Gait planning method for time-lag reservoir of lower limb prosthesis

Abstract

The invention belongs to the field of lower limb prosthesis control, and particularly relates to a gait planning method for a time-lag reservoir of a lower limb prosthesis. A hip joint angle is used as input, a knee joint angle is used as output, time sequence information existing in input data is fully utilized, and an accurate reference trajectory with high disturbance adaptability is generated. Meanwhile, hardware deployment of the model is realized by adopting the FPGA, and experiments prove that by adopting the FPGA as a hardware implementation platform, the frequency requirement of artificial limb control can be met, meanwhile, the precision is high, and accurate and precise artificial limb control can be realized. According to the method, the adaptability of the artificial limb to gait mutation is improved, and a new solution is provided for artificial limb control.

Description

Technical Field

[0001] The invention belongs to the field of lower limb prosthesis control, and in particular relates to a time-delay reserve pool gait planning method for a lower limb prosthesis. Background Art

[0002] There are more than 85 million disabled people in my country. The loss of lower limbs seriously affects the quality of life of people with lower limb disabilities. At the same time, due to the loss of residual limbs, the disabled people's bodies are unbalanced on both sides, which may cause further damage. In order to restore the disabled people's motor ability before amputation, it is necessary to develop lower limb prostheses with better performance.

[0003] The lower limb prostheses currently in common use are passive prostheses, which are easy to use but cannot provide the active excitation required for walking and cannot adapt to bionics, and can only meet the most basic needs. Active prostheses can output active torque and provide better support, so the development of high-performance active prostheses is of great practical significance.

[0004] In the development of active prostheses, control is a key technical difficulty. How to achieve stable and accurate prosthesis control is the key to prosthesis design. Currently, the widely used control methods include impedance control based on finite state machines and position control based on reference trajectories. Among them, the finite state machine control method has a large number of control parameters, and the control process is discontinuous during gait conversion, so it cannot achieve completely reliable prosthesis control. The control method based on reference trajectories uses fewer model parameters. By accurately modeling the reference trajectory, the stability of the prosthesis can be improved. The current reference trajectory modeling method usually uses a static model to model the joint angles of the lower limbs. This method is simple to implement and can better fit the gait trajectory during stable walking. However, since the static model has no memory, and the gait data of human motion naturally contains time series information, the use of the static model cannot utilize the time series relationship. At the same time, when the gait changes suddenly, the adaptability to interference is poor. Therefore, it is necessary to propose a model that can use time series information to model the joint angles and improve the adaptability of the prosthesis to interference. Summary of the invention

[0005] The purpose of the present invention is to provide a time-delayed reservoir gait planning method for lower limb prosthesis. The present invention proposes to use a time-delayed feedback reservoir for modeling hip and knee joint angle mapping to improve the accuracy and disturbance adaptability of prosthesis control. The method of the present invention only needs to use the hip joint angle on the prosthesis side as the input of the model, and the input data collection relies on the IMU sensor unit at the hip joint, which is simple to implement and uses fewer sensors. At the same time, the present invention proposes to use FPGA to deploy the time-delayed feedback reservoir model to improve the calculation speed of the model.

[0006] The present invention proposes a time-delay reserve pool gait planning method for a lower limb prosthesis, and the specific steps are as follows:

[0007] (1) Modeling of a time-delay feedback reservoir, wherein the time-delay feedback reservoir model includes an input layer, a reservoir and an output layer, wherein: a random mask is added to the input layer to achieve time multiplexing of input samples within a delay period, and the masked samples are input into the reservoir; the reservoir contains an oscillator, which generates the state response of each virtual node to the input data; the states of the virtual nodes are linearly superimposed in the output layer to finally generate the output signal of the reservoir; the input of the model is the hip joint angle, and the output is the knee joint angle; 70% of the data in the gait data set is used as the training set to calculate the weight of the output layer of the time-delay feedback reservoir, which is then used for simulation experiments and deployment of the time-delay feedback reservoir model;

[0008] (2) A time-delay feedback reservoir model is set up on the ZYNQ (Zynq-7000, Xilinx, American) development board. The time-delay feedback reservoir model is located at the PL end of the ZYNQ development board. The PS end of the ZYNQ development board is responsible for data format conversion and serial communication. The PL end and the PS end communicate through BRAM. The PL end uses registers to store the states of nonlinear nodes in the time-delay feedback reservoir. The storage method is fixed-point numbers to balance the hardware usage area and the accuracy of the model.

[0009] (3). Use serial port polling mode to communicate with STM32 microprocessor (STM32F4, ST, Switzerland) to obtain hip joint angle transmitted by STM32 microprocessor, and transmit the mapped knee joint angle back to STM32 microprocessor for subsequent control; wherein: the input end of STM32 microprocessor is respectively connected to IMU sensor unit, ZYNQ and six-dimensional force sensor, STM32 microprocessor is bidirectionally connected to actuator, the input end of actuator is connected to the output end of angle encoder, the output end of actuator is connected to motor, STM32 microprocessor and motor are respectively connected to power supply system, STM32 microprocessor communicates with IMU sensor unit, ZYNQ and six-dimensional force sensor to realize input hip joint angle signal collection, hip and knee joint mapping and ankle joint force measurement; after calculating the reference knee joint angle, the motor is driven by actuator to realize the movement of prosthesis.

[0010] In the present invention, the oscillator is a Mackey-Glass oscillator.

[0011] The beneficial effects of the present invention are:

[0012] 1. The short-term memory capacity of the delayed feedback reserve pool can make full use of the temporal information in the gait data and improve the mapping ability of the model;

[0013] 2. By selecting model parameters, the reserve pool can produce different dynamics for different inputs, making the model sensitive to input changes and able to adapt to gait mutations;

[0014] 3. Hardware implementation on FPGA can improve the computation speed and enable the deployment of larger models;

[0015] 4. The computational burden on the controller is small, and the computational frequency meets the requirements of prosthetic control. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 Modeling processes for time-delayed feedback reservoirs.

[0017] Figure 2 The architecture diagram of the time-delay feedback reserve pool system implemented for ZYNQ.

[0018] Figure 3 Schematic diagram of the control architecture of a lower limb prosthesis. (a) is a physical diagram of the power supply of a lower limb prosthesis machine, and (b) is the prosthesis control architecture and its components. DETAILED DESCRIPTION

[0019] The present invention is further described below by way of embodiments in conjunction with the accompanying drawings.

[0020] Example 1: A subject with right thigh amputation (170cm, 68kg), with residual limb accounting for 60% of the thigh, and wearing prosthesis for 25 years. In the prosthesis wearing experiment, the subject walked at a constant speed on the treadmill at 2.5 and 3.0km / h, and the walking time was about 50-60 seconds. During the walking process, he was subjected to a random tension of 10%-20% of his body weight to simulate the disturbance encountered during walking. Handrails are provided on both sides of the treadmill to prevent the subject from falling due to instability during walking. In the experiment, the subject wore a belt with a cable around his waist, and the tension was applied through the cable. The magnitude of the tension was recorded by the tension machine at the other end of the cable. The gait data of the subject's walking was collected by a motion capture system. The experimental results show that at both speeds, the subject can achieve stable walking, and can still remain stable when disturbed, proving that the use of a time-delay feedback reserve pool can learn the gait pattern of healthy people when walking, and generate an accurate reference trajectory that is highly adaptable to gait mutations.

[0021] Figure 1This is the modeling process of the time-delay feedback reservoir. The input continuous hip joint angle data is multiplied by a random mask after sampling and holding operations to achieve time multiplexing. The output of the preprocessing stage is J(t). The input data generates a dynamic response in the nonlinear node, and the control equation of the nonlinear node uses a Mackey-Glass oscillator. At the same time, there are N virtual nodes on the delay loop in the reservoir. The interval time of each virtual node is θ, and the total delay time is τ. The weights of the output layer are trained using linear regression or ridge regression through the node state of the reservoir, and the knee joint angle is finally output. The trained weights are used to fit the knee joint angle in the experiment.

[0022] Figure 2 The architecture diagram of the time-delay feedback reservoir system implemented for ZYNQ is divided into two parts: the processing system (PS) and the FPGA (PL). The PS has an ARM Cortex-A9 processor, which is responsible for the conversion of input and output data formats and the implementation of serial communication. Specifically, the PS converts the floating-point data transmitted by the STM32 into fixed-point numbers, and then passes them to the PL through the BRAM. The data is first time-multiplexed through a mask, and then the state of the reservoir is stored in the PL through registers, and the input data is responded to. The nonlinear node of the reservoir uses a Mackey-Glass oscillator, and the virtual node is mapped through the output layer to obtain the knee joint angle, which is finally passed to the STM32 by the PS.

[0023] Figure 3 It is the overall framework for controlling lower limb prostheses, including Figure 3 (a) is a real picture of the lower limb prosthesis. Figure 3 (b) is the control component and parts of the prosthesis. As shown in the figure, the processor and motor of the prosthesis are powered by the same power system. The STM32 microprocessor communicates with the IMU sensor unit, ZYNQ and the six-dimensional force sensor through CAN, UART and SPI Port respectively to realize the input hip joint angle signal acquisition, hip and knee joint mapping and ankle joint force measurement. After the reference knee joint angle is calculated, it is transmitted to the actuator drive motor through CAN communication to realize the movement of the prosthesis. The angle encoder is used to encode the joint angle to realize the underlying motion control of the joint motor. The control frequency of the prosthesis is maintained at 100Hz, and the clock frequency of the PL end is 100MHz.

Claims

1. A time-delay reserve pool gait planning method for lower limb prosthesis, characterized in that The specific steps are as follows: (1) Modeling of a time-delay feedback reservoir, wherein the time-delay feedback reservoir model includes an input layer, a reservoir and an output layer, wherein: a random mask is added to the input layer to realize time multiplexing of input samples within a delay period, and the masked samples are input into the reservoir; the reservoir contains an oscillator, which generates the state response of each virtual node to the input data; the states of the virtual nodes are linearly superimposed in the output layer to finally generate the output signal of the reservoir; the input of the model is the hip joint angle, and the output is the knee joint angle; 70% of the data in the gait data set is used as the training set to calculate the weight of the output layer of the time-delay feedback reservoir, which is then used for simulation experiments and deployment of the time-delay feedback reservoir model; (2) A time-delay feedback reservoir model is set up on the ZYNQ development board. The time-delay feedback reservoir model is located on the PL side of the ZYNQ development board. The PS side of the ZYNQ development board is responsible for data format conversion and serial communication. The PL side and the PS side communicate through BRAM. The PL side uses registers to store the states of nonlinear nodes in the time-delay feedback reservoir. The storage method is fixed-point numbers to balance the hardware usage area and the accuracy of the model. (3). Use serial port polling mode to communicate with the STM32 microprocessor to obtain the hip joint angle transmitted by the STM32 microprocessor, and transmit the mapped knee joint angle back to the STM32 microprocessor for subsequent control; wherein: the input end of the STM32 microprocessor is respectively connected to the IMU sensor unit, ZYNQ and the six-axis force sensor, the STM32 microprocessor is bidirectionally connected to the actuator, the input end of the actuator is connected to the output end of the angle encoder, the output end of the actuator is connected to the motor, the STM32 microprocessor and the motor are respectively connected to the power supply system, the STM32 microprocessor communicates with the IMU sensor unit, ZYNQ and the six-axis force sensor respectively, to realize the input hip joint angle signal acquisition, hip and knee joint mapping and ankle joint force measurement; after calculating the reference knee joint angle, the motor is driven by the actuator to realize the movement of the prosthesis.

2. The time-delay reserve pool gait planning method for a lower limb prosthesis according to claim 1, characterized in that The oscillator used is a Mackey-Glass oscillator.

Images