CN-122005271-A - Self-adaptive closed-loop exoskeleton system based on multi-mode brain-computer interface

CN122005271ACN 122005271 ACN122005271 ACN 122005271ACN-122005271-A

Abstract

The invention relates to a self-adaptive closed-loop exoskeleton system based on a multi-mode brain-computer interface, which comprises a) a multi-mode signal acquisition unit, b) a self-adaptive nerve decoding algorithm module, c) a three-joint exoskeleton, a redundancy switching unit, a spinal cord electrical stimulation module and a control system, wherein the three-joint exoskeleton controls the movement of an exoskeleton joint in real time along with the movement intention through a layered control strategy and dynamically adjusts gait parameters, the redundancy switching unit is switched to an EEG full-function mode within 200ms when an ECoG signal is abnormal, and the spinal cord electrical stimulation module is used for providing closed-loop auxiliary stimulation when an EEG signal is weak. The invention can provide high-precision and high-reliability motion control, has self-adaptive complex environment interaction capability, can promote nerve function remodeling and rehabilitation, and provides a safe solution for motion control of spinal cord injury patients.

Inventors

LI WENLING
ZHENG JIE
ZHANG JINGWEN
ZHANG DI
SONG XIAOLEI
HU TIANQI
ZHAO XIANYING
WANG ZIHAO
SUN CHENGLIN

Assignees

河北医科大学第二医院

Dates

Publication Date: 20260512
Application Date: 20260409

Claims (9)

1. An adaptive closed-loop exoskeleton system based on a multi-mode brain-computer interface, which is characterized by comprising: a) The multi-mode signal acquisition unit comprises an invasive ECoG stereotactic brain deep electrode implanted in a motor cortex lower limb functional area and a non-invasive dry electrode EEG cap, wherein the invasive ECoG stereotactic brain deep electrode is a columnar electrode, the electrode surface of the invasive ECoG stereotactic brain deep electrode is covered with a 10 mu m dexamethasone slow-release layer and a 50 mu m parylene insulating layer, and the non-invasive dry electrode EEG cap adopts an Ag/AgCl sintered electrode and conductive gel micro-storage tank structure, and has impedance of <5k omega; b) The self-adaptive nerve decoding algorithm module adopts a CNN-LSTM-transducer mixed network architecture, recursively updates weights every 15s, and outputs 7-dimensional lower limb movement intentions, wherein the 7-dimensional lower limb movement intentions comprise left/right hip flexion and extension, left/right knee flexion and extension, left/right ankle flexion and extension and stillness; c) The three-joint exoskeleton controls the exoskeleton joint to move along with the movement intention in real time through a layered control strategy, and dynamically adjusts gait parameters; d) A redundancy switching unit that switches to an EEG full function mode within 200ms when the ECoG signal is abnormal; e) The spinal cord electric stimulation module is used for providing closed-loop auxiliary stimulation when the electroencephalogram signal is weak, comprises an electrode array arranged outside the dura mater, and outputs stimulation current with the frequency of 40Hz and the pulse width of 300 mu s.
2. The self-adaptive closed-loop exoskeleton system of claim 1, wherein the ECoG electrode array has a sampling rate of 1kHz, a differential amplification circuit is used for inhibiting metal artifacts, and the ECoG electrode array is wirelessly powered at 13.56MHz and is used for transmitting 402-405 MHz UHF data.
3. The adaptive closed-loop exoskeleton system of claim 1 wherein said adaptive neural decoding algorithm module comprises: Time-frequency characteristic extraction, namely separating a beta frequency band of 13-30Hz and a gamma frequency band of 30-100Hz through short-time Fourier transformation; Spatial feature enhancement, namely common space mode CSP filtering focusing C3/C4/Cz region; and (3) real-time calibration, namely updating model parameters every 15 seconds, and compensating signal drift.
4. The adaptive closed-loop exoskeleton system of claim 1 wherein said hierarchical control strategy comprises: high-level control, namely converting movement intention into joint angle track, wherein the hip joint is flexed at 0-30 degrees; middle layer control, namely model predictive control MPC optimizes gait stability, predicts time domain by 0.5s, sampling period by 20ms, and cost function J=Σ (attitude error 2+moment 2); bottom control, PID motor tracking, proportional gain Kp=2.5, integration time Ti=0.1 s.
5. The adaptive closed-loop exoskeleton system of claim 1 wherein said dynamically adjusting gait parameters further comprises employing a terrain adaptation mechanism comprising: and setting up an environment map based on the laser radar and the RGB-D camera, and adjusting the step length and the leg lifting height to cross the obstacle in real time.
6. The adaptive closed-loop exoskeleton system of claim 1 wherein said neural function remodeling method comprises: Progressive disengagement, namely decreasing the power assisting proportion according to the circumference, starting from 100%, and gradually decreasing by 15% every week until the power assisting proportion is 30%; hebbian plasticity training-exoskeleton responds within 150ms after motor imagery begins.
7. The adaptive closed-loop exoskeleton system of claim 1, further comprising: and the myoelectricity monitoring module is used for collecting the rectus femoris/tibialis anterior EMG and cutting off the output when the stimulating current exceeds the threshold value.
8. The adaptive closed-loop exoskeleton system of claim 1 further comprising a gambling rehabilitation interface wherein the virtual character actions are synchronized with the exoskeleton in real time.
9. The adaptive closed-loop exoskeleton system of claim 5, wherein the steps of constructing an environment map based on the laser radar and the RGB-D camera and adjusting the step size and the leg lifting height in real time across the obstacle are as follows: Step 1, synchronously acquiring multiple sensors, namely outputting 640 multiplied by 480 depth maps by an RGB-D camera at 30fps, and outputting point clouds by a 16-line laser radar at 10Hz, wherein two paths of data are synchronously time-synchronized by PTP on Jetson Xavier NX, so that the time deviation is ensured to be less than 1 millisecond; Step 2, preprocessing and fusing point clouds, namely performing 2cm voxel filtering on the original point cloud of the laser radar, dividing the ground by using RANSAC, and only reserving points within a range of 0-50cm above the ground; Step 3, local 2.5D grid mapping, namely projecting the mixed point cloud to two-dimensional grids of 5cm multiplied by 5cm, wherein each grid stores the maximum height h_max, updating the grid map at the frequency of 30Hz, and maintaining a sliding window of 2s recently; Step 4, obstacle detection and classification, namely calculating a height difference delta h=h_max-h_group of each grid, and marking the area as an obstacle if delta h is larger than 2cm and 3 continuous grids are all met; Step 5, the gait parameters are decided in real time, namely the leg lifting height H_step=max (5 cm, h+3cm) is calculated according to the barrier height H and is not more than 40% of the leg length of a user, the step length is adjusted according to the barrier width w, wherein the step length is increased by 20% when w is less than 10cm, the step length is increased by 10% when 10-20cm, the step length is decreased by 20% when w is more than 20cm, the reference step length L 0 is preset by the user, all parameters are subjected to 2Hz low-pass filtering, and if the height change of adjacent frames is more than 3cm, a 300ms Bezier curve is adopted for smooth transition; Step 6, track generation and optimization, namely inputting the adjusted H_step and L_step into a min-jerk track generator to obtain an angle-time sequence of the hip joint, the knee joint and the ankle joint, wherein the middle layer model is predicted and controlled in a 0.5s prediction domain to optimize joint moment; and 7, carrying out safety and exception processing, namely, if the coincidence of RGB-D and LiDAR point clouds is not detected in 5 continuous frames, canceling an obstacle mark, if the angular speed of an exoskeleton joint exceeds 120 degrees/s or the plantar force is greater than 150N, immediately triggering sudden stop, automatically degrading the system when a sensor fails, shortening the step length by 20%, and fixing the leg lifting height to be 5cm.

Description

Self-adaptive closed-loop exoskeleton system based on multi-mode brain-computer interface Technical Field The invention relates to the technical field of brain-computer interfaces, in particular to a self-adaptive closed-loop exoskeleton system based on a multi-mode brain-computer interface. Background Spinal cord injury causes interruption of communication between the brain and the spinal motor center, which causes paralysis of lower limbs (such as paraplegia or quadriplegia). The prior art relies on invasive surgery (such as epidural electric stimulation, EES) or passive rehabilitation training, and has the defects of unnatural control that EES needs to be provided with a stimulation program and cannot respond to the movement intention of a patient in real time, poor adaptability, incapability of adapting to complex terrains (such as slopes and obstacles) or adjusting gait parameters (such as step length and speed) in real time, long-term dependence that the patient needs to continuously wear external equipment and nerve function recovery is limited (only auxiliary walking capacity is recovered). Brain-computer interface (BCI) technology presents a technical bottleneck in treating spinal cord injury, with non-invasive BCI (e.g., EEG) signals being noisy, with low decoding accuracy, typically <80%. While invasive BCIs (e.g., ECoG) require craniotomy implantation of electrodes, long term stability is inadequate, and signal attenuation is typically >0.03 dB/day. Therefore, the existing spinal cord injury rehabilitation exoskeleton has three defects that the brain electrical signal decoding accuracy is insufficient, the traditional EEG is easy to interfere, the single degree of freedom control cannot meet the requirement of complex actions, and the passive training mode inhibits nerve plasticity. Disclosure of Invention The invention aims to solve the problems of non-naturalization, poor adaptability and insufficient recovery of nerve functions of spinal cord injury patients in the prior art, and provides a brain control nerve recovery device based on an electroencephalogram signal-self-adaptive algorithm-exoskeleton closed-loop system, so as to realize the control of the natural motion, namely, decoding the motion intention (such as walking and turning) of the patients in real time, directly driving the exoskeleton, the interaction of self-adaptive environment, namely, dynamically adjusting gait parameters (step frequency and step length) to cope with complex terrains, and the remodeling of the nerve functions, namely, promoting the plasticity of spinal nerves through closed-loop training and gradually deviating from the assistance of the exoskeleton. To achieve the above object, the present invention provides an adaptive closed-loop exoskeleton system based on a multi-modal brain-computer interface, including: a) The multi-mode signal acquisition unit comprises an invasive ECoG stereotactic brain deep electrode implanted in a motor cortex lower limb functional area and a non-invasive dry electrode EEG cap, wherein the invasive ECoG stereotactic brain deep electrode is a columnar electrode, the electrode surface of the invasive ECoG stereotactic brain deep electrode is covered with a 10 mu m dexamethasone slow-release layer and a 50 mu m parylene insulating layer, and the non-invasive dry electrode EEG cap adopts an Ag/AgCl sintered electrode and conductive gel micro-storage tank structure, and has impedance of <5k omega; b) The self-adaptive nerve decoding algorithm module adopts a CNN-LSTM-transducer mixed network architecture, recursively updates weights every 15s, and outputs 7-dimensional lower limb movement intentions, wherein the 7-dimensional lower limb movement intentions comprise left/right hip flexion and extension, left/right knee flexion and extension, left/right ankle flexion and extension and stillness; c) The three-joint exoskeleton controls the exoskeleton joint to move along with the movement intention in real time through a layered control strategy, and dynamically adjusts gait parameters; d) A redundancy switching unit that switches to an EEG full function mode within 200ms when the ECoG signal is abnormal; e) The spinal cord electric stimulation module is used for providing closed-loop auxiliary stimulation when the electroencephalogram signal is weak, comprises an electrode array arranged outside the dura mater, and outputs stimulation current with the frequency of 40Hz and the pulse width of 300 mu s. Preferably, the sampling rate of the ECoG electrode array is 1kHz, a differential amplification circuit is adopted to inhibit metal artifacts, and the signal transmission of the ECoG electrode array adopts 13.56MHz wireless power supply and 402-405 MHz UHF data transmission. Preferably, the adaptive neural decoding algorithm module includes: Time-frequency characteristic extraction, namely separating a beta frequency band of 13-30Hz and a gamma frequency band of 30-100Hz through short-tim