CN110193830B - Ankle joint gait prediction method based on RBF neural network - Google Patents
Ankle joint gait prediction method based on RBF neural network Download PDFInfo
- Publication number
- CN110193830B CN110193830B CN201910437364.9A CN201910437364A CN110193830B CN 110193830 B CN110193830 B CN 110193830B CN 201910437364 A CN201910437364 A CN 201910437364A CN 110193830 B CN110193830 B CN 110193830B
- Authority
- CN
- China
- Prior art keywords
- gait
- prediction
- neural network
- model
- gait prediction
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B25—HAND TOOLS; PORTABLE POWER-DRIVEN TOOLS; MANIPULATORS
- B25J—MANIPULATORS; CHAMBERS PROVIDED WITH MANIPULATION DEVICES
- B25J9/00—Programme-controlled manipulators
- B25J9/0006—Exoskeletons, i.e. resembling a human figure
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B25—HAND TOOLS; PORTABLE POWER-DRIVEN TOOLS; MANIPULATORS
- B25J—MANIPULATORS; CHAMBERS PROVIDED WITH MANIPULATION DEVICES
- B25J9/00—Programme-controlled manipulators
- B25J9/16—Programme controls
- B25J9/1602—Programme controls characterised by the control system, structure, architecture
- B25J9/161—Hardware, e.g. neural networks, fuzzy logic, interfaces, processor
Abstract
The invention discloses an ankle joint gait prediction method based on a RBF neural network, and belongs to the field of human body gait motion prediction. The method comprises the following operation steps: 1) And based on the RBF neural network, the parameters of the lower limb model are quickly approximated. 2) And a gait prediction algorithm is designed, and a robust item is added into the algorithm, so that the robustness of the system is improved. 3) Based on the optical motion measurement system, the lower limb gait capture is designed and realized, and an experimental foundation is laid for gait prediction. 4) And analyzing the gait prediction result, and evaluating the performance of the gait prediction method. The method has innovativeness and simulation basis, can overcome the defects that the traditional gait prediction method depends on an accurate musculoskeletal model, needs a large amount of gait data to train the gait model and the like, has stronger robustness and higher prediction precision, and has great guiding significance for realizing the follow-up control of the lower limb exoskeleton.
Description
Technical Field
The invention relates to an ankle joint gait prediction method based on a RBF neural network, which is applied to the field of human-computer cooperative motion control of lower extremity exoskeleton.
Background
The lower limb exoskeleton is a wearable man-machine integrated mechanical device, can reconstruct and enhance the lower limb movement capability of a wearer, and has wide application prospect in the fields of helping the old and the disabled, medical rehabilitation, military disaster relief and the like. The follow-up control of the lower extremity exoskeleton is the key to realizing the auxiliary function of the exoskeleton. In the traditional human-computer cooperative control of the lower limb exoskeleton, physical sensors such as angle/angular velocity, force/moment and the like are utilized to acquire human kinematics or dynamics data in real time to serve as input of an exoskeleton control system. This approach typically results in the exoskeleton's movements lagging behind those of the wearer due to the time delay in information transfer. In addition, some external skeleton control systems use a sensing sensor to measure a human biological signal such as an electroencephalogram (EEG) signal or an Electromyogram (EMG) signal, and the like, so that although the biological signal is advanced with respect to the motion of the human body and the motion lag can be solved, the EEG and EMG signal are unstable and are easily interfered by static electricity, sweat, and the like.
Human lower limb gait prediction is used as an alternative method, and potential possibility is provided for man-machine cooperative motion control of the lower limb exoskeleton. Traditional gait prediction methods rely on accurate musculoskeletal models or require extensive gait data to train the gait model. The RBF neural network-based ankle joint gait prediction method adopts the RBF neural network to predict the lower limb model, and compared with the traditional method, the RBF neural network-based ankle joint gait prediction method can quickly approach the parameters of the lower limb model.
Disclosure of Invention
The invention aims to overcome the defects in the prior art and provides an ankle joint gait prediction method based on a RBF neural network. The method has the advantages of high convergence speed, strong robustness and high prediction precision.
In order to achieve the purpose, the invention adopts the following technical scheme:
an ankle joint gait prediction method based on an RBF neural network comprises the following operation steps:
1) Based on the RBF neural network, realizing the rapid approximation of the parameters of the ankle joint model;
2) A gait prediction algorithm is designed, and a robust item is added into the algorithm, so that the robustness of the system is improved;
3) Based on an optical motion measurement system, ankle joint gait capture is designed and realized, and an experimental foundation is laid for gait prediction;
4) And analyzing the gait prediction result, and evaluating the performance of the gait prediction method.
Compared with the prior art, the invention has the following obvious outstanding characteristics and obvious advantages:
1. the gait prediction method solves the problems that the traditional gait prediction method depends on an accurate musculoskeletal model and needs a large amount of gait data to train the model and the like. 2. According to the ankle joint gait prediction method based on the RBF neural network, the gait prediction of the lower limb ankle joint is realized, and the prediction method has strong robustness and prediction precision. 3. The invention can be applied to the follow-up control of the lower limb exoskeleton and realizes the man-machine cooperative motion of the lower limb exoskeleton.
Drawings
Fig. 1 is a flow chart of ankle joint gait prediction based on RBF neural network of the invention.
FIG. 2 is a schematic diagram of the NDI Optotrak Certus three-dimensional motion measurement system of the present invention.
Fig. 3 is a diagram illustrating a result of ankle flexion and extension gait prediction according to a first embodiment of the invention.
Fig. 4 shows the result of prediction of ankle eversion and inversion gait according to the first embodiment of the invention.
Fig. 5 shows the predicted result of the ankle internal rotation/external rotation gait according to the first embodiment of the invention.
Detailed Description
The preferred embodiments of the present invention are described in detail below with reference to the attached drawing figures:
as shown in fig. 1, an ankle gait prediction method based on an RBF neural network includes the following operation steps:
1) Based on the RBF neural network, the rapid approximation of the ankle joint model parameters is realized:
the lagrangian equation for the lower limb ankle model 9 is defined as:
wherein D (theta) is an inertia term,the joint is in a sagittal plane, namely the rotation angle of a longitudinal section dividing the human body into a left part and a right part along the front-back direction of the human body, and T is an external factor;
because the wearer lower limb ankle joint model 9 cannot be accurately modeled, the RBF neural network 7 is adopted to approximate D (theta),and G (θ) three model parameters:
D(θ)=D SNN (θ)+E D (2)
G(θ)=G SNN (θ)+E G (4)
in the formula (5), W D ,W C And W G Weights for modelling neural networks, xi D ,Ξ C Xi and xi G Is the output of the hidden layer gaussian basis function, andas shown in formula (10) and formula (11);
wherein the content of the first and second substances,are respectively the weight valueW D ,W C ,W G Is estimated by the estimation of (a) a,
2) Designing a gait prediction algorithm, adding a robust item into the algorithm, and improving the robustness of the system:
the RBF neural network-based ankle joint gait prediction method has the core that an RBF neural network 7 is used for approximating a model matrix of a lower limb ankle joint model 9: d, C and G; meanwhile, a robust term is added into the algorithm, so that the gait prediction algorithm 6 has stronger interference resistance, and a gait prediction result 11 of the next period is obtained based on a proportional integral term;
to design the gait prediction algorithm 6, the following definitions are given:
e(t)=θ d (t)-θ(t) (9)
wherein theta is d (t) is the target trajectory 4, θ (t) is the next cycle gait prediction result 11;
thus:
wherein Λ >0; then:
the gait prediction algorithm 6 is designed as follows:
wherein, K p >O,K i O, tm is the model-based estimate term, T r Is a robust term for overcoming neural network modeling errors, and has:
T r =K r sgn(r) (18)
finally setting gait prediction algorithm parameter [ K p K i Λ K r ]Wherein K is p And K i For the PID parameters of the prediction algorithm, Λ is the coefficient of r (t) in equation (12), K r Are robust term coefficients.
3) Based on the optical motion measurement system, the ankle joint gait capture is designed and realized, and an experimental foundation is laid for gait prediction:
the purpose of the gait capture experiment is to acquire gait data of two continuous gait cycles in the human body walking process in real time; taking the gait of the previous cycle as a target track 4, inputting the gait into gait prediction 10, comparing the gait 5 of the next cycle with a gait prediction result 11 of the next cycle, and evaluating the gait prediction performance through an evaluation module 13;
as shown in fig. 2, an NDI Optotrak Certus three-dimensional motion measuring system 3 is adopted to collect gait data of the lower limbs of the human body in various gait modes; the gait capture experimental process comprises two steps of gait capture 1 and gait analysis 2; the position sensors 14, 15 are optical instruments that detect infrared light emitted from the inside of an object within a specific range; a system control unit 17 for controlling the operation of the position sensor and the additional connector, the system control unit having a port 20 connected to an ac power supply; the identification points 18 are infrared light emitting diodes that are tracked by the position sensor when they are activated by the strobe; the mark point connector collects all the mark points 18, and is connected with the system control unit 17 and controlled by the system control unit 17; the support frame 16 is used for adjusting the height of the position sensor;
After the position sensors 14 and 15 track the identification point 18 attached to the human body, the three-dimensional coordinate information of the identification point 18 is transmitted to a musculoskeletal model in the system, and the musculoskeletal model is corrected to be matched with a tested person; then, the gait analysis 2 is performed in the host computer 21, and the three-dimensional coordinate data of all the identification points attached to the lower limbs of the human body is converted into the rotation angle data of each joint and is input to the gait prediction 10 as the target trajectory 4 in real time.
In this embodiment, the parameters of gait prediction 1 are set as follows: [ K ] p K i Λ K r ]=[200 100 100 0.1](ii) a Parameters for gait prediction 2 are set as follows: [ K ] p K i Λ K r ]=[10 5 0.1 0.1](ii) a And (4) carrying out a gait capture experiment, and acquiring flexion and extension, inversion and eversion and internal rotation and external rotation gait data of two continuous gait cycles in a walking gait mode.
4) Analyzing the gait prediction result, and evaluating the performance of the gait prediction method:
adopting a root-mean-square error as a performance evaluation index of gait prediction; firstly, the root mean square error between the gait prediction result 11 of the next period and the gait 5 of the next period acquired by the gait acquisition experiment is calculated to be used as the performance evaluation index of gait prediction, and then the maximum prediction error and the minimum prediction error are calculated.
The gait prediction results show that the gait prediction results of the gait prediction 1 and the gait prediction 2 are basically consistent with the experimentally measured trend of the actual track of the ankle joint in all directions. For ankle flexion-extension movements, the root mean square error for gait prediction 1 and gait prediction 2 were 0.050rad and 0.051rad, respectively, as shown in figure 3; for ankle eversion and inversion movements, the root mean square error of gait prediction 1 and gait prediction 2 are 0.129rad and 0.121rad, respectively, as shown in FIG. 4; for the ankle internal rotation and external rotation movement, the root mean square error of gait prediction 1 and gait prediction 2 is 0.162rad and 0.198rad respectively, as shown in fig. 5; the gait prediction effect of the gait prediction 1 is better than that of the gait prediction 2.
The gait prediction method for the ankle joint of the lower limb of the human body is realized through the RBF neural network-based ankle joint gait prediction method, the prediction method is simple in structure, has strong robustness and prediction accuracy, and can overcome many defects in the existing method. The method can be applied to follow-up control of the lower limb exoskeleton and has great guiding significance for realizing man-machine cooperative motion of the lower limb exoskeleton.
Claims (3)
1. An ankle joint gait prediction method based on an RBF neural network is characterized by comprising the following operation steps:
1) Based on the RBF neural network, realizing the rapid approximation of the parameters of the ankle joint model;
2) A gait prediction algorithm is designed, and a robust item is added into the algorithm, so that the robustness of the system is improved;
3) Based on an optical motion measurement system, ankle joint gait capture is designed and realized, and an experimental foundation is laid for gait prediction;
4) Analyzing the gait prediction result and evaluating the performance of the gait prediction method;
the specific process of the step 1) is as follows:
the lagrange equation defining the lower limb ankle joint model (9) is:
wherein D (theta) is an inertia term,is the Coriolis force and centrifugal force term, G (theta) is the gravity termTheta is the rotation angle of the joint in the sagittal plane, i.e., the longitudinal section dividing the human body into left and right parts along the front-back direction of the human body, and T is an external factor;
Because the lower limb ankle joint model (9) of the wearer cannot be accurately modeled, the RBF neural network (7) is adopted to approach D (theta),and G (θ) three model parameters:
D(θ)=D SNN (θ)+E D (2)
G(θ)=G SNN (θ)+E G (4)
in the formula (5), W D ,W C And W G Weights for modelling neural networks, xi D ,Ξ C Xi and xi G Is the output of the hidden layer gaussian basis function, andas shown in formula (10) and formula (11);
wherein the content of the first and second substances,are respectively a weight W D ,W C ,W G Is estimated by the estimation of (a) a,
the specific process of the step 2) is as follows:
the core of the method is that an RBF neural network (7) is used for approximating a model matrix of a lower limb ankle joint model (9): d, C and G; meanwhile, a robust term is added into the algorithm, so that the gait prediction algorithm (6) has stronger interference resistance, and a gait prediction result (11) of the next period is obtained based on a proportional integral term;
to design the gait prediction algorithm (6), the following definitions are given:
e(t)=θ d (t)-θ(t) (9)
wherein theta is d (t) is the target trajectory (4), and theta (t) is the next cycle gait prediction result (11);
thus:
wherein Λ >0; then:
the gait prediction algorithm (6) is designed as follows:
wherein, K p >0,K i >0,T m For model-based estimation terms, T r Is a robust term for overcoming neural network modeling errors, and has:
T r =K r sgn(r) (18)
Finally setting gait prediction algorithm parameter [ K p K i Λ K r ](ii) a Wherein K p And K i For the PID parameters of the prediction algorithm, Λ is the coefficient of r (t) in equation (12), K r Is a robust term coefficient.
2. The RBF neural network-based ankle gait prediction method according to claim 1, characterized in that the specific process of step 3) is as follows:
the purpose of the gait capture experiment is to acquire gait data of two continuous gait cycles in the human body walking process in real time; taking the gait of the previous cycle as a target track (4), inputting the gait into gait prediction (10), comparing the gait (5) of the next cycle with a gait prediction result (11) of the next cycle, and evaluating the gait prediction performance through an evaluation module (13);
acquiring gait data of the lower limbs of the human body in various gait modes by adopting an NDI Optotrak Certus three-dimensional motion measuring system (3); the gait capturing experiment process comprises two steps of gait capturing (1) and gait analyzing (2); the position sensor (14, 15) is an optical instrument capable of detecting infrared light emitted from the inside of the object within a specific range; a system control unit (17) for controlling the operation of the position sensor and the additional connector, a port (20) of the system control unit being connected to an alternating current power supply; the identification points (18) are infrared light emitting diodes which are tracked by a position sensor when activated by a strobe; the mark point connector collects all the mark points (18), and is connected with the system control unit (17) and controlled by the system control unit (17); the support frame (16) is used for adjusting the height of the position sensor;
After the position sensors (14, 15) track the identification points (18) attached to the human body, the three-dimensional coordinate information of the identification points (18) is transmitted to a musculoskeletal model in the system, and the musculoskeletal model is corrected to be matched with a tested person; then, gait analysis (2) is carried out in a host computer (21), three-dimensional coordinate data of all identification points attached to the lower limbs of the human body are converted into rotation angle data of each joint, and the rotation angle data are input into gait prediction (10) as a target track (4) in real time.
3. The RBF neural network-based ankle gait prediction method according to claim 1, characterized in that the specific process of the step 4) is as follows:
adopting a root-mean-square error as a performance evaluation index of gait prediction; firstly, the root mean square error between a gait prediction result (11) of the next period and the gait (5) of the next period obtained by a gait acquisition experiment is calculated and used as a performance evaluation index of gait prediction, and then the maximum prediction error and the minimum prediction error are calculated.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910437364.9A CN110193830B (en) | 2019-05-24 | 2019-05-24 | Ankle joint gait prediction method based on RBF neural network |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910437364.9A CN110193830B (en) | 2019-05-24 | 2019-05-24 | Ankle joint gait prediction method based on RBF neural network |
Publications (2)
Publication Number | Publication Date |
---|---|
CN110193830A CN110193830A (en) | 2019-09-03 |
CN110193830B true CN110193830B (en) | 2022-10-11 |
Family
ID=67751694
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910437364.9A Active CN110193830B (en) | 2019-05-24 | 2019-05-24 | Ankle joint gait prediction method based on RBF neural network |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110193830B (en) |
Families Citing this family (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111267071A (en) * | 2020-02-14 | 2020-06-12 | 上海航天控制技术研究所 | Multi-joint combined control system and method for exoskeleton robot |
CN112947071B (en) * | 2021-01-28 | 2022-04-01 | 河北工业大学 | Backstepping-based lower limb exoskeleton control method |
CN113538514B (en) * | 2021-07-14 | 2023-08-08 | 厦门大学 | Ankle joint movement tracking method, system and storage medium |
CN113829339B (en) * | 2021-08-02 | 2023-09-15 | 上海大学 | Exoskeleton movement coordination method based on long-short-term memory network |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105213153A (en) * | 2015-09-14 | 2016-01-06 | 西安交通大学 | Based on the lower limb rehabilitation robot control method of brain flesh information impedance |
CN106821391A (en) * | 2017-03-23 | 2017-06-13 | 北京精密机电控制设备研究所 | Body gait acquisition analysis system and method based on inertial sensor information fusion |
WO2018066151A1 (en) * | 2016-10-06 | 2018-04-12 | Cyberdyne株式会社 | Gait abnormality assistance device and gait abnormality assistance method |
CN108098736A (en) * | 2016-11-24 | 2018-06-01 | 广州映博智能科技有限公司 | A kind of exoskeleton robot auxiliary device and method based on new perception |
CN109381184A (en) * | 2018-10-15 | 2019-02-26 | 刘丹 | A kind of wearable smart machine control method that auxiliary is carried |
Family Cites Families (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
KR101131776B1 (en) * | 2004-10-15 | 2012-04-06 | 혼다 기켄 고교 가부시키가이샤 | Gait generator of legged mobile robot |
CN101598618A (en) * | 2009-04-29 | 2009-12-09 | 天津大学 | Precise walker dynamometer system calibration method based on radial basis function neural network |
US9682005B2 (en) * | 2012-02-24 | 2017-06-20 | Massachusetts Institute Of Technology | Elastic element exoskeleton and method of using same |
US9044859B2 (en) * | 2013-10-31 | 2015-06-02 | Disney Enterprises, Inc. | Method for gait generation and tracking control for a bipedal walking robot |
-
2019
- 2019-05-24 CN CN201910437364.9A patent/CN110193830B/en active Active
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105213153A (en) * | 2015-09-14 | 2016-01-06 | 西安交通大学 | Based on the lower limb rehabilitation robot control method of brain flesh information impedance |
WO2018066151A1 (en) * | 2016-10-06 | 2018-04-12 | Cyberdyne株式会社 | Gait abnormality assistance device and gait abnormality assistance method |
CN108098736A (en) * | 2016-11-24 | 2018-06-01 | 广州映博智能科技有限公司 | A kind of exoskeleton robot auxiliary device and method based on new perception |
CN106821391A (en) * | 2017-03-23 | 2017-06-13 | 北京精密机电控制设备研究所 | Body gait acquisition analysis system and method based on inertial sensor information fusion |
CN109381184A (en) * | 2018-10-15 | 2019-02-26 | 刘丹 | A kind of wearable smart machine control method that auxiliary is carried |
Also Published As
Publication number | Publication date |
---|---|
CN110193830A (en) | 2019-09-03 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110193830B (en) | Ankle joint gait prediction method based on RBF neural network | |
Fang et al. | 3D human gesture capturing and recognition by the IMMU-based data glove | |
CN106156524B (en) | A kind of online gait planning method of Intelligent lower limb power assisting device | |
CN110327187B (en) | Model-free control method with prior moment for exoskeleton | |
CN203149575U (en) | Interactive upper limb rehabilitation device based on microsensor | |
CN109953761A (en) | A kind of lower limb rehabilitation robot sensory perceptual system and motion intention inference method | |
CN102567638A (en) | Interactive upper limb rehabilitation system based on micro-sensor | |
Wang et al. | Inertial odometry using hybrid neural network with temporal attention for pedestrian localization | |
Palani et al. | Real-time joint angle estimation using mediapipe framework and inertial sensors | |
Li et al. | Estimation of continuous elbow joint movement based on human physiological structure | |
CN110456902A (en) | It is mobile to control the skeleton pattern in computer system to track user | |
Minh et al. | Model Predictive Control for Modeling and Simulation of Human Gait Motions | |
Saputra et al. | Human joint skeleton tracking using multiple kinect azure | |
Wang et al. | Arbitrary spatial trajectory reconstruction based on a single inertial sensor | |
Jiang et al. | Motor ability evaluation of the upper extremity with point-to-point training movement based on end-effector robot-assisted training system | |
Ren et al. | A gait trajectory control scheme through successive approximation based on radial basis function neural networks for the lower limb exoskeleton robot | |
Zhang et al. | Motion velocity, acceleration and energy expenditure estimation using micro flow sensor | |
CN111716334A (en) | Upper limb exoskeleton control method based on sliding mode impedance | |
Lin et al. | Using hybrid sensoring method for motion capture in volleyball techniques training | |
Nandy et al. | Modern methods for affordable clinical gait analysis: theories and applications in healthcare systems | |
Luo et al. | A virtual reality system for arm and hand rehabilitation | |
Jin | Design of intelligent perception module based on wireless sensor network and basketball sports attitude | |
CN112762934B (en) | Lower limb movement direction prediction device and method | |
Jiang et al. | Digital magnetic tensor sensor with ANN measurement model for human joint motion sensing in sagittal plane | |
CN114224325A (en) | Gait analysis system and method for calculating joint moment and angle by using inertial sensor |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |