Training assisting robot for cognitive rehabilitation training

Abstract

The utility model discloses a training assisting robot for cognitive rehabilitation training. The training auxiliary robot comprises an auxiliary training control host, a double-end mechanical arm support, a left mechanical arm, a right mechanical arm, a left servo sensing glove and a right servo sensing glove. The auxiliary training control host is provided with a multi-modal fusion neural network module and a reinforcement learning mechanical arm control module. The training assisting robot for cognitive rehabilitation training can detect electromyographic signals, force signals and torque signals of a rehabilitation patient in real time, so that a corresponding mechanical arm control decision is generated, and the patient is assisted in cognitive rehabilitation training.

Description

Technical Field

[0001] This utility model relates to the field of rehabilitation training technology, specifically to a training assist robot for cognitive rehabilitation training. Background Technology

[0002] Although modern medicine has effectively reduced the mortality rate of stroke, it cannot prevent postoperative sequelae, resulting in a postoperative disability rate of 70% to 80% for stroke patients, with 42% of patients unable to care for themselves. Therefore, postoperative rehabilitation is particularly important. With the vigorous development of intelligent control and robotics technology, medical rehabilitation robots have gradually become an indispensable tool in the postoperative rehabilitation process.

[0003] Clinical results show that compared to manual rehabilitation therapy, the intervention of rehabilitation robots can effectively improve treatment efficiency, and rehabilitation robots have better precision than human hands. Adding robots with electromyographic (EMG) feedback can achieve even better training results. Especially for stroke patients, rehabilitation training can promote muscle tissue regeneration and strengthening; the addition of EMG feedback can train the entire reflex arc of the nervous system. If effective rehabilitation training is conducted after stroke surgery and the end of drug treatment, the likelihood of patients regaining normal motor function is greatly increased, thereby reducing the incidence of postoperative motor dysfunction and cognitive impairment.

[0004] Effective rehabilitation training can significantly reduce the incidence of postoperative syndromes in stroke patients. Current clinical research indicates that through systematic rehabilitation training, muscle damage can be repaired, and with repeated stimulation, the nervous system can also repair and regenerate. Motor dysfunction and daily living ability impairments in stroke patients can also be significantly improved. Regarding cognitive rehabilitation training, existing equipment requires a certain level of muscle strength from patients. However, most stroke patients and cardiovascular patients experience both motor and cognitive impairments in the early postoperative period. Therefore, these patients often have insufficient muscle strength in the early postoperative period and cannot complete cognitive rehabilitation training based on existing equipment. This causes patients with cardiovascular disease to miss the golden period for cognitive rehabilitation training in the early postoperative period.

[0005] Currently, cognitive rehabilitation training robot technology has made some progress in helping patients restore cognitive function. For example, the Second Affiliated Hospital of Xi'an Jiaotong University School of Medicine published a patent, "An Intelligent Robot for Improving Cognition in Alzheimer's Patients" (application number: CN201910941464.5), which has a positive effect on promoting cognitive rehabilitation. However, existing technologies still have significant shortcomings, especially in training methods and interaction approaches. Traditional cognitive training robots typically rely on voice prompts or visual feedback to guide patients during training, primarily interacting with them through language. However, a purely voice-based approach cannot fully engage the patient's multiple senses, nor can it effectively guide individual motor coordination and limb activities. For some cognitive tasks requiring motor coordination and fine motor control, simple voice guidance often fails to provide sufficient assistance, thus limiting the training effect and patient participation. Especially for patients with limited motor abilities, training methods lacking tactile feedback and physical interaction often fail to fully mobilize their cognitive abilities, leading to unsatisfactory rehabilitation results. Utility Model Content

[0006] In view of the needs of addressing the aforementioned practical problems and the existing technologies involved, the purpose of this invention is to provide a training assist robot for cognitive rehabilitation training. This invention's training assist robot can collect force signals, torque signals, and electromyographic signals of rehabilitation training patients in real time and generate corresponding auxiliary motion decisions based on these signals, guiding the robotic arm and servo-driven gloves to assist patients in completing cognitive rehabilitation training. Simultaneously, this invention can continuously optimize auxiliary motion commands during the training process using reinforcement learning methods, improving the patient's performance in completing rehabilitation cognitive training tasks.

[0007] The objective of this utility model is achieved through at least one of the following technical solutions.

[0008] A training assistive robot for cognitive rehabilitation training includes an assistive training control host, a dual-head robotic arm support, a left robotic arm, a right robotic arm, a left servo sensor glove, and a right servo sensor glove.

[0009] The auxiliary training control unit and the dual-head robotic arm support are located in front of the patient. The dual-head robotic arm support is mounted on the auxiliary training control unit, and the two are physically connected. The dual-head robotic arm support connects the left and right robotic arms. The left and right robotic arms are connected to the auxiliary training control unit via wires, and both have accessory slots at their ends. The left robotic arm is connected to the left servo sensor glove at its end via an accessory slot, and the right robotic arm is connected to the right servo sensor glove at its end via an accessory slot. The left and right servo sensor gloves are connected to the auxiliary training control unit via wires.

[0010] The left and right robotic arms collect force and torque signals and send these signals to the auxiliary training control host via signal lines. The left and right servo sensor gloves collect the patient's hand movement and force signals through built-in sensors and send these signals to the auxiliary training control host via signal lines. The auxiliary training control host performs multimodal fusion decision-making and reinforcement learning based on the input signals to obtain optimized control commands for the left and right robotic arms, left and right servo sensor gloves, and sends the corresponding control commands to these gloves to assist the patient in completing cognitive training tasks.

[0011] Furthermore, the auxiliary training control host is equipped with a multimodal fusion neural network module and a reinforcement learning robotic arm control module. The multimodal fusion neural network module and the reinforcement learning robotic arm control module are based on existing graph neural networks and deep deterministic policy gradient (DDPG) algorithms, respectively. They perform multimodal fusion decision-making and reinforcement learning on the force and torque signals transmitted from the left and right robotic arms, as well as the hand motion and force signals collected by the left and right servo sensing gloves, to obtain optimized control commands for the left and right robotic arms, the left and right servo sensing gloves.

[0012] Furthermore, the dual-head robotic arm support adopts a triangular support structure, which can enhance stability, reduce the swaying interference of the left and right robotic arms on the support during movement, and improve the accuracy of operation.

[0013] Furthermore, the dual-head robotic arm support allows for quick disassembly and assembly, facilitating transportation and deployment in different treatment environments.

[0014] Furthermore, the left and right servo sensing gloves incorporate multiple highly sensitive sensors, including pressure sensors, bending sensors, and tactile sensors, to accurately capture hand movements and force changes.

[0015] Furthermore, the finger portions of the left and right servo sensing gloves are made of flexible materials to mimic the natural movements of the human hand, and the built-in sensors can detect and respond to subtle finger movements.

[0016] Furthermore, high-resolution absolute encoders are installed at the joints of the left and right robotic arms to monitor and control their positions and orientations in real time, ensuring operational precision.

[0017] Furthermore, the auxiliary training control host has an overload protection function. When the left robotic arm, right robotic arm, left servo sensor glove, or right servo sensor glove is subjected to a force exceeding the design range, the system will automatically disconnect the power supply to protect the equipment and user safety.

[0018] Furthermore, the multimodal fusion neural network module uses an existing fusion algorithm based on graph neural networks to take force signals, torque signals, and hand movement and force signals as inputs to the neural network. The neural network predicts the patient's movement intention and obtains corresponding auxiliary movement decisions. The auxiliary training control host sends corresponding auxiliary movement commands to the left robotic arm, right robotic arm, left servo sensing glove, and right servo sensing glove according to the auxiliary movement decisions to assist the patient in completing cognitive training tasks.

[0019] Furthermore, the reinforcement learning robotic arm control module evaluates the completion of cognitive rehabilitation training tasks in real time during the cognitive rehabilitation training process, optimizes auxiliary motion decisions through the existing deep deterministic policy gradient (DDPG) reinforcement learning algorithm, thereby optimizing the corresponding auxiliary motion commands and improving the patient's completion of cognitive training tasks.

[0020] Compared with the prior art, the beneficial effects of this utility model are:

[0021] The training assist robot for cognitive rehabilitation described in this utility model can automatically collect force signals, torque signals and electromyographic signals of patients during the training process, and perform fusion analysis on these signals through a multimodal fusion neural network to predict the patient's movement intentions and generate corresponding movement assist commands.

[0022] This invention utilizes a reinforcement learning mechanism to continuously optimize motor assistive instructions based on the patient's completion of cognitive training tasks, thereby improving the patient's cognitive training task completion and enhancing the effectiveness of cognitive rehabilitation treatment.

[0023] This invention overcomes the shortcomings of existing cognitive rehabilitation training robot technologies by introducing innovative technologies such as a robotic arm and a servo-sensing glove. Through the physical intervention of the robotic arm and servo-sensing glove, this invention can guide patients to complete simple cognitive training movements and tasks, fully mobilizing their motor coordination and physical participation. For example, the servo-sensing glove can provide precise force feedback, guiding patients to complete hand movements and fine motor operations, while the robotic arm can assist patients in performing specific body movements or tasks, helping them complete more complex movements during training. Through this multi-sensory, whole-body participation training method, not only can patients' cognitive responses be improved, but their motor function recovery can also be promoted. Attached Figure Description

[0024] Figure 1 This is a schematic diagram of the structure of a training assist robot for cognitive rehabilitation training according to an embodiment of the present invention;

[0025] Figure 2This is a control principle diagram of a training assist robot for cognitive rehabilitation training according to an embodiment of the present invention;

[0026] Figure 3 This is a flowchart illustrating the workflow of a training assist robot for cognitive rehabilitation training according to an embodiment of the present invention. Detailed Implementation

[0027] To more clearly illustrate the technical solutions in the embodiments of this utility model or the prior art, the accompanying drawings and technical solutions used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this utility model. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. The specific implementation of this utility model will be further described below in conjunction with the accompanying drawings.

[0028] Example:

[0029] A training assistive robot for cognitive rehabilitation training, such as Figure 1 As shown, it includes an auxiliary training control host 1, a dual-head robotic arm support 2, a left robotic arm 3, a right robotic arm 4, a left servo sensor glove 5, and a right servo sensor glove 6.

[0030] The auxiliary training control host 1 and the dual-head robotic arm support 2 are located in front of the patient. The dual-head robotic arm support 2 is mounted on the auxiliary training control host 1 and the two are physically connected. The dual-head robotic arm support 2 connects the left robotic arm 3 and the right robotic arm 4. The left robotic arm 3 and the right robotic arm 4 are connected to the auxiliary training control host 1 through wires, and both have accessory slots at their ends. The left robotic arm 3 is connected to the left servo sensor glove 5 at its end through the accessory slot, and the right robotic arm 4 is connected to the right servo sensor glove 6 at its end through the accessory slot. The left servo sensor glove 5 and the right servo sensor glove 6 are connected to the auxiliary training control host 1 through wires.

[0031] The left robotic arm 3 and the right robotic arm 4 collect force and torque signals and send these two types of signal data to the auxiliary training control host 1 via signal lines. The left servo sensor glove 5 and the right servo sensor glove 6 collect the patient's hand movement and force signals through built-in sensors and send the signal data to the auxiliary training control host 1 via signal lines. The auxiliary training control host 1 performs multimodal fusion decision-making and reinforcement learning based on the input signals to obtain optimized control commands for the left robotic arm 3, the right robotic arm 4, the left servo sensor glove 5, and the right servo sensor glove 6, and sends the corresponding control commands to the left robotic arm 3, the right robotic arm 4, the left servo sensor glove 5, and the right servo sensor glove 6 to assist the patient in completing cognitive training tasks.

[0032] In one embodiment, the auxiliary training control host 1 is equipped with a multimodal fusion neural network module and a reinforcement learning robotic arm control module. The multimodal fusion neural network module and the reinforcement learning robotic arm control module are based on existing graph neural networks and deep deterministic policy gradient (DDPG) algorithms, respectively. They perform multimodal fusion decision-making and reinforcement learning on the force and torque signals transmitted from the left robotic arm 3 and the right robotic arm 4, as well as the hand motion and force signals collected by the left servo sensing glove 5 and the right servo sensing glove 6, to obtain optimized control commands for the left robotic arm 3, the right robotic arm 4, the left servo sensing glove 5, and the right servo sensing glove 6.

[0033] In one embodiment, the dual-head robotic arm support 2 adopts a triangular support structure, which can enhance stability, reduce the swaying interference of the left robotic arm 3 and the right robotic arm 4 on the support during movement, and improve the accuracy of operation.

[0034] In one embodiment, the dual-head robotic arm support 2 allows for quick disassembly and assembly, facilitating transportation and deployment in different treatment environments.

[0035] In one embodiment, the left servo sensing glove 5 and the right servo sensing glove 6 are hand rehabilitation devices developed by PalmMing, which have multiple built-in high-sensitivity sensors, including pressure sensors, bending sensors and tactile sensors, to accurately capture hand movements and force changes.

[0036] In one embodiment, the finger portions of the left servo sensing glove 5 and the right servo sensing glove 6 are made of flexible material to mimic the natural movements of the human hand, and the built-in sensors are able to detect and respond to the subtle movements of the fingers.

[0037] In one embodiment, the left robotic arm 3 and the right robotic arm 4 are selected from the Danish Universal Robots UR3e robotic arms, which are equipped with high-resolution absolute encoders at their joints to monitor and control the position and orientation of the left robotic arm 3 and the right robotic arm 4 in real time, ensuring the accuracy of the operation.

[0038] In one embodiment, the auxiliary training control host 1 has an overload protection function. When the left robotic arm 3, right robotic arm 4, left servo sensor glove 5, or right servo sensor glove 6 are subjected to forces exceeding the design range, the system will automatically disconnect the power supply to protect the equipment and user safety.

[0039] In one embodiment, the multimodal fusion neural network module, referencing the multimodal fusion model proposed in the literature "A Multimodal Information Fusion Model for Robot Action Recognition with Time Series", uses force signals, torque signals, and hand motion and force signals as inputs to the neural network. The neural network predicts the patient's movement intention and obtains corresponding auxiliary movement decisions. The auxiliary training control host 1 sends corresponding auxiliary movement commands to the left robotic arm 3, right robotic arm 4, left servo sensing glove 5, and right servo sensing glove 6 according to the auxiliary movement decisions, assisting the patient in completing cognitive training tasks.

[0040] In one embodiment, the reinforcement learning robotic arm control module evaluates the completion of cognitive rehabilitation training tasks in real time during cognitive rehabilitation training. It optimizes auxiliary motion decisions using the Deep Deterministic Policy Gradient (DDPG) reinforcement learning algorithm proposed in the reference "Adaptive PID computed-torque control of robot manipulators based on DDPG reinforcement learning," thereby optimizing the corresponding auxiliary motion commands and improving the patient's completion of cognitive training tasks.

[0041] like Figure 2 and Figure 3 As shown, when a patient undergoing cognitive rehabilitation sits in the chair in front of the training assistive robot, they need to wear servo-sensor gloves on both hands and begin the corresponding cognitive training task. During the execution of the cognitive training task, the training assistive robot will collect the patient's force signals, torque signals, and electromyographic signals in real time and generate corresponding auxiliary motion decisions based on these signals. This guides the robotic arm and servo-driven gloves to assist the patient in completing the cognitive rehabilitation training. Simultaneously, the training assistive robot will optimize the auxiliary motion commands based on the task completion status, thereby improving the effectiveness of cognitive rehabilitation training.

[0042] The above description is merely a specific embodiment of this utility model, but the protection scope of this utility model is not limited thereto. Any changes or substitutions conceived without inventive effort should be included within the protection scope of this utility model. Therefore, the protection scope of this utility model should be determined by the scope defined in the claims.

Claims

1. A training assistive robot for cognitive rehabilitation training, characterized in that, It includes an auxiliary training control host (1), a dual-head robotic arm support (2), a left robotic arm (3), a right robotic arm (4), a left servo sensor glove (5), and a right servo sensor glove (6); The auxiliary training control host (1) and the dual-head robotic arm support (2) are located in front of the patient. The dual-head robotic arm support (2) is installed on the auxiliary training control host (1) and the two are physically connected. The dual-head robotic arm support (2) connects the left robotic arm (3) and the right robotic arm (4). The left robotic arm (3) and the right robotic arm (4) are connected to the auxiliary training control host (1) through wires, and both have accessory slots at their ends. The left robotic arm (3) is connected to the left servo sensor glove (5) at its end through the accessory slot, and the right robotic arm (4) is connected to the right servo sensor glove (6) at its end through the accessory slot. The left servo sensor glove (5) and the right servo sensor glove (6) are connected to the auxiliary training control host (1) through wires, respectively. The left robotic arm (3) and the right robotic arm (4) collect force signals and torque signals, and send the two types of signal data to the auxiliary training control host (1) through signal lines; the left servo sensing glove (5) and the right servo sensing glove (6) collect the patient's hand movement and force signals through built-in sensors, and send the signal data to the auxiliary training control host (1) through signal lines; the auxiliary training control host (1) performs multimodal fusion decision-making and reinforcement learning based on the input signals, obtains optimized control commands for the left robotic arm (3), the right robotic arm (4), the left servo sensing glove (5) and the right servo sensing glove (6), and sends the corresponding control commands to the left robotic arm (3), the right robotic arm (4), the left servo sensing glove (5) and the right servo sensing glove (6) to assist the patient in completing the cognitive training task.

2. The training assistive robot for cognitive rehabilitation training according to claim 1, characterized in that, The auxiliary training control host (1) is equipped with a multimodal fusion neural network module and a reinforcement learning robotic arm control module. The multimodal fusion neural network module and the reinforcement learning robotic arm control module are based on existing algorithms to perform multimodal fusion decision-making and reinforcement learning on the force signals and torque signals transmitted from the left robotic arm (3) and the right robotic arm (4), as well as the hand motion and force signals collected by the left servo sensing glove (5) and the right servo sensing glove (6), to obtain optimized control commands for the left robotic arm (3), the right robotic arm (4), the left servo sensing glove (5), and the right servo sensing glove (6).

3. The training assist robot for cognitive rehabilitation training according to claim 1, characterized in that, The dual-head robotic arm support (2) adopts a triangular support structure.

4. The training assist robot for cognitive rehabilitation training according to claim 1, characterized in that, The dual-head robotic arm support (2) allows for quick disassembly and assembly, facilitating transportation and deployment in different treatment environments.

5. A training assistive robot for cognitive rehabilitation training according to claim 1, characterized in that, The left servo sensing glove (5) and the right servo sensing glove (6) are equipped with multiple sensors, including pressure sensors, bending sensors and tactile sensors, to accurately capture hand movements and force changes.

6. A training assist robot for cognitive rehabilitation training according to claim 5, characterized in that, The finger portions of the left servo sensing glove (5) and the right servo sensing glove (6) are made of flexible material to mimic the natural movement of the human hand, and the built-in sensors can detect and respond to the subtle movements of the fingers.

7. A training assistive robot for cognitive rehabilitation training according to claim 1, characterized in that, The joints of the left robotic arm (3) and the right robotic arm (4) are equipped with absolute encoders to monitor and control the position and posture of the left robotic arm (3) and the right robotic arm (4) in real time, so as to ensure the accuracy of operation.

8. A training assist robot for cognitive rehabilitation training according to claim 1, characterized in that, The auxiliary training control host (1) has an overload protection function. When the left robotic arm (3), right robotic arm (4), left servo sensor glove (5) or right servo sensor glove (6) is subjected to a force that exceeds the design range, the system will automatically disconnect the power supply to protect the equipment and user safety.

9. A training assist robot for cognitive rehabilitation training according to claim 1, characterized in that, The multimodal fusion neural network module uses existing fusion algorithms to take force signals, torque signals, and hand movement and force signals as inputs to the neural network. The neural network predicts the patient's movement intention and obtains corresponding auxiliary movement decisions. The auxiliary training control host (1) sends corresponding auxiliary movement instructions to the left robotic arm (3), right robotic arm (4), left servo sensing glove (5), and right servo sensing glove (6) according to the auxiliary movement decisions to assist the patient in completing the cognitive training task.

10. A training assistive robot for cognitive rehabilitation training according to claim 1, characterized in that, During cognitive rehabilitation training, the reinforcement learning robotic arm control module evaluates the completion of cognitive rehabilitation training tasks in real time, optimizes auxiliary movement decisions through existing algorithms, and thus optimizes the corresponding auxiliary movement commands to improve the patient's completion of cognitive training tasks.

Images