction with low calculation cost and high robustness.
Enables personalized gait parameter optimization for lower limb exoskeletons, achieving fast convergence and high accuracy in meeting user needs.
Smart Images

Figure 2024156227000001_ABST
Abstract
Description
[Technical field]
[0001] The present invention relates to a method for optimizing the walking parameters of a skeleton in the technical field of rehabilitation robots, and in particular to a method for optimizing the walking parameters of an exoskeleton for lower limb rehabilitation training. [Background technology]
[0002] Only 3-10% of spinal cord injury or stroke patients can recover to walk 50m on their own even after 3 years of illness due to central nervous system damage. Long-term, standard and scientific rehabilitation exercise training can reconstruct the relevant nervous system functional areas and allow patients to regain their ability to move independently. China is rapidly entering an ultra-aging society, and rehabilitation facilities and basic rehabilitation services are needed for disabled and semi-disabled elderly people. Lower limb rehabilitation exoskeleton can provide systematic rehabilitation training to patients without relying on physical therapists, and gait design is one of the key technologies of lower limb rehabilitation exoskeleton. Gait is the trajectory that lower limb rehabilitation exoskeleton makes the wearer walk, and individual gait design for patients plays an important role in improving the efficiency of rehabilitation training and accelerating the recovery of lower limbs.
[0003] However, conventional technologies lack consideration of individual walking parameters. Since each patient has different parameters of leg and limb structure and individual differences in the degree of motor function disorder, ideal rehabilitation effects can only be achieved by adopting an effective walking optimization method for each patient. If walking is generated based only on the patient's physical characteristic parameters, it cannot fully reflect the user's preference for walking parameters, so it is necessary to complete the optimization of the user's walking parameters through feedback information of the user's walking preferences. Summary of the Invention [Problem to be solved by the invention]
[0004] In order to solve the problems described in the Background Art, the present invention proposes a method for optimizing the walking parameters of an exoskeleton for lower limb rehabilitation training. The walking of the exoskeleton is quantified, and the walking is clustered by a Gaussian mixture model based on the walking and the compatibility of the exoskeleton with walking, and the usefulness of the walking is calculated by a multi-arm banded algorithm, thereby predicting the posterior probability that the user's needs will be satisfied by the walking, and optimizing the walking parameters. The present invention can predict the walking parameters of a lower limb exoskeleton that will satisfy the user's needs, and has low computational cost, fast convergence speed, and high robustness. [Means for solving the problem]
[0005] In order to achieve the above object, the technical solution of the present invention is as follows:
[0006] Step 1) sets the initial walking of the lower limb rehabilitation exoskeleton, and after the user wears the lower limb rehabilitation exoskeleton and conducts a walking experiment, obtains the exoskeleton-walking compatibility with the current walking;
[0007] Step 2) clustering the gaits in the gait database that have been subjected to the gait experiment using a Gaussian mixture model based on the compatibility of the exoskeleton-gait corresponding to the gait that has been subjected to the current gait experiment, and obtaining gait clusters; 3) calculating the utility of the gait cluster and updating the utility of the corresponding gait in the gait database;
[0008] 4) sampling the gaits in the gait database based on the current utility of the gait to obtain a sampled gait set, and further selecting the gait with the highest utility from the gait set;
[0009] The method further comprises a step 5) in which the user wears the lower limb rehabilitation exoskeleton and conducts a walking experiment based on the walking with the highest current utility, and then obtains the compatibility of the exoskeleton with the current walking, and repeats steps 2) to 4) a planned number of times, and determines the walking parameters corresponding to the final walking as the optimal walking parameters. Specifically, step 2) is
[0010] Step 2.1) using the Bayesian information criterion to calculate the corresponding Bayesian information criterion operator before and after doubling the number of clusters in the current Gaussian mixture model, calculate the difference between the two, obtain the difference between the operators, and adjust based on the difference between the operators to obtain the final number of clusters in the Gaussian mixture model;
[0011] Step 2.2) clustering the gaits in the gait database for which the gait experiment was performed using the gait mixture Gaussian model based on the number of clusters of the current Gaussian mixture model and the compatibility of the exoskeleton-gait corresponding to the gait for which the current gait experiment was performed, and obtaining initial gait clusters;
[0012] and step 2.3) obtaining an updated gait cluster as a final gait cluster by adjusting the coverage area of the current gait cluster based on the minimum coverage area. In step 2.1), the step of adjusting based on the operator difference to obtain the final number of clusters of the Gaussian mixture model is specifically
[0013] The Bayesian information criterion operator after doubling is subtracted from the Bayesian information criterion operator before doubling to obtain a difference between the operators. If the difference between the operators is less than 0, the number of clusters in the current Gaussian mixture model is doubled as the number of clusters in the final Gaussian mixture model; otherwise, the number of clusters in the current Gaussian mixture model is not changed.
[0014] The formula for the Gaussian mixture model is as follows: JPEG2024156227000002.jpg10170
[0015] where G represents the walking mixture Gaussian model, (α K ,θ K ) (L,H,W,T,f) represents the K-th Gaussian distribution cluster in the gait mixture Gaussian model, L is the stride length, H is the stride height, W is the stride width, T is the gait cycle, f is the exoskeleton-gait compatibility, and α K is the probability of the Kth walking Gaussian distribution, and θ K represents the probability density function of the K-th walking Gaussian distribution, and K is the number of clusters in the Gaussian mixture model.
[0016] In step 2.3), for each initial walking cluster, the coverage area of each initial walking cluster is calculated by the following formula: JPEG2024156227000003.jpg23144 where Cov represents the coverage area of each initial walking cluster, and σ i represents the standard deviation of the i-th dimension of the gait cluster, and N represents the dimension of the gait parameter.
[0017] If the coverage area of the current walking cluster is smaller than the minimum coverage area, update the current walking cluster with the current minimum coverage area as the coverage area of the current walking cluster; otherwise, do not change the coverage area of the current walking cluster.
[0018] The formula for calculating the minimum coverage area is as follows: JPEG2024156227000004.jpg14155 where minTh is the minimum covered area, Th is the current minimum covered area, d is the convergence rate, and t is the number of iterations. The walking parameters include stride length, stride width, stride height, and walking cycle.
[0019] In step 3), the walking clusters are treated as arms of the multi-arm banded algorithm, and the utility corresponding to the walking clusters is calculated by the multi-arm banded algorithm.
[0020] In the multi-arm bandit algorithm in step 3), the formula for calculating the utility of the walking cluster is as follows: JPEG2024156227000005.jpg22170
[0021] JPEG2024156227000006.jpg71170 Effect of the Invention
[0022] The present invention has the following beneficial effects: The present invention can predict the walking parameters of a lower limb exoskeleton that meets the user's needs, with low computational cost, fast convergence speed, and high robustness. [Brief description of the drawings]
[0023] [Figure 1] FIG. 1 is a detailed flowchart of the method for optimizing gait parameters of an exoskeleton for lower limb rehabilitation training. [Diagram 2] FIG. 2 is a schematic diagram of stride length and step width parameters of walking. [Diagram 3] FIG. 3 is a schematic diagram of walking stride height and gait cycle parameters. [Figure 4] FIG. 4 is an overall flowchart of a method for optimizing gait parameters of an exoskeleton for lower limb rehabilitation training. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS
[0024] In order to clarify the objectives, technical solutions and advantages of the present invention, the present invention will be described in more detail with reference to the accompanying drawings and examples. It should be understood that the specific examples described herein are merely for the purpose of interpreting the present invention and do not limit the present invention. In addition, the technical features related to each embodiment of the present invention described below can be combined with each other as long as they are not mutually inconsistent. As shown in FIG. 1 and FIG. 4, the present invention comprises the following steps.
[0025] 1) The initial walking of the lower limb rehabilitation exoskeleton is set, and after the user wears the lower limb rehabilitation exoskeleton and performs a walking experiment, the compatibility of the exoskeleton with the current walking is obtained. In a specific implementation, the compatibility of the exoskeleton with the current walking is the user's subjective evaluation of the exoskeleton. However, the compatibility of the exoskeleton with the current walking is a numerical value.
[0026] In this example, there were a total of eight groups in the initial experimental walk, the results of which are shown in Table 1.
[0027] Table 1 shows the initial gait parameters. JPEG2024156227000007.jpg104170
[0028] 2) Based on the compatibility of the exoskeleton-gait corresponding to the gait in the current walking experiment, the gaits in the walking database in the walking experiment are clustered by a Gaussian mixture model to obtain a gait cluster. Note that the gait parameters corresponding to the gait of each group include step length, step width, step height, and walking cycle, as shown in Figures 2 and 3. Specifically, 2) is as follows:
[0029] 2.1) Use the Bayesian information criterion to calculate the corresponding Bayesian information criterion operators before and after doubling the number of clusters in the current Gaussian mixture model, calculate the difference between the two, obtain the difference between the operators, and make adjustments based on the difference between the operators to obtain the final number of clusters in the Gaussian mixture model. In 2.1), the final number of clusters in the Gaussian mixture model is obtained by adjusting the operator difference.
[0030] The operator difference is obtained by subtracting the Bayes Information Criterion operator after doubling from the Bayes Information Criterion operator before doubling. That is, ΔBIC = BIC K -BIC 2K , but BIC K represents the Bayesian information criterion operator before doubling, and BIC 2K represents the Bayesian Information Criterion operator after doubling. If the difference ΔBIC between the operators is smaller than 0, it indicates that the number of existing clusters is insufficient to accurately fit the model. In that case, the number of clusters in the final Gaussian mixture model is doubled to be the number of clusters in the current Gaussian mixture model. Otherwise, the number of clusters in the current Gaussian mixture model is not changed, i.e., kept at K.
[0031] In this embodiment, the calculation formulas of the Bayesian Information Criterion operator corresponding to before and after doubling the number of clusters of the current Gaussian mixture model are as follows: JPEG2024156227000008.jpg41170ΔBIC>0 means that the number of existing clusters is sufficient to accurately fit the model, and there is no need to increase the number of clusters.
[0032] 2.2) Based on the number of clusters in the current Gaussian mixture model and the compatibility of the exoskeleton-gait corresponding to the gait for which the current walking experiment was performed, the gaits for which the walking experiment was performed in the walking database are clustered using the gait Gaussian mixture model to obtain initial gait clusters.
[0033] The formula for the Gaussian mixture model is as follows: JPEG2024156227000009.jpg14170
[0034] JPEG2024156227000010.jpg74170
[0035] The parameters of the walking mixture Gaussian model are solved by the expectation maximization algorithm. JPEG2024156227000011.jpg78170
[0036] 2.3) Obtain the updated gait cluster as the final gait cluster by adjusting the coverage area of the current gait cluster based on the minimum coverage area.
[0037] For each initial walking cluster, the coverage area of each initial walking cluster is calculated by the following formula: The formula is as follows: JPEG2024156227000012.jpg36136 where Cov represents the coverage area of each initial walking cluster, and σ i represents the standard deviation of the i-th dimension of the gait cluster, and N represents the dimension of the gait parameter.
[0038] At the beginning of the optimization process of walking parameters, the number of sampling points is small, so the fitting of the model is not accurate. If the coverage area of the current walking cluster is smaller than the minimum coverage area, it may not be possible to explore the unknown area. At this time, the minimum coverage area of the Gaussian model should be limited, that is, the coverage area of the current walking cluster is updated with the current minimum coverage area; otherwise, the coverage area of the current walking cluster is not changed.
[0039] JPEG2024156227000013.jpg14129 where minTh is the minimum covered area, Th is the current minimum covered area, d is the convergence rate, and t is the number of iterations.
[0040] In this embodiment, the calculation formulas for the covered area and the minimum covered area of the walking cluster are as follows: JPEG2024156227000014.jpg23170The initial minimum coverage area Th is 1.25e-5 and the convergence rate is 0.8. JPEG2024156227000015.jpg18169
[0041] For Cov < minTh, correct the variance of the model. The minimum variance is log51e - 5 = 0.1, and the corrected variance is [0.1, 0.1, 0.1, 0.1, 46.4]. 3) Calculate the utility of the walking cluster and update the utility of the corresponding walking in the walking database.
[0042] Regard the walking cluster as each arm of the multi - armed bandit algorithm, and calculate the utility corresponding to the walking cluster by the multi - armed bandit algorithm. The calculation formula for the utility of the walking cluster is as follows: JPEG2024156227000016.jpg27169
[0043] JPEG2024156227000017.jpg70170
[0044] 4) Continuously sample the walking in the walking database, and exclude the walking that obviously does not meet the user's satisfaction and expectation by the pruning algorithm, so as to gradually narrow the search range and accelerate the progress of convergence. Sample the walking in the walking database based on the current utility of the walking to obtain a sampled walking set, and further select the walking with the highest utility from the walking set. The specific formula is as follows: JPEG2024156227000018.jpg54170
[0045] 5) Based on the walking with the highest current utility, after the user wears the lower - limb rehabilitation exoskeleton and conducts a walking experiment, obtain the fitness of the exoskeleton - walking in the current walking, and repeat 2) - 4) until the planned number of times. Take the walking parameters corresponding to the final walking as the optimal walking parameters. In this embodiment, the sampled walking set is shown in Table 2.
[0046] Table 2 is a table of the sampled walking set. JPEG2024156227000019.jpg99170
[0047] According to the calculation results, the walking of the second group is more suitable for the current user, so the next walking parameters are set to be a step length of 1.37 m, a step height of 0.12 m, a step width of 0.20 m, and a walking cycle of 1.08 s. The above specific examples are merely illustrative for explaining the principles and effects of the present invention, and are not intended to limit the present invention. Those skilled in the art may modify or change the above examples without departing from the spirit and scope of the present invention. Therefore, all equivalent modifications and changes made without departing from the spirit and technical ideas disclosed in the present invention are still included in the scope of the claims of the present invention.
Claims
1. A method for optimizing gait parameters of an exoskeleton for lower limb rehabilitation training, comprising: Step 1) of setting the initial walking of the lower limb rehabilitation exoskeleton, having the user wear the lower limb rehabilitation exoskeleton and conduct a walking experiment, and then obtaining the exoskeleton-walking compatibility for the current walking; 2) clustering the gaits in the gait database that have been subjected to the gait experiment using a Gaussian mixture model based on the compatibility of the exoskeleton-gait corresponding to the gait that has been subjected to the current gait experiment, and obtaining gait clusters; 3) calculating the utility of the gait cluster and updating the utility of the corresponding gait in the gait database; 4) sampling the gaits in the gait database based on the current utility of the gait to obtain a sampled gait set, and further selecting the gait with the highest utility from the gait set; and 5) a step in which the user wears the lower limb rehabilitation exoskeleton and performs a walking experiment based on the walking with the highest current utility, and then obtains the exoskeleton-walking compatibility for the current walking, and repeats steps 2) to 4) a planned number of times to determine the walking parameters corresponding to the final walking as the optimal walking parameters. A method for optimizing walking parameters of an exoskeleton for lower limb rehabilitation training, comprising:
2. Specifically, step 2) is Step 2.1) of calculating the Bayesian information criterion operators corresponding to the number of clusters of the current Gaussian mixture model before and after doubling the number of clusters by the Bayesian information criterion, performing subtraction on both to obtain the difference between the operators, and adjusting based on the difference between the operators to obtain the number of clusters of the final Gaussian mixture model; Step 2.2) clustering the gaits for which the walking experiment has been performed in the walking database using the walking Gaussian mixture model based on the number of clusters of the current Gaussian mixture model and the compatibility of the exoskeleton-walking corresponding to the gait for which the current walking experiment has been performed, and obtaining initial walking clusters; 2.3) obtaining an updated gait cluster as a final gait cluster by adjusting the coverage area of the current gait cluster based on the minimum coverage area; The method for optimizing gait parameters of an exoskeleton for lower limb rehabilitation training according to claim 1.
3. In step 2.1), the number of clusters in the final Gaussian mixture model is adjusted based on the operator difference to specifically: Obtain a difference between the Bayesian information criterion operator after doubling by subtracting the Bayesian information criterion operator after doubling from the Bayesian information criterion operator before doubling, and if the difference between the operators is less than 0, set the number of clusters in the final Gaussian mixture model to be twice the number of clusters in the current Gaussian mixture model; otherwise, do not change the number of clusters in the current Gaussian mixture model. The method for optimizing gait parameters of an exoskeleton for lower limb rehabilitation training according to claim 2.
4. The formula for the Gaussian mixture model is as follows: where G represents the walking mixture Gaussian model, (α K ,θ K ) (L,H,W,T,f) represents the K-th Gaussian distribution cluster in the gait mixture Gaussian model, L is the stride length, H is the stride height, W is the stride width, T is the gait cycle, f is the exoskeleton-gait compatibility, and α K is the probability of the Kth walking Gaussian distribution, and θ K represents the probability density function of the K-th walking Gaussian distribution, and K is the number of clusters in the Gaussian mixture model. The method for optimizing gait parameters of an exoskeleton for lower limb rehabilitation training according to claim 2.
5. In step 2.3), for each initial walking cluster, calculate the coverage area of each initial walking cluster by the following formula, which is as follows: where Cov represents the coverage area of each initial walking cluster, and σ i represents the standard deviation of the i-th dimension of the gait cluster, N represents the dimension of the gait parameters, If the coverage area of the current walking cluster is smaller than the minimum coverage area, update the current walking cluster with the current minimum coverage area as the coverage area of the current walking cluster; otherwise, do not change the coverage area of the current walking cluster. The method for optimizing gait parameters of an exoskeleton for lower limb rehabilitation training according to claim 2.
6. The formula for calculating the minimum coverage area is as follows: where minTh is the minimum covered area, Th is the current minimum covered area, d is the convergence rate, and t is the number of iterations. The method for optimizing gait parameters of an exoskeleton for lower limb rehabilitation training according to claim 5.
7. The walking parameters include a stride length, a stride width, a stride height, and a walking cycle. The method for optimizing gait parameters of an exoskeleton for lower limb rehabilitation training according to claim 1.
8. In step 3), the walking clusters are treated as arms of a multi-arm banded algorithm, and the utility corresponding to the walking clusters is calculated by the multi-arm banded algorithm. The method for optimizing gait parameters of an exoskeleton for lower limb rehabilitation training according to claim 1.
9. In the multi-arm bandit algorithm of step 3), the calculation formula for the utility of the walking cluster is as follows: The method for optimizing gait parameters of an exoskeleton for lower limb rehabilitation training according to claim 8.
