Bioinspired learning systems based on human-computer collaborative training and their application methods

Abstract

This invention discloses a bio-inspired learning system based on human-machine collaborative training and its usage method. The bio-inspired learning system includes a robot learning unit and a bio-inspired teaching unit. Embodiments of this invention enable the robot to acquire human motion data in real time, analyze and extract key features through deep learning algorithms, thereby quickly adapting to different movement patterns. During machine learning, dynamic interaction between humans and machines is achieved. The robot can not only observe human actions but also adjust its own movement strategies based on feedback, achieving more natural and fluid collaboration. Through human-machine collaborative learning, the robot can perform more accurate imitation and learning, significantly improving its motor skill learning efficiency and reducing sample dependence and sample quality problems in traditional learning methods.

Description

Technical Field

[0001] The embodiments of the present invention relate to the field of robotics technology, and in particular to a bio-inspired learning system based on human-machine collaborative training and its usage method. Background Technology

[0002] With the rapid development of artificial intelligence and robotics, intelligent robots are increasingly being applied in numerous fields, especially in the service industry, medical assistance, and education. To enable robots to interact better with humans and enhance their autonomous learning and adaptability, researchers have begun exploring how to improve robots' motor skills and decision-making abilities by observing human bio-inspired behavior. Existing research primarily focuses on using sensors, vision systems, and machine learning algorithms to analyze human actions so that robots can imitate or learn these actions.

[0003] Furthermore, traditional machine learning-related motion learning methods often rely on static datasets and procedural action sequences, lacking real-time feedback and dynamic adaptation capabilities during training. Additionally, existing robots typically struggle to effectively recognize and understand diverse human movements in complex environments, resulting in low learning efficiency and insufficient adaptability. Summary of the Invention

[0004] This invention provides a bio-inspired learning system based on human-machine collaborative training and its usage method, which enables robots to gradually improve their own motor abilities by observing and analyzing human actions in real time, so as to better simulate human motor characteristics. It has the advantages of high learning efficiency and strong adaptability.

[0005] In a first aspect, embodiments of the present invention provide a bio-inspired learning system based on human-computer collaborative training, comprising:

[0006] The robot learning unit includes a robot perception module, a robot motion execution and balance control module, a robot information integration and transmission module, a robot motion learning module, and robot limb end pressure sensors. The robot perception module is located in the robot's head and various limbs. The robot motion execution and balance control module is used to drive the robot's limb movement. The robot information integration and transmission module and the robot motion learning module are located in the robot's processor. The robot limb end pressure sensors are distributed at the ends of the robot's limbs.

[0007] The bio-inspired teaching unit includes a human motion information acquisition module, a human limb end-pressure acquisition module, a human information integration and transmission module, and a robot status feedback module. The human motion information acquisition module is used to collect information on changes in skeletal points during human movement by a human instructor. The human limb end-pressure acquisition module is used to collect multi-point pressure information on both feet during human movement by a human instructor. The skeletal point change information and the multi-point pressure information on both feet are transmitted to the robot motion learning module and the robot motion execution and balance control module via the human information integration and transmission module to generate a robot following gait. The robot following gait is fed back to the human instructor through the robot status feedback module.

[0008] In one embodiment, the robot perception module is used to acquire perception information, which includes optical perception information, ground three-dimensional point cloud perception information, and limb end-effector pressure perception information. The optical perception information and the ground three-dimensional point cloud perception information are collected from the robot's first-person perspective, and the three-dimensional point cloud is mapped to the environmental coordinate system.

[0009] In one embodiment, the robot state feedback module is used to provide state information, which includes learning accuracy numerical information, feedback guidance information, and robot scene environment information.

[0010] In one embodiment, the logic for generating the feedback guidance information is as follows: if the range of motion of the human instructor's previous gait exceeds the robot's movement threshold, the instructor is prompted to adjust their gait to within the range of motion of the robot; if the consistency of the human instructor's gait is below the threshold, the instructor is prompted to conduct bio-inspired teaching with a more consistent gait; if the consistency of the human instructor's gait exceeds the threshold, the instructor is prompted to move with differentiated gait parameters.

[0011] In one embodiment, the human motion information acquisition module employs multiple visual depth sensors placed in the scene, and the multiple visual depth sensors, combined with a skeletal point analysis algorithm, reconstruct multiple skeletal motion sequences of the human teacher.

[0012] In one embodiment, the robot's actual motion state is based on the robot's own zero-point torque Z, and the gait M generated by the robot's motion learning module. robot If restrictions are imposed, the robot's real-time motion state is M. physical =f(M robot ,Z), where f is the boundary function that ensures the robot does not fall.

[0013] Secondly, embodiments of the present invention provide a method for using a bio-inspired learning system based on human-computer collaborative training, applied to the bio-inspired learning system based on human-computer collaborative training as described in the first aspect, the method comprising:

[0014] Step 1: Set up the robot learning unit and the bio-inspired teaching unit in areas A and B, respectively, which are the same size and have the same environmental configuration;

[0015] Step 2: A human instructor configures the motion names of the robot's learning units;

[0016] Step 3: The human instructor performs an action in area B;

[0017] Step 4: The human motion information acquisition module collects information on changes in skeletal points during human motion, and the human limb end pressure acquisition module collects multi-point pressure information on both feet during human motion. The skeletal point change information and the multi-point pressure information on both feet are transmitted to the robot motion learning module via the human information integration and transmission module. The robot motion learning module, in conjunction with the motion execution and balance control module, generates the robot's following gait.

[0018] Step 5: The robot's gait is fed back to the human instructor via the robot information integration and transmission module;

[0019] Step 6: The human instructor performs the next action in area B and repeats steps 4 and 5;

[0020] Step 7: The robot saves the training data collected in steps 4 to 7, and trains the robot's motion learning module based on the training data;

[0021] Step 8: Based on the subsequent actions of the human instructor, repeat steps 3 through 7.

[0022] In one embodiment, the human instructor performs an action in region B, including:

[0023] When a human instructor performs an action in region B, the human motion information acquisition module and the human limb end pressure acquisition module collect time series t1 to t2. n Human motion data, real-time environmental data, and plantar pressure data during the period, wherein the real-time environmental data is represented by a three-dimensional point cloud E∈R T×P×3 The representation is as follows: T is the number of time sampling points, and P is the number of point clouds; the human motion data includes a sequence of skeletal points moving over time, and the motion of all bones is represented by a multidimensional matrix M. human ∈R T×J×KThe representation is as follows: T is the number of time sampling points, J is the number of skeletal points, and K is the spatial description for each skeletal point in space. The spatial description is adjusted according to the features of the machine learning algorithm to the three-dimensional coordinates of the head-to-end of the skeletal point, or the coordinates of one end of the skeletal point plus the quaternion pose of the skeletal point in three-dimensional space. The plantar pressure data uses a multidimensional matrix P∈p. T×S The expression is given, where T is the number of time sampling points and S is the number of foot sensors.

[0024] In one embodiment, the method of use further includes:

[0025] The 3D point cloud matrix E, the multidimensional matrix M, and the multidimensional matrix P are used as training information and synthesized into a tensor input to the robot's machine learning model. During training, the human first repeats the action c times (c>3). Then, based on the repeated actions, the robot generates its own skeletal point motion reference sequence M in conjunction with the network model. robot ∈R T×J×K The self-skeleton point motion reference sequence generates a rotation matrix sequence J for each joint using a geometric inverse kinematics algorithm. This self-skeleton point motion reference sequence is input into the robot motion execution and balance control module. The robot's actual motion state is based on the robot's own zero-point torque Z and the gait M generated by machine learning. robot If restrictions are imposed, the robot's real-time motion state is M. physical =f(M robot ,Z), where f is the boundary function to ensure the robot does not fall. The robot motion execution and balance control module first adjusts the joints based on the rotation matrix sequence J, combined with its torque position or model prediction, to generate the robot's physical motion gait M. physical ; with M human With M physical The difference between the parameters is used as an evaluation function to assess the robot's motion process, and the robot's model parameters are trained through gradient optimization.

[0026] In one embodiment, the method of use further includes:

[0027] The robot generates feedback information, which includes a scalarized M. human With M physical The difference between them, and the judgment of M human Whether the skeletal pose exceeds the robot physics motion threshold; if it does, the feedback information suggests limiting the human movement; and determining M during different training sessions of the same movement. human If the consistency is lower than a preset value, random motion prompt information is generated in the feedback information.

[0028] This invention includes: a bio-inspired learning system based on human-machine collaborative training and its usage method. The bio-inspired learning system includes a robot learning unit and a bio-inspired teaching unit. The robot learning unit includes a robot perception module, a robot motion execution and balance control module, a robot information integration and transmission module, a robot motion learning module, and robot limb end pressure sensors. The robot perception module is located in the robot's head and various limbs. The robot motion execution and balance control module drives the robot's limb movement. The robot information integration and transmission module and the robot motion learning module are located within the robot's processor. The robot limb end pressure sensors are distributed throughout the robot. The robot's limbs and extremities; the bio-inspired teaching unit includes a human motion information acquisition module, a human limb pressure acquisition module, a human information integration and transmission module, and a robot state feedback module. The human motion information acquisition module collects information on skeletal changes during human movement by a human instructor, and the human limb pressure acquisition module collects multi-point pressure information from both feet during human movement by a human instructor. The skeletal change information and multi-point pressure information from both feet are transmitted to the robot motion learning module and the robot motion execution and balance control module via the human information integration and transmission module to generate the robot's following gait. The robot's following gait is then fed back to the human instructor via the robot state feedback module. Based on this, the embodiments of the present invention enable the robot to gradually improve its own motor ability by observing and analyzing human movements in real time, so as to better simulate human movement characteristics. It has the advantages of high learning efficiency and strong adaptability.

[0029] Other features and advantages of the invention will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures particularly pointed out in the description, claims, and drawings. Attached Figure Description

[0030] The accompanying drawings are provided to further understand the technical solutions of the present invention and constitute a part of the specification. They are used together with the embodiments of the present invention to explain the technical solutions of the present invention, and do not constitute a limitation on the technical solutions of the present invention.

[0031] Figure 1 A physical distribution block diagram of a bio-inspired learning system based on human-computer collaborative training, provided as an embodiment of the present invention;

[0032] Figure 2 A flowchart illustrating the usage method of a bio-inspired learning system based on human-computer collaborative training, as provided in one embodiment of the present invention;

[0033] Figure 3This is a schematic diagram illustrating a feasible example of robot sensor logical position configuration under the training platform of this invention, provided as an embodiment of the invention.

[0034] Figure 4 This is a schematic diagram illustrating a feasible robot leg joint configuration example under the training platform of the present invention, as provided in one embodiment of the present invention. Detailed Implementation

[0035] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0036] It should be understood that in the description of the embodiments of the present invention, "multiple" (or "amounts") means two or more, "greater than," "less than," and "exceeding" are understood to exclude the stated number, while "above," "below," and "within" are understood to include the stated number. If "first," "second," etc., are used in the description, they are only for the purpose of distinguishing technical features and should not be construed as indicating or implying relative importance, or implicitly indicating the number of indicated technical features, or implicitly indicating the order of the indicated technical features.

[0037] With the rapid development of artificial intelligence and robotics, intelligent robots are increasingly being applied in numerous fields, especially in the service industry, medical assistance, and education. To enable robots to interact better with humans and enhance their autonomous learning and adaptability, researchers have begun exploring how to improve robots' motor skills and decision-making abilities by observing human bio-inspired behavior. Existing research primarily focuses on using sensors, vision systems, and machine learning algorithms to analyze human actions so that robots can imitate or learn these actions.

[0038] Furthermore, traditional machine learning-related motion learning methods often rely on static datasets and procedural action sequences, lacking real-time feedback and dynamic adaptation capabilities during training. Additionally, existing robots typically struggle to effectively recognize and understand diverse human movements in complex environments, resulting in low learning efficiency and insufficient adaptability.

[0039] To address the aforementioned problems in the existing technology, this invention provides a bio-inspired learning system based on human-machine collaborative training and its usage method. The bio-inspired learning system includes a robot learning unit and a bio-inspired teaching unit. The robot learning unit includes a robot perception module, a robot motion execution and balance control module, a robot information integration and transmission module, a robot motion learning module, and robot limb end-effector pressure sensors. The robot perception module is located in the robot's head and various limbs. The robot motion execution and balance control module drives the robot's limb movement. The robot information integration and transmission module and the robot motion learning module are located within the robot's processor. The robot limb end-effector... Pressure sensors are distributed at the extremities of the robot's limbs. The bio-inspired teaching unit includes a human motion information acquisition module, a human limb pressure acquisition module, a human information integration and transmission module, and a robot state feedback module. The human motion information acquisition module collects information on skeletal changes during human movement, while the human limb pressure acquisition module collects multi-point pressure information from both feet. The skeletal change information and multi-point pressure information are transmitted via the human information integration and transmission module to the robot motion learning module and the robot motion execution and balance control module, generating a robot following gait. This gait is then fed back to the human teacher via the robot state feedback module. Based on this, the embodiments of the present invention enable the robot to gradually improve its motor abilities by observing and analyzing human movements in real time, thus better simulating human movement characteristics. It has the advantages of high learning efficiency and strong adaptability.

[0040] like Figure 1 As shown, Figure 1 This is a physical distribution block diagram of a bio-inspired learning system based on human-computer collaborative training, provided in one embodiment of the present invention.

[0041] 1-A represents the robot learning unit, which may include:

[0042] 1-A1 is the robot perception module, which is distributed in a fixed scene, the robot's head and various limbs in the form of optical and pressure sensors;

[0043] 1-A2 is the robot motion and balance control module, used to drive the robot's limb movements;

[0044] 1-A3 is the robot information integration and transmission module, which is placed in the robot processor in software form;

[0045] 1-A4 is the robot motion learning module, which is placed in the robot processor in the form of an algorithm;

[0046] 1-A5 are pressure sensors at the ends of the robot's limbs, distributed at the ends of each limb.

[0047] 1-B is a biology-inspired teaching unit, which may include:

[0048] 1-B1 is for human educators;

[0049] 1-B2 is a human motion information acquisition module, and multiple three-dimensional sensing devices placed in a fixed area of ​​the scene;

[0050] 1-B3 is a pressure acquisition module for human extremities, which is distributed in the form of a sensor array on the palms and soles of the human teacher;

[0051] 1-B4 is the robot status feedback module, which uses VR / AR devices to provide feedback on the environment, robot status, and prompts.

[0052] The human body information integration and transmission module (not shown in the figure) can transmit information on changes in skeletal points and pressure information at multiple points on both feet to the robot's motion learning module.

[0053] Understandably, the robot perception module 1-A1 is used to acquire perception information, which includes optical perception information, ground 3D point cloud perception information, and limb end-effector pressure perception information. Among them, the optical perception information and ground 3D point cloud perception information are collected from the robot's first-person perspective, and the 3D point cloud is mapped to the environmental coordinate system.

[0054] It is understandable that the human limb end pressure acquisition module 1-B3 is characterized by the following: the pressure sensor used in the human limb end pressure acquisition module 1-B3 has multiple pressure detection points, and the pressure sensor used by the robot perception module 1-A1 to acquire limb end pressure perception information has the same size, pressure detection point position and sampling rate as the pressure sensor used in the human limb end pressure acquisition module.

[0055] It is understandable that the robot's state feedback has three attributes: 1) Learning accuracy value, which is fed back to the human instructor by the display screen, reflecting the degree of fit between the robot's current learning-based motion state and the human instructor's teaching motion state; 2) Feedback guidance information: This information is fed back to the human instructor by the display screen, including guidance information for the instructor such as "increase / decrease movement speed", "increase / decrease leg height", and "try to keep the same as the previous movement". This guidance information is generated by the robot's motion learning module 1-A4; 3) Robot scene environment information, which is obtained by digitally modeling the robot's environment or by collecting data from a fixed camera placed in the scene.

[0056] Understandably, the logic for generating feedback guidance information is as follows: if the range of motion of the human instructor's previous gait exceeds the robot's movement threshold, the instructor is prompted to adjust their gait to within the robot's range of motion. If the consistency of the human instructor's gait is below the threshold, the instructor is prompted to use a more consistent gait for bio-inspired teaching (i.e., maintain consistency in each teaching movement as much as possible). If the consistency of the human instructor's gait exceeds the threshold, the instructor is prompted to use differentiated gait parameters.

[0057] Understandably, the human motion information acquisition module 1-B2 is characterized by using multiple visual depth sensors placed in the scene, combined with a skeletal point analysis algorithm to reconstruct multiple skeletal motion sequences of a human teacher.

[0058] Understandably, the robot motion and balance control module 1-A2 is characterized by the following: the robot's actual motion state is based on the robot's own zero-point torque Z, and the gait M generated by the robot motion learning module 1-A4. robot If restrictions are imposed, the robot's real-time motion state is M. physical =f(M robot ,Z), where f is the boundary function that ensures the robot does not fall.

[0059] Where E represents the boundary range of the robot's supporting legs.

[0060] Understandably, before a human instructor lifts or lowers their leg, the robot's status feedback module outputs teaching prompts, and the human instructor adjusts their movements based on these prompts, gait type, and the actual environment.

[0061] like Figure 2 As shown, Figure 2 This is a flowchart illustrating the usage method of a bio-inspired learning system based on human-computer collaborative training, according to an embodiment of the present invention. The usage method is as follows:

[0062] Step 1: Set up the robot learning unit and the bio-inspired teaching unit in areas A and B, respectively, which are the same size and have the same environmental configuration;

[0063] Step 2: Human instructors configure the movement names of the robot's learning units;

[0064] Step 3: The human instructor performs an action in area B;

[0065] Step 4: The human motion information acquisition module collects information on changes in skeletal points during human motion, and the human limb end pressure acquisition module collects multi-point pressure information on both feet during human motion. The skeletal point change information and multi-point pressure information on both feet are transmitted to the robot motion learning module via the human information integration and transmission module. The robot motion learning module, in conjunction with the motion execution and balance control module, generates the robot's following gait.

[0066] Step 5: The robot follows the movement gait and transmits the information back to the human instructor via the robot information integration and transmission module;

[0067] Step 6: The human instructor performs the next action in area B and repeats steps 4 and 5;

[0068] Step 7: The robot saves the training data collected in steps 4 to 7, and trains the robot's motion learning module based on the training data;

[0069] Step 8: Based on the subsequent actions of the human instructor, repeat steps 3 through 7.

[0070] like Figure 3 As shown, Figure 3 This is a feasible example of robot sensor logical position configuration under the training platform described in this invention, wherein: 3-1 is the robot head optical sensor, 3-2 is the robot internal posture sensor, 3-3 is the robot limb end pressure sensor, 3-4 is the multi-degree-of-freedom robot upper limb, and 3-5 is the multi-degree-of-freedom robot lower limb.

[0071] Figure 4 This is a feasible example of robot leg joint configuration under the training platform described in this invention, wherein 4-1 to 4-6 are the robot's 6 degrees of freedom actuators, and 4-7 is the robot's foot pressure sensor array.

[0072] The usage method of this training platform is as follows:

[0073] 1. Divide the area into two units of the same size and with the same environment. The robot learning unit (2-A) is for placing robots, and the bio-inspired teaching unit (2-B) is for human educators to participate in activities.

[0074] 2. Before learning begins, the human instructor and the robot maintain the same posture (any initial posture is acceptable). First, the human instructor defines a movement name (e.g., "single leg lift") and inputs the gait into the robot. Then, the human and the robot begin bio-inspired teaching.

[0075] 3. When a human performs a single action, the human motion information acquisition module and the human plantar pressure acquisition module collect time series t1~t2. nHuman motion data and real-time environmental data during the process. Environmental data can be obtained through 3D point cloud E∈R. T×P×3 The representation is as follows: T represents the number of time-series sampling points, and P represents the number of point clouds. Human motion data includes a sequence of skeletal points moving over time; the motion of all bones can be represented by a multidimensional matrix M. human ∈R T×J×K The description is as follows: T represents the number of time sampling points, J represents the number of skeletal points, and K represents the spatial description of each skeletal point segment in space. This description can be adjusted based on machine learning algorithm features to represent the 3D coordinates of the head-to-end of the skeletal segment, or the coordinates of one end of the skeletal segment plus the quaternion pose of the skeletal segment in 3D space. Plantar pressure data can be represented using a multidimensional matrix P∈p. T×S The description is provided, where T is the number of time sampling points and S is the number of plantar sensors.

[0076] 4. The M, P, and E matrices are used as training information and synthesized as tensors, which are then input into the robot's machine learning model. The machine learning model can be built using frameworks such as Deep Neural Networks (DNNs), Long Short-Term Memory Networks (LSTMs) combined with Generative Adversarial Networks (GANs). During training, the human first repeats the action c times (c>3), and then the robot generates its own skeletal point motion reference sequence M based on the repeated actions and the network model. robot ∈R T×J×K The sequence generates a rotation matrix sequence J for each joint using a geometric inverse attitude algorithm (e.g., spatial matrix transformation combined with the law of cosines). This sequence is then input into the robot's balance control module. The balance control module can employ algorithms such as the Zero-Point Moment Algorithm (ZMP) and Model Prediction (MPC). The robot's actual motion state is based on its own zero-point moment Z and the gait M generated by machine learning. robot If restrictions are imposed, the robot's real-time motion state is M. physical =f(M robot The equation is: f(J, Z), where f is the boundary function to prevent the robot from falling. The balance control module first adjusts the joints based on J, combined with its torque position or model predictions, to generate the robot's physical gait M. physical .

[0077] 5. With M human With M physical The differences between the parameters are used as an evaluation function to assess the robot's motion process. The robot's model parameters are trained using gradient optimization. The gradient optimization algorithm can be adjusted according to the specific model.

[0078] 6. The robot generates feedback information, which includes a scalarized M. human With M physical The difference between them can be calculated using norm(M) as one method. human -Mphysical Simultaneously determine: 1) M human Does the skeletal pose exceed the robot's physical motion threshold? If it does, the feedback message will suggest limiting the human motion. 2) In different training sessions of the same movement, M human Consistency. If the consistency is lower than the preset value, random exercise prompts will be generated in the feedback information (such as "increase / decrease exercise speed", "increase / decrease leg height"), otherwise the prompt will be "try to keep the same as the previous exercise".

[0079] 7. Repeat steps 3-5.

[0080] The following is an example of an implementation of the present invention:

[0081] Both the robot learning unit (1-A) and the teaching inspiration unit (1-B) are 5m*5m flat areas with green plastic flooring.

[0082] The robot's motion is driven by a bus driver with angle feedback.

[0083] At the algorithm and software levels, the balance control algorithm can employ a zero-point torque algorithm. The robot's machine learning model can be implemented using Long Short-Term Memory / Adversarial Networks (LSTM / GAN). Information integration and transmission are achieved using Wi-Fi as the physical layer medium. The balance algorithm, information integration and transmission software, and balance control algorithm are all processed internally within the robot in software form. Intel processors can be selected.

[0084] At the sensor level, robot perception and human motion acquisition modules can be implemented using point cloud-vision sensors, with the realSense-D435i being a feasible model. For end-effector pressure sensors, carbon resistive film sensor arrays can be used. Robot status feedback can be achieved using various wearable devices such as HoloLens.

[0085] Regarding the application of bio-inspired learning systems based on human-computer collaborative training, one implementation example is as follows:

[0086] 1) The robot has 24 degrees of freedom, of which the two arms have a total of 12 degrees of freedom and the two legs have a total of 12 degrees of freedom.

[0087] 2) Human educators define the type of movement as "squatting".

[0088] 3) A human instructor repeats the squatting motion 10 times. The platform generates a human skeletal motion sequence based on AlphaPose and point cloud clustering, forming a training set.

[0089] 4) The robot uses the training set and the adversarial network (GAN) model to generate a feasible "squat-up" motion sequence.

[0090] 5) The human instructor performs one squat.

[0091] 6) The robot's internal GAN ​​algorithm combines the motion data to generate a set of motion sequences.

[0092] 7) The robot uses the Zero Moment of Motion (ZMP) algorithm to adjust the generated motion sequence and generate a physical motion sequence.

[0093] 8) The physical motion sequence generates 24 (number of joints) * 3 (target position, target velocity, maximum load torque) dimensional joint information through the inverse kinematics algorithm (IK), and inputs it into the robot joint driver through the CAN bus protocol. The driver will drive the robot to repeat the action once.

[0094] 9) The robot compares the differences between human motion and machine physical motion through difference calculation, takes the difference as the evaluation value, and uses the gradient descent method to optimize the model parameters.

[0095] 10) Repeat steps 4-9.

[0096] Based on this, the present invention proposes a bio-inspired learning system based on human-machine collaborative training and its usage method. By acquiring human motion sequences, the robot can gradually observe and learn human actions in a real-time environment. The present invention has at least the following significant advantages:

[0097] Real-time learning and interactive mimicry: This system enables robots to acquire human motion data in real time, analyze and extract key features through deep learning algorithms, and thus quickly adapt to different movement patterns.

[0098] Dynamic interaction: In the process of machine learning, dynamic interaction between humans and machines is realized. The robot can not only observe human actions, but also adjust its own movement strategy according to the feedback, so as to achieve more natural and smooth collaboration.

[0099] Improved learning efficiency: Through human-machine collaborative learning, robots can perform more accurate imitation and learning, significantly improving the learning efficiency of their motor skills and reducing sample dependence and sample quality problems in traditional learning methods.

[0100] The above provides a detailed description of the preferred embodiments of the present invention. However, the present invention is not limited to the above embodiments. Those skilled in the art can make various equivalent modifications or substitutions without departing from the spirit of the present invention. All such equivalent modifications or substitutions are included within the scope defined by the claims of the present invention.

Claims

1. A bio-inspired learning system based on human-computer collaborative training, characterized in that, include: The robot learning unit includes a robot perception module, a robot motion execution and balance control module, a robot information integration and transmission module, a robot motion learning module, and robot limb end pressure sensors. The robot perception module is located in the robot's head and various limbs. The robot motion execution and balance control module is used to drive the robot's limb movement. The robot information integration and transmission module and the robot motion learning module are located in the robot's processor. The robot limb end pressure sensors are distributed at the ends of the robot's limbs. The bio-inspired teaching unit includes a human motion information acquisition module, a human limb end-pressure acquisition module, a human information integration and transmission module, and a robot status feedback module. The human motion information acquisition module is used to collect information on changes in skeletal points during human movement by a human instructor. The human limb end-pressure acquisition module is used to collect multi-point pressure information on both feet during human movement by a human instructor. The skeletal point change information and the multi-point pressure information on both feet are transmitted to the robot motion learning module and the robot motion execution and balance control module via the human information integration and transmission module to generate a robot following gait. The robot following gait is fed back to the human instructor through the robot status feedback module.

2. The bio-inspired learning system based on human-machine collaborative training according to claim 1, characterized in that, The robot perception module is used to acquire perception information, which includes optical perception information, ground three-dimensional point cloud perception information, and limb end-effector pressure perception information. The optical perception information and the ground three-dimensional point cloud perception information are collected from the robot's first-person perspective, and the three-dimensional point cloud is mapped to the environmental coordinate system.

3. The bio-inspired learning system based on human-computer collaborative training according to claim 1, characterized in that, The robot state feedback module is used to provide state information, which includes learning accuracy numerical information, feedback guidance information, and robot scene environment information.

4. The bio-inspired learning system based on human-machine collaborative training according to claim 3, characterized in that, The logic for generating the feedback guidance information is as follows: if the range of motion of the human instructor's previous gait exceeds the robot's movement threshold, the instructor is prompted to adjust their gait to within the range of motion of the robot; if the consistency of the human instructor's gait is below the threshold, the instructor is prompted to conduct bio-inspired teaching with a more consistent gait. If the gait consistency of human instructors exceeds a threshold, prompts are given to guide instructors to move with differentiated gait parameters.

5. The bio-inspired learning system based on human-machine collaborative training according to claim 1, characterized in that, The human motion information acquisition module uses multiple visual depth sensors placed in the scene. The multiple visual depth sensors, combined with the skeletal point analysis algorithm, reconstruct multiple skeletal motion sequences of the human teacher.

6. The bio-inspired learning system based on human-machine collaborative training according to claim 1, characterized in that, The robot's actual motion state is based on the robot's own zero-point torque Z and the gait M generated by the robot's motion learning module. robot If restrictions are imposed, the robot's real-time motion state is M. physical =f(M robot ,Z), where f is the boundary function that ensures the robot does not fall.

7. A method for using a bio-inspired learning system based on human-computer collaborative training, characterized in that, Applied to the bio-inspired learning system based on human-machine collaborative training as described in any one of claims 1 to 6, the method of use includes: Step 1: Set up the robot learning unit and the bio-inspired teaching unit in areas A and B, respectively, which are the same size and have the same environmental configuration; Step 2: A human instructor configures the motion names of the robot's learning units; Step 3: The human instructor performs an action in area B; Step 4: The human motion information acquisition module collects information on changes in skeletal points during human motion, and the human limb end pressure acquisition module collects multi-point pressure information on both feet during human motion. The skeletal point change information and the multi-point pressure information on both feet are transmitted to the robot motion learning module via the human information integration and transmission module. The robot motion learning module, in conjunction with the motion execution and balance control module, generates the robot's following gait. Step 5: The robot's gait is fed back to the human instructor via the robot information integration and transmission module; Step 6: The human instructor performs the next action in area B and repeats steps 4 and 5; Step 7: The robot saves the training data collected in steps 4 to 7, and trains the robot's motion learning module based on the training data; Step 8: Based on the subsequent actions of the human instructor, repeat steps 3 through 7.

8. The method of use according to claim 7, characterized in that, The human instructor performs an action in area B, including: When a human instructor performs an action in region B, the human motion information acquisition module and the human limb end pressure acquisition module collect time series t1 to t2. n Human motion data, real-time environmental data, and plantar pressure data during the period, wherein the real-time environmental data is represented by a three-dimensional point cloud E∈R T×P×3 The representation is as follows: T is the number of time sampling points, and P is the number of point clouds; the human motion data includes a sequence of skeletal points moving over time, and the motion of all bones is represented by a multidimensional matrix M. human ∈R T ×J×K The representation is as follows: T is the number of time sampling points, J is the number of skeletal points, and K is the spatial description for each skeletal point in space. This spatial description is adjusted based on machine learning algorithm features to represent the 3D coordinates of the head and tail of the skeletal segment, or the coordinates of one end of the skeletal segment plus the quaternion pose of the skeletal segment in 3D space. The plantar pressure data uses a multidimensional matrix P∈p. T×S The expression is given, where T is the number of time sampling points and S is the number of foot sensors.

9. The method of use according to claim 8, characterized in that, The method of use also includes: The 3D point cloud matrix E, the multidimensional matrix M, and the multidimensional matrix P are used as training information and synthesized into a tensor input to the robot's machine learning model. During training, the human first repeats the action c times (c>3). Then, based on the repeated actions, the robot generates its own skeletal point motion reference sequence M in conjunction with the network model. robot ∈R T×J×K The self-skeleton point motion reference sequence generates a rotation matrix sequence J for each joint using a geometric inverse kinematics algorithm. This self-skeleton point motion reference sequence is input into the robot motion execution and balance control module. The robot's actual motion state is based on the robot's own zero-point torque Z and the gait M generated by machine learning. robot If restrictions are imposed, the robot's real-time motion state is M. physical =f(M robot ,Z), where f is the boundary function to ensure the robot does not fall. The robot motion execution and balance control module first adjusts the joints based on the rotation matrix sequence J, combined with its torque position or model prediction, to generate the robot's physical motion gait M. physical ; with M human With M physical The difference between the parameters is used as an evaluation function to assess the robot's motion process, and the robot's model parameters are trained through gradient optimization.

10. The method of use according to claim 9, characterized in that, The method of use also includes: The robot generates feedback information, which includes a scalarized M. human With M physical The difference between them, and the judgment of M human Whether the skeletal pose exceeds the robot physics motion threshold; if it does, the feedback information suggests limiting the human movement; and determining M during different training sessions of the same movement. human If the consistency is lower than a preset value, random motion prompt information is generated in the feedback information.

Images