Model-free control method, system and device based on brain-like safety monitoring and medium

Abstract

The invention discloses a model-free control method, system and equipment based on brain-like safety monitoring and a medium, and the method comprises the steps: building a kinetic model of an n-dimensional mechanical arm and a camera, constructing a mixed vector field in a Cartesian space, collecting the obstacle avoidance data of a human under a dynamic obstacle environment, and carrying out the monitoring of a dynamic obstacle. The method comprises the following steps: establishing a BSIM model, establishing a BISM (brain-like safety monitoring) model by using a pulse neural network, generating a global reference trajectory based on a Tau-J theory, performing local correction on the global reference trajectory by using a virtual potential field (VPF) and an admittance relationship to obtain a real-time safety trajectory, designing a state-adaptive function (SAF), combining the SAF with the BSIM model, and establishing a BISM-MFC control rate; and proving the asymptotic stability of the BISM-MFC closed-loop system by using a Lyapunov stability theory. According to the method, the sensitivity of the system to the disturbance is reduced, the influence of the disturbance on the state trajectory is reduced at a relatively low calculation cost, and safe obstacle avoidance and trajectory planning in a dynamic obstacle environment are realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of robot control, in particular to a model-free control method and system based on brain-like safety monitoring, equipment and medium, which can be used for arm chapter control and trajectory tracking of robots, and is especially suitable for real-time decision-making and safety planning in dynamic obstacle environment. BACKGROUND

[0002] Robot systems are widely used in manufacturing, medical, logistics and exploration fields, but the uncertainty in dynamic environment (such as sensor error, system noise, environmental change and collision, etc.) brings challenges to safety control. Especially in the dynamic environment of unknown disturbance and obstacles, the robot needs to adjust the trajectory and behavior in real time to avoid collision and ensure task completion.

[0003] The ISO / TS15066 protocol emphasizes the importance of speed and separation monitoring (SSM) for safe operation, which ensures the safety of the robot by maintaining a minimum safety distance. However, the minimum safety distance of SSM needs to be manually set, which limits its adaptability and effectiveness in dynamic environments; Spiking neural network (SNN) is inspired by biological neurons and has advantages in processing time-sensitive and dynamic data. However, after intelligent decision-making, the trajectory needs to be re-planned to ensure stable and safe control of the robot; the admittance control method based on virtual potential field (VPF) realizes local trajectory planning through virtual force and admittance relationship, which ensures safe local trajectory planning under unknown disturbance. However, frequent trajectory changes and unknown disturbances may cause the robot system to oscillate and lose control; vector field technology is widely used in path tracking control due to its robustness and low computational complexity. The non-uniform vector field method does not need to consider the wind effect and can handle speed differences and angle problems. The hybrid vector field does not require prior knowledge of the path, but requires the trajectory time to be constant and known. Given that the robot trajectory is usually unknown and changes in real time, the vector field can be used to limit the convergence direction of the trajectory tracking error, which can effectively suppress external disturbances and has low computational cost. SUMMARY

[0004] To solve the above problems, the present application provides a model-free control method based on brain-like safety monitoring, system, equipment and medium, which reduces the sensitivity of the system to disturbance, reduces the influence of disturbance on state trajectory with low computational cost, and realizes safe obstacle avoidance and trajectory planning in dynamic obstacle environment.

[0005] As an aspect of the present application, a model-free control method based on brain-like safety monitoring is first provided, characterized by comprising the following steps:

[0006] Step S1: Establish the dynamics model of the n-dimensional robot arm and camera, and construct a hybrid vector field in Cartesian space;

[0007] Step S2: Collecting the obstacle avoidance data of human in dynamic obstacle environment, and constructing brain-inspired safety monitoring model BISM by using spiking neural network SNN;

[0008] Step S3: Generating global reference trajectory based on Tau-J theory, and locally correcting the global reference trajectory by using virtual potential field VPF and admittance relationship to obtain real-time safety trajectory, and calculating tracking error of end effector;

[0009] Step S4: Designing state attractor function SAF, and combining it with BSIM model to construct BISM-MFC control rate;

[0010] Step S5: Proving the asymptotic stability of BISM-MFC closed-loop system by using Lyapunov stability theory.

[0011] Preferably, the n-dimensional robot arm dynamics model in step S1 is:

[0012]

[0013] wherein, respectively represent the position, velocity and acceleration of the end effector, M x (x)∈R n×n , C x ∈R n ×n , G x ∈R n respectively represent the mass matrix, Coriolis matrix and centripetal matrix, and gravity vector, u∈R n is the control input, F e ∈R n is the external force, d x ∈R n is an unknown but bounded disturbance, Δ x ∈R n is a bounded uncertainty.

[0014] Preferably, by using the camera dynamics model, the position of the obstacle in the robot arm coordinate system can be obtained, and the specific conversion formula is:

[0015]

[0016] wherein, x o is the position of the obstacle in the robot arm coordinate system, x co is the position of the obstacle in the camera coordinate system, is the conversion matrix from the camera coordinate system to the robot arm coordinate system.

[0017] Preferably, the mixed vector field is:

[0018]

[0019] where Φ i (x v )∈R n denotes a sub-vector field, x v = [x v1 , x v2 ,..., x vi ,..., x vn ] T , x vi ∈R, x v is the coordinate point in the sub-vector field, k a , k b is a positive proportional coefficient, respectively used to control the intensity and direction of the vector field, Δ m is a proper small constant, is the position variable in the i-th dimension, I i is the unit vector in the i-th dimension.

[0020] Preferably, the SNN network in the step S2 uses a LIF neuron model, mainly composed of two fully connected layers and a LIF layer; the SNN network adopts a cross-entropy loss function, the optimizer is adaptive moment estimation, the learning rate is 0.01, the batch size is 32, and the training period is set to 5000 cycles. After training, the classification accuracy of the BISM model is more than 98%, and the average inference time of a single sample is only 7x10 -5 -4 seconds.

[0021] Preferably, the global reference trajectory in the step S3 is generated based on the Tau-J theory, and the specific formula is as follows:

[0022]

[0023] wherein, is the global reference trajectory, is the velocity obtained by taking the first derivative of the global reference trajectory, is the acceleration obtained by taking the second derivative of the global reference trajectory, X0=x s -x g , x s ∈R n is the starting position, is the target position, t g =t-t s , t s ∈R n is the starting time, T g =T-T s , T∈R is the target time, k t ∈(0, 0.5) is a kinematic constant that determines the motion.

[0024] Preferably, the non-contact external force generated by the virtual potential field in step S3 is:

[0025]

[0026] wherein, j is the jth obstacle, o j is the center position of the obstacle, r j is the radius of the boundary circle of the obstacle, k v is a positive constant that determines the amplitude of the force in the virtual potential field, k B is a positive constant that defines the threshold of the range of the potential field force, when the robot enters this range, the potential field force begins to act.

[0027] Preferably, the local correction of the global reference trajectory using the virtual potential field (VPF) and admittance relationship to obtain a real-time safety trajectory is:

[0028]

[0029] wherein, x r is the shaped reference trajectory, M d is the desired inertia, B d is the damping, k d is the stiffness matrix.

[0030] Preferably, the BISM model adjusts the desired trajectory to:

[0031]

[0032] wherein, t e is the BISM model output "time when the robot stops", ts is the start time of the output conversion from "robot stop" to "robot run".

[0033] Preferably, the tracking error of the end effector is:

[0034] e = x - x d (8)

[0035] wherein, x d ∈ R n denotes the desired trajectory and x d , are all bounded.

[0036] Preferably, the mathematical expression of the state attraction function (SAF) in step S4 is:

[0037]

[0038] wherein, e iThe state error in the i-th dimension has global convergence and boundedness;

[0039] Preferably, the control rate of the BISM-MFC method is:

[0040]

[0041] Wherein, k c is a constant, and the error is attracted The change rate of the tracking error directly reflects the dynamic deviation rate of the system to the trajectory tracking.

[0042] As a second aspect of the application, a model-free control system based on brain-like safety monitoring is provided, characterized in that the system comprises:

[0043] The brain-like safety monitoring model construction unit: by establishing the dynamics model of the n-dimensional robot arm and the camera, and constructing the mixed vector field in the Cartesian space; collecting the obstacle avoidance data of humans in the dynamic obstacle environment, and using the spiking neural network SNN to construct the brain-like safety monitoring model;

[0044] The BISM-MFC control rate construction unit: based on the Tau-J theory, a global reference trajectory is generated, and the global reference trajectory is locally corrected by using the virtual potential field VPF and the admittance relationship, to obtain a real-time safety trajectory, and the tracking error of the end effector is calculated; a state attraction function SAF is designed, and combined with the BISM model, to construct the BISM-MFC control rate;

[0045] The verification unit: the asymptotic stability of the BISM-MFC closed-loop system is proved by using the Lyapunov stability theory, so as to ensure that the tracking error and its derivative converge to a small neighborhood near the origin under unknown disturbance and bounded initial condition.

[0046] The third aspect of the application provides a computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to realize the steps of the method in any one of claims 1 to 7.

[0047] The fourth aspect of the application provides a computer device, which comprises a memory, a processor, and a computer program stored on the memory and executable on the processor, and the computer program is executed by the processor to realize the steps of the method in any one of claims 1 to 7.

[0048] Compared with the prior art, the application has the following beneficial effects:

[0049] (1) The model-free control method based on brain-like safety monitoring greatly reduces the tracking error and overshoot in the x direction, and is more easily convergent and returns to steady state after the disturbance disappears. Meanwhile, the time consumption per cycle is reduced by about 12%.

[0050] (2) The trajectory controlled by the method has smaller tracking error, and can be adjusted more quickly in a dynamic obstacle environment, and is more easily adapted to complex and variable environments.

[0051] (3) The method significantly reduces the influence of disturbance on the state trajectory at low computational cost, realizes high-precision trajectory tracking, millisecond-level real-time response, and strong robustness to unknown disturbances, and can better complete adaptive obstacle avoidance in a dynamic environment. BRIEF DESCRIPTION OF DRAWINGS

[0052] Figure 1 Flowchart of the model-free control method based on brain-like safety monitoring;

[0053] Figure 2 Trajectories of different initial states in the two-dimensional vector field Φ(x v ) in the embodiment;

[0054] Figure 3 Learning and cross-entropy loss curves of the SNN network in the embodiment, (a) is the learning curve; (b) is the cross-entropy loss curve;

[0055] Figure 4 Block diagram of the BISM-MFC closed-loop system in the embodiment;

[0056] Figure 5 Trajectories of different control methods in the embodiment;

[0057] Figure 6 Tracking error phase diagrams of different control methods in the embodiment, (a) is the x direction; (b) is the y direction;

[0058] Figure 7 Tracking errors of different control methods in the embodiment, (a) is the x direction; (b) is the y direction;

[0059] Figure 8 Control inputs of different control methods in the embodiment, (a) is the x direction; (b) is the y direction. DETAILED DESCRIPTION

[0060] The technology described below can have various modifications and can be various embodiments, and specific embodiments are described in detail herein with reference to the accompanying drawings. However, this does not mean that the technology described below is limited to specific embodiments. It should be understood that the present application includes all similar modifications, equivalents and alternatives without departing from the spirit and technical scope of the technology described below.

[0061] Figure 1 Flowchart of the model-free control method based on brain-like safety monitoring of the present application. As shown in Figure 1 the specific implementation of the method proposed by the present application is as follows:

[0062] Step S1: Establish the dynamics model of the n-dimensional robot arm and the camera, and construct the hybrid vector field in the Cartesian space;

[0063] The n-dimensional robot arm dynamics model in step S1 is:

[0064]

[0065] wherein, respectively represent the position, velocity and acceleration of the end effector, M x (x)∈R n×n , C x ∈R n ×n , G x ∈R n respectively represent the mass matrix, the Coriolis matrix and the centripetal matrix, the gravity vector, u∈R n is the control input, F e ∈R n is the external force, d x ∈R n is an unknown but bounded disturbance, Δ x ∈R n is a bounded uncertainty.

[0066] Preferably, by using the camera dynamics model, the position of the obstacle in the robot arm coordinate system can be obtained, and the specific conversion formula is:

[0067]

[0068] wherein, x o is the position of the obstacle in the robot arm coordinate, x co is the position of the obstacle in the camera coordinate system, is the conversion matrix from the camera coordinate system to the robot arm coordinate system.

[0069] Preferably, the hybrid vector field is:

[0070]

[0071] where Φ i (x v )∈R n represents a sub-vector field, x v = [x v1 , x v2 ,..., x vi ,..., x vn ] T , x vi ∈R, x v is the coordinate point in the sub-vector field, k a , k b is a positive proportional coefficient, respectively used to control the vector field intensity and direction, Δ m is a properly small constant, is the position variable in the i-th dimension, I i is the unit vector in the i-th dimension. See Figure 2 , which shows the trajectories of different initial states in a two-dimensional vector field.

[0072] After establishing the model of the uncertain robot system, the specific parameter configuration is as follows:

[0073] G x = J(q) -T G(q), Δ x = J(q) -T Δ, d x = J(q) -T d, M x = J(q) -T M(q)J(q) -1 , J(q) is the Jacobian matrix, d x is the external disturbance in the joint space, Δ x is the internal uncertainty.

[0074] where:

[0075]

[0076] where g = 9.8 N / kg, l1 and l2 are 1 meter, k S = cos(q1)cos(q2) + sin(q1) + sin(q2), k C = cos(q1)sin(q2) - sin(q1)cos(q2), M1 and M2 are both 5 kg.

[0077] Step S2: Collecting human obstacle avoidance data in a dynamic obstacle environment, and constructing a brain-inspired safety monitoring (BISM) model using a spiking neural network (SNN);

[0078] Preferably, in step S2, a BMIS model containing LIF neurons is built. Four Vicon cameras and marker points are used to collect data of human motion in a dynamic obstacle environment. 20 participants are selected, and 20000 data samples are collected and divided into training set, validation set and test set according to the ratio of 7:1:3. The SNN network uses a LIF neuron model, which mainly consists of two fully connected layers and a LIF layer. The input dimension of the fully connected layer is 10, the output dimension is 15, the input dimension of the LIF layer is 10, the output dimension is 15, the input dimension of the fully connected layer is 15, and the output dimension is 2. The leakage rate of the LIF model is 0.9, the optimizer is adaptive moment estimation, the initial learning rate is 0.01, the batch size is 32, the epoch number is 5000, the classification loss is cross-entropy loss, and the classification output is "robot stop" and "robot run". After training, the classification accuracy of the BISM model is more than 98%, and the average inference time of a single sample is only 7×10 -5 seconds. Referring to Figure 3 , the learning and cross-entropy loss curves of the SNN network are shown, (a) is the learning curve; (b) is the cross-entropy loss curve.

[0079] Step S3: Generating a global reference trajectory based on the Tau-J theory, and locally correcting the global reference trajectory using a virtual potential field (VPF) and admittance relationship to obtain a real-time safety trajectory, and calculating the tracking error of the end effector;

[0080] Preferably, the global trajectory is generated based on the Tau-J theory, and the generated trajectory is locally planned using the admittance relationship based on the globally convergent and bounded virtual potential field, which specifically includes:

[0081] Generating a global reference trajectory:

[0082]

[0083] wherein, is the global reference trajectory, is the velocity obtained by taking the first derivative of the global reference trajectory, is the acceleration obtained by taking the second derivative of the global reference trajectory, X0=x s -x g , x s ∈R n is the starting position, t g = t - t s t s ∈ R n T g = T - T s T t ∈ R is the target time, k j ∈ (0, 0.5) is the kinematic constant that determines the motion.

[0084] The resulting contact-free external force of each obstacle after the superposition of the virtual potential field is:

[0085]

[0086] where, j is the jth obstacle, o j is the center position of the obstacle, r v is the radius of the obstacle boundary circle, k B is a positive constant that defines the threshold of the range of the potential field force, when the robot enters this range, the potential field force begins to act.

[0087] Then, the trajectory is shaped through the admittance relationship:

[0088]

[0089] where, x r is the corrected reference trajectory, M d is the desired inertia, B d is the damping, k d is the stiffness matrix.

[0090] Finally, the selection of the desired trajectory by the robot arm is:

[0091]

[0092] where, t e is the output of the BISM model “time when the robot stops”, ts is the start time of the output conversion from “robot stop” to “robot run”.

[0093] Calculate the tracking error of the end effector

[0094] e = x - x d (9)

[0095] where, x d ∈ R n represents the desired trajectory and x d , are all bounded.

[0096] Step S4: Design the state-attracted function (SAF) and combine it with the BSIM model to construct the BSIM-MFC control law;

[0097] See Figure 4 The block diagram of the BISM-MFC closed-loop system is shown. Preferably, the mathematical expression of the state attraction function SAF in step S4 is:

[0098]

[0099] Among them, e i Let be the state error in the i-th dimension, which has global convergence and boundedness;

[0100] The control rate of the BISM-MFC method is:

[0101]

[0102] Where, k c For positive constants, attracting error The meaning of the tracking error change rate is to directly reflect the dynamic deviation rate of the system in tracking the trajectory.

[0103] Construct the BISM-MFC control law according to the above formula, and configure the following parameters: k a k b k c The values ​​are set to 220, 10000, and 2200 respectively; the total simulation time is 10s, the policy step size is 0.001s; the initial state is x(0) = [-1,1]. T and Three obstacles are set up, with force constraint k. b =0.3, and the obstacle positions are o1=[0,0] T o2 = [0,1] T o3 = [0, -1] T The values ​​are 0.3, 0.5, and 0.2 respectively.

[0104] Uncertainty

[0105] External disturbance d x Select as:

[0106]

[0107] in, The computational cost is calculated based on the total time t of the robot system's 10,000 cycles. T measure.

[0108] Step S5: The asymptotic stability of the BISM-MFC (brain-inspired safety monitoring based model-free control) closed-loop system is proved by Lyapunov stability theory, which ensures that the tracking error and its derivative converge to a small neighborhood around the origin under unknown disturbances and bounded initial conditions.

[0109] Preferably, the stability proof of the BISM-MFC closed-loop system includes:

[0110] A Lyapunov function is selected:

[0111]

[0112] Regarding t to V E Take the differential:

[0113]

[0114] Where Y=k c +Λ m , Λ m is the minimum eigenvalue of the matrix Since Φ(e),

[0115] Δ x , d x are bounded. The appropriate parameter k c can be selected to make Y>0, and by Young's inequality, we know:

[0116]

[0117] Where α∈(0,1) is a constant. Select the κ class function ρ(·) as:

[0118]

[0119] When , we have:

[0120]

[0121] Therefore, the attraction error E can be expressed as:

[0122]

[0123] Where β E is a κL class function, and γ E (·) is a κ class function.

[0124] Therefore, it is proved that the attraction error E is constrained and guided to a small neighborhood of the origin, and the consistency between the convergence direction of the tracking error and the direction of the vector field is ensured. In addition, by selecting a larger parameter k c , the bound of the attraction error E becomes smaller.

[0125] Another Lyapunov function is selected as:

[0126]

[0127] The differential of V e with respect to t is:

[0128]

[0129] And:

[0130]

[0131] By selecting appropriate parameters, the following can be obtained:

[0132]

[0133] That is:

[0134]

[0135] Therefore, the boundary of the tracking error can be expressed as:

[0136]

[0137] eand converge to a small neighborhood close to the origin. And by selecting a larger parameter k a and k b , the bounds of eand can be further reduced.

[0138] Referring to Figure 5 , the trajectory generated by the BISM-MFC method is obviously smoother and reaches the target point more accurately, while the trajectories controlled by the NNC and PID methods have greater fluctuations and errors. Moreover, the BISM-MFC can effectively adjust the trajectory to avoid obstacles, while the NNC and PID methods exhibit obvious hysteresis and instability when adjusting the trajectory.

[0139] Referring to Figure 6 , the error phase diagram of the BISM-MFC method shows that its tracking error quickly converges to a small neighborhood near the origin, indicating that the system is stable and responds quickly. In contrast, the error fluctuations of the NNC and PID methods are larger, the convergence speed is slower, and there is an obvious divergence trend in some stages.

[0140] Referring toFigure 7 The BISM-MFC method has the minimum tracking error amplitude in both x and y directions, and maintains a low error level throughout the whole process. The NNC method has the second minimum error, while the PID method has the maximum error, especially in the dynamic change stage, and the error peak of the PID method is much higher than that of the other methods.

[0141] Referring to Figure 8 The control input of the BISM-MFC method changes relatively smoothly without large fluctuations, indicating that the control strategy is efficient and has small disturbance to the system. The control input of the NNC method fluctuates obviously, and the control input of the PID method fluctuates most dramatically, which is easy to cause system oscillation.

[0142] Although the present application has been described in detail with general description and specific embodiments above, some modifications or improvements can be made on the basis of the present application. The above description is only the preferred embodiments of the present application, and does not limit the patent scope of the present application. Other changes and modifications made by those skilled in the art without departing from the spirit and protection scope of the present application are still included in the protection scope of the present application.

Claims

1. A model-free control method based on neuromorphic safety monitoring, characterized in that, Includes the following steps: Step S1: Establish a dynamic model of the n-dimensional robotic arm and camera, and construct a hybrid vector field in Cartesian space; Step S2: Collect obstacle avoidance data of humans in dynamic obstacle environments, and use spiking neural networks (SNNs) to construct a brain-like safety monitoring model (BISM). Step S3: Generate a global reference trajectory based on Tau-J theory, and use the relationship between the virtual potential field (VPF) and admittance to locally correct the global reference trajectory to obtain a real-time safe trajectory, and calculate the tracking error of the end effector; Step S4: Design the State Attraction Function (SAF) and combine it with the BISM model to construct the BISM-MFC control law; Step S5: Prove the asymptotic stability of the BISM-MFC closed-loop system using Lyapunov stability theory.

2. The model-free control method based on neuromorphic safety monitoring according to claim 1, characterized in that, The n-dimensional robotic arm dynamics model in step S1 is as follows: in, M represents the position, velocity, and acceleration of the end effector, respectively. x (x)∈R n×n C x ∈R n×n G x ∈R n Let u ∈ R represent the mass matrix, Coriolis matrix, centripetal matrix, and gravity vector, respectively. n To control the input, F e ∈R n For external force, d x ∈R n For an unknown but bounded perturbation, Δ x ∈R n This is bounded uncertainty; Using the camera dynamics model, the position of the obstacle in the robotic arm coordinate system can be obtained. The specific transformation formula is as follows: Where, x o Let x be the position of the obstacle in the robot arm's coordinate system. co The position of the obstacle in the camera coordinate system. This is the transformation matrix from the camera coordinate system to the robot arm coordinate system; The hybrid vector field constructed in the Cartesian space is: Where, Φ i (x v )∈R n Denotes a sub-vector field, x v =[x v1 ,x v2 ,...x vi ,...,x vn ] T ,x vi ∈R,x v For the sub-vector field at coordinate point k a k b These are proportionality coefficients, used to control the vector field intensity and direction, respectively. m For a reasonably small constant, Let I be the position variable in the i-th dimension. i Let be the unit vector in the i-th dimension.

3. The model-free control method based on neuromorphic safety monitoring according to claim 1, characterized in that, In step S2, the SNN network uses a LIF neuron model, mainly composed of two fully connected layers and one LIF layer. The SNN network employs a cross-entropy loss function, an adaptive moment estimator (AME) optimizer, a learning rate of 0.01, a batch size of 32, and a training epoch of 5000 epochs. After training, the BISM model achieves a classification accuracy exceeding 98%, with an average inference time of 7 × 10⁻⁶ per sample. -5 Second.

4. The model-free control method based on neuromorphic safety monitoring according to claim 1, characterized in that, In step S3, a global reference trajectory is generated based on Tau-J theory, and the specific formula is as follows: in, For global reference trajectory, The velocity obtained by taking the first derivative of the global reference trajectory The acceleration obtained by taking the second derivative of the global reference trajectory is X0 = x s -x g x s ∈R n Starting position For the target position, t g =tt s ,t s ∈R n T is the start time. g =TT s T∈R is the target time, k t ∈(0,0.5) are the kinematic constants that determine the motion.

5. The model-free control method based on neuromorphic safety monitoring according to claim 1, characterized in that, The non-contact external force generated by the virtual potential field in step S3 is: in, j is the j-th obstacle, o j r is the center position of the obstacle. j Let k be the radius of the boundary circle of the obstacle. v k is a positive constant used to determine the boundary of a force, and also a positive constant to determine the magnitude of the force. B A positive constant defines the threshold range of the potential force. The method of using the virtual potential field (VPF) and admittance relationship to locally correct the global reference trajectory to obtain a real-time safe trajectory specifically involves: Where, x r For the corrected reference trajectory, M d For the expected inertia, B d For damping, k d Here is the stiffness matrix.

6. The model-free control method based on neuromorphic safety monitoring according to claim 1, characterized in that, The BISM model will adjust the desired trajectory as follows: Among them, t e Output "Robot Stop Time" for the BISM model, where ts is the start time of the output transition from "Robot Stop" to "Robot Start". The tracking error of the end effector is: e=x-x d (8) Where, x d ∈R n Describe the desired trajectory and x d , Both are bounded.

7. The model-free control method based on neuromorphic safety monitoring according to claim 1, characterized in that, The mathematical expression for the state attraction function SAF in step S4 is: Among them, e i Let be the state error in the i-th dimension, which has global convergence and boundedness; The control rate of the BISM-MFC method is: Where, k c For positive constants, attracting error To track the rate of error change, reflecting the dynamic deviation rate of the system in tracking the trajectory.

8. A model-free control system based on neuromorphic safety monitoring, characterized in that, The system includes: Brain-like safety monitoring model construction unit: Establish a dynamic model of an n-dimensional robotic arm and camera, and construct a hybrid vector field in Cartesian space; collect obstacle avoidance data of humans in dynamic obstacle environments, and use spiking neural networks (SNN) to construct a brain-like safety monitoring model; BISM-MFC control law building unit: Generates a global reference trajectory based on Tau-J theory, and uses the relationship between the virtual potential field (VPF) and admittance to locally correct the global reference trajectory to obtain a real-time safe trajectory, and calculates the tracking error of the end effector; designs the state attraction function (SAF) and combines it with the BSIM model to construct the BSIM-MFC control law; Verification Unit: Using Lyapunov stability theory, the asymptotic stability of the BISM-MFC closed-loop system is proved, ensuring that under unknown perturbation and bounded initial conditions, the tracking error and its derivative converge to a small neighborhood near the origin.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.

Images