CN-121806684-B - Exoskeleton position prediction and closed-loop stepping control system and method

CN121806684BCN 121806684 BCN121806684 BCN 121806684BCN-121806684-B

Abstract

The application discloses an exoskeleton position prediction and closed-loop stepping control system and method, wherein a microcontroller unit completes myoelectric signal acquisition, neural network reasoning, pulse generation and closed-loop correction functions, a myoelectric signal acquisition unit acquires surface myoelectric signals for preprocessing and feature extraction, digital myoelectric features are output, a motor position feedback unit acquires absolute position and speed information of a closed-loop stepping motor in real time, a neural network operation unit splices the digital myoelectric features and the absolute position and speed information of the closed-loop stepping motor into multidimensional input vectors, a three-dimensional space position increment of an exoskeleton is predicted at the next moment, a double-closed-loop control unit converts a predicted value of the three-dimensional space position increment into a stepping motor control instruction, and closed-loop correction is carried out by combining with a motor real-time position, zero-delay tracking control is realized through a prediction-execution parallel assembly line, high-precision low-delay tracking of exoskeleton movement is realized, and high integration and light weight of the system are realized.

Inventors

PENG JIN
REN YUHAN
PENG YINGJIE
XIONG XINYU
MA MINGYUE
Zhao Daifan
WANG HAISHI
TAN FEIFEI

Assignees

成都信息工程大学

Dates

Publication Date: 20260505
Application Date: 20260306

Claims (6)

1. An exoskeleton position prediction and closed loop stepper control system, comprising: the microcontroller unit is used for executing electromyographic signal acquisition, neural network reasoning, pulse generation and closed loop correction functions in a single chip; the electromyographic signal acquisition unit is connected to the microcontroller unit and is used for acquiring surface electromyographic signals from the big arms at the two sides of a user, preprocessing the surface electromyographic signals and extracting the characteristics, and outputting digital electromyographic characteristics; the motor position feedback unit is connected to the microcontroller unit and used for acquiring the absolute position and speed information of the closed-loop stepping motor in real time; the neural network operation unit is integrated in the microcontroller unit, realizes a one-dimensional convolution regression network based on CMSIS-NN library, splices the digital myoelectric characteristics and the absolute position and speed information of the closed-loop stepping motor into a multidimensional input vector, and predicts the three-dimensional space position increment of the exoskeleton at the next moment; the double closed-loop control unit is connected to the microcontroller unit and is used for converting the predicted value of the three-dimensional space position increment into a stepping motor control instruction, carrying out closed-loop correction by combining the real-time position of the motor and realizing zero delay tracking control through a prediction-execution parallel pipeline; The dual closed loop control unit includes: The position loop is used for comparing the position increment predicted value with the real-time position of the motor, generating a position error signal and outputting a speed reference value by adopting a PD control law; the speed loop is used for calculating the correction quantity of the pulse frequency of the motor according to the position error and carrying out amplitude limiting processing by adopting a PI control law; The pulse generation timer is used for generating a stepping motor driving pulse according to the corrected frequency to realize control of entering the next predicted position in advance; the system adopts a sliding time window to buffer and splice the myoelectric signal and the motor state; The sliding window is circularly stored in the SRAM through a DMA double-buffer mechanism; the neural network operation unit calculates a time domain root mean square value, a zero crossing rate and a wavelength respectively for the two-sided surface electromyographic signals in the sliding time window to form a 6-dimensional electromyographic characteristic vector; performing first-order and second-order differential operation on the real-time angle information acquired by the motor position feedback unit to obtain angular velocity and angular acceleration, and combining the angle values to form a 3-dimensional motor dynamic state vector; And jointly splicing the 6-dimensional myoelectricity characteristic vector, the 3-dimensional motor dynamic state vector and the motor control quantity from the previous control period into a multi-dimensional input vector.
2. The exoskeleton position prediction and closed loop stepper control system of claim 1, wherein the electromyographic signal acquisition unit comprises a differential amplification circuit, a trap circuit and an analog-to-digital conversion circuit; The differential amplifying circuit adopts a low-noise instrument amplifier to amplify signals; the trap circuit is used for filtering power frequency interference; the analog-to-digital conversion circuit is formed by cascading an external ADC with an on-chip ADC in the microcontroller unit.
3. The exoskeleton position prediction and closed-loop stepper control system of claim 1, wherein the motor position feedback unit employs a hybrid closed-loop stepper motor, the hybrid closed-loop stepper motor integrating a magnetic encoder; The magnetic encoder feeds back absolute angle position data of the motor rotor to the microcontroller unit in real time through the SPI communication interface; the microcontroller unit performs a filtering process on the received absolute angular position data.
4. The exoskeleton position prediction and closed loop stepping control system of claim 1, wherein the hierarchy of the one-dimensional convolution regression network comprises, in order, a first convolution layer, a first activation layer, a first averaging pooling layer, a second convolution layer, a second activation layer, a second averaging pooling layer, a flattening layer, a full connection layer, a third activation layer, and an output regression layer; the 3 nodes of the output regression layer correspond to the position increment of the exoskeleton in three directions of a space X, Y, Z respectively; The one-dimensional convolution regression network has undergone INT8 full integer quantization prior to execution.
5. The exoskeleton position prediction and closed loop stepper control system of claim 1 wherein the neural network arithmetic unit is further configured to map the three-dimensional spatial position increments to angular increments of the shoulder, elbow and wrist joints, respectively; smoothing and filtering the angle increment; and superposing the joint angle increment after the smoothing treatment with the current actual angle of the joint obtained by the motor position feedback unit to generate a target tracking position of each joint, and sending the target tracking position into the double closed-loop control unit.
6. A method for exoskeleton position prediction and closed loop stepper control, wherein the method is applied to the exoskeleton position prediction and closed loop stepper control system of any one of claims 1 to 5, and comprises: Collecting surface electromyographic signals of the big arms at the two sides of a user; Acquiring absolute position and speed information of a closed-loop stepping motor in real time; Splicing myoelectricity characteristics in the sliding window with absolute position and speed information into a multidimensional input vector; The step of splicing myoelectric characteristics, absolute position and speed information in the sliding window into a multidimensional input vector comprises the following steps: respectively calculating a time domain root mean square value, a zero crossing rate and a wavelength for the myoelectric signals on the two side surfaces in the sliding time window to form a 6-dimensional myoelectric characteristic vector; performing first-order and second-order differential operation on the real-time angle information acquired by the motor position feedback unit to obtain angular velocity and angular acceleration, and combining the angle values to form a 3-dimensional motor dynamic state vector; The 6-dimensional myoelectricity characteristic vector, the 3-dimensional motor dynamic state vector and the motor control quantity from the previous control period are spliced together to form a multi-dimensional input vector; a three-dimensional position increment predicted value of the exoskeleton at the next moment is obtained by reasoning through a one-dimensional convolution regression network solidified in the micro controller chip; And converting the position increment predicted value into a stepping motor pulse instruction, and carrying out closed-loop correction on the position of the stepping motor and the real-time position of the stepping motor, thereby realizing zero-delay tracking control of exoskeleton movement.

Description

Exoskeleton position prediction and closed-loop stepping control system and method Technical Field The application relates to the technical field of man-machine interaction, in particular to an exoskeleton position prediction and closed-loop stepping control system and method. Background Currently, an intention recognition scheme based on a joint angle sensor or a moment sensor is generally adopted in an upper limb assistance/rehabilitation exoskeleton system, and the movement intention of a wearer is judged by collecting angle change of joints such as shoulders, elbows and the like or handle moment information. Such schemes typically acquire sensory data at sampling frequencies no greater than 200Hz and rely on a computer or high performance embedded platform to run complex deep learning models for intent resolution and motion planning. Although the method can realize a certain degree of motion following, the method depends on an external computing platform, so that the whole system is large in size and high in power consumption, the response delay from signal acquisition to motion execution is generally more than 300ms, and real-time and natural interaction requirements are difficult to meet. To further increase the response speed, some studies have attempted to introduce surface electromyographic signals as the source of intended perception, taking advantage of the property that their neuromuscular activity leads the actual motion in order to achieve motion prediction. However, limited by the limited computing power and storage resources of the traditional microcontroller, the existing embedded scheme can only extract amplitude characteristics of the electromyographic signals and judge based on a simple threshold value, so that accurate and advanced prediction of the spatial position track of the upper limb in the future cannot be realized. In addition, the system adopts an open loop stepping motor driving mode, and the step-out risk caused by abrupt load change or high-speed movement exists, so that the position control precision is poor, the composite movement track formed by the shoulder, elbow and wrist joints is difficult to stably and accurately track, and the requirement of high-speed and high-precision auxiliary of one step in advance cannot be met. In summary, the prior art has remarkable limitations in the links of intention recognition, motion prediction and drive control of an exoskeleton system, or relies on an external computing platform with high power consumption and large volume to cause large response delay and poor portability, or is limited by embedded computing force although an electromyographic signal is tried to be adopted, so that simple real-time action triggering can be realized and the prediction capability is lacking, and meanwhile, the accuracy and reliability are insufficient by adopting open-loop stepping control at an execution level. Therefore, a highly integrated lightweight solution is urgently needed, and high-speed acquisition of the electromyographic signals on the surfaces of the large arms on two sides, real-time motion intention prediction based on an artificial intelligent model and high-precision closed-loop driving control of a stepping motor can be synchronously completed in a single microcontroller chip, so that the response speed, prediction capacity and tracking precision of the exoskeleton are obviously improved on the premise of ensuring the compactness and low power consumption of the system. Disclosure of Invention The application aims to overcome the defects of the prior art, provides an exoskeleton position prediction and closed-loop stepping control system and method, and can solve the defects of large intended identification delay, need of an external AI accelerator, easiness in open-loop stepping, high system volume power consumption and the like in the prior art, and realize zero-delay tracking of shoulder-elbow-wrist composite motion. The aim of the application is achieved by the following technical scheme: in a first aspect, the present application provides an exoskeleton position prediction and closed loop stepper control system comprising: the microcontroller unit is used for executing electromyographic signal acquisition, neural network reasoning, pulse generation and closed loop correction functions in a single chip; the electromyographic signal acquisition unit is connected to the microcontroller unit and is used for acquiring surface electromyographic signals from the big arms at the two sides of a user, preprocessing the surface electromyographic signals and extracting the characteristics, and outputting digital electromyographic characteristics; the motor position feedback unit is connected to the microcontroller unit and used for acquiring the absolute position and speed information of the closed-loop stepping motor in real time; the neural network operation unit is integrated in the microcontroller unit, realizes a one-dimensional con