Mechanical arm brain-like control method and device based on pulse neural network

Abstract

The invention relates to the technical field of mechanical arm control, in particular to a mechanical arm brain-like control method and device based on a spiking neural network, and the method comprises the steps that a cerebellum module is constructed based on the spiking neural network, and an initial torque instruction of a target mechanical arm is obtained through the cerebellum module; constructing a thalamus module based on the initial moment instruction, and adjusting the initial moment instruction by using the thalamus module to obtain a final moment instruction; constructing a brainstem module by using the final moment instruction to obtain a feed-forward moment; and a spinal cord module is constructed based on the pulse codes and the feed-forward torque, and a mechanical arm control model is constructed in combination with the cerebellum module, the thalamus module and the brainstem module so as to output a torque control instruction. Therefore, the problems that in the related technology, in a complex working scene, due to the influences of the size, the quality and the like of an operation object, the working requirements of the mechanical arm are remarkably different when the mechanical arm executes different tasks, the control precision is reduced, the universality is damaged, and the mechanical arm cannot adapt to the control tasks under the complex load condition are solved.

Description

Technical Field

[0001] The present application relates to the technical field of robotic arm control, and in particular to a method and device for brain-like control of a robotic arm based on a pulse neural network. Background Art

[0002] As an automatic control device that imitates the functions of human arms and can complete various operations, the robotic arm has been gradually applied to various industrial fields since the 1980s, such as assembly, painting and welding. With the development of drive and material technology, the degree of freedom of the robotic arm has gradually increased while the cost has continued to decrease, and it can complete more sophisticated operations. In recent years, it has gradually expanded to medical care, services, and life entertainment. However, compared with creatures in nature, the most advanced robotic arms cannot be compared with the operating capabilities of even ants, and it is even more difficult to achieve the dexterity of human arms.

[0003] In the related technology, with the development of brain science and computing hardware, humans have gradually gained a clearer understanding of biological, especially human physiological structure and function. For example, an adaptive control framework based on ANN (Artificial Neural Network) can be used, which uses neural networks to imitate human learning ability to a certain extent, providing a solution for controlling robotic arms in complex application environments. In addition, the neurons of SNN (Spiking Neural Network) transmit information between neurons through discrete pulses, and can use cell membrane potential to store time-space information, so it has a natural advantage in robotic arm control. Therefore, SNN and the human central nervous system can be combined to control the robotic arm, and SNN has a strong ability to adapt to the environment.

[0004] However, in related technologies, in complex work scenarios, due to the influence of the size and mass of the operated object, the working requirements of the robotic arm when performing different tasks may vary significantly, thereby damaging the versatility of the ANN-based adaptive control algorithm. The SNN-based algorithm has low control accuracy, poor dynamic performance during the control process, relies on the dynamic model for gravity compensation, and cannot adapt well to control tasks under various complex load conditions, and is in urgent need of improvement. Summary of the invention

[0005] The present application provides a brain-like control method and device for a robotic arm based on a pulse neural network to solve the problem in the related art that in complex work scenarios, due to the influence of the size and mass of the operation object, the work requirements of the robotic arm when performing different tasks may have significant differences, thereby damaging the versatility of the ANN-based adaptive control algorithm, while the SNN-based algorithm has low control accuracy, poor dynamic performance during the control process, relies on a dynamic model for gravity compensation, and cannot adapt well to control tasks under various complex load conditions.

[0006] The first aspect of the present application provides a brain-like control method for a robotic arm based on a pulse neural network, which is applied to a model construction stage, wherein the method comprises the following steps: constructing a cerebellum module of the robotic arm control model based on the pulse neural network, and using the cerebellum module to obtain an initial torque instruction of a target robotic arm; constructing a thalamus module of the robotic arm control model based on the initial torque instruction, and using the thalamus module to adjust the initial torque instruction to obtain a final torque instruction corresponding to the initial torque instruction; constructing a brainstem module of the robotic arm control model using the final torque instruction to obtain a feedforward torque of the target robotic arm; constructing a spinal cord module of the robotic arm control model based on pulse coding and the feedforward torque, and constructing the robotic arm control model in combination with the spinal cord module, the cerebellum module, the thalamus module and the brainstem module, so as to output a torque control instruction of the target robotic arm using the robotic arm control model.

[0007] Optionally, in one embodiment of the present application, the cerebellum module for constructing a robotic arm control model based on a pulse neural network includes: constructing an initial cerebellum module of the robotic arm control model according to the pulse neural network and non-excited pulse neurons; training the initial cerebellum module through the oblique circular trajectory, planar figure eight trajectory, planar circular trajectory, different gravity compensation modes and MSE (Mean Squared Error) of the target robotic arm to obtain the cerebellum module.

[0008] Optionally, in one embodiment of the present application, the thalamus module of the robotic arm control model constructed based on the initial torque instruction includes: constructing the initial thalamus module of the robotic arm control model based on a fully connected pulse neural network and neurons; based on the initial torque instruction, training the initial thalamus module through a first loss function to obtain the thalamus module.

[0009] Optionally, in one embodiment of the present application, the brainstem module of the robotic arm control model constructed using the final torque instruction includes: establishing a first part of the brainstem module based on a pulse neural network, a pulse neuron and a second loss function; constructing a second part of the brainstem module based on a single-layer neural network and the final torque instruction, so as to obtain the brainstem module using the first part and the second part.

[0010] Optionally, in one embodiment of the present application, the spinal cord module that constructs the robotic arm control model based on pulse coding and the feedforward torque includes: obtaining the speed information pulse string and the position information pulse string of the target robotic arm through nerve fibers and pulse coding based on the planned trajectory and the actual trajectory; and constructing the spinal cord module according to the speed information pulse string, the position information pulse string, the feedforward torque and neurons.

[0011] The second aspect of the present application provides a brain-like control method for a robotic arm based on a pulse neural network, which is applied to a model application stage, wherein the method includes the following steps: obtaining control instructions for a robotic arm in a working state; inputting the control instructions into a pre-built robotic arm control model to output a torque control instruction for the robotic arm, wherein the pre-built robotic arm control model is composed of at least one of a cerebellum module, a thalamus module, a brainstem module and a spinal cord module; and controlling the robotic arm to output a corresponding torque based on the torque control instruction.

[0012] The third aspect of the present application provides a brain-like control device for a robotic arm based on a pulse neural network, which is applied to a model construction stage, wherein the device includes: a first construction module, which is used to construct a cerebellum module of the robotic arm control model based on the pulse neural network, so as to obtain the initial torque instruction of the target robotic arm using the cerebellum module; a second construction module, which is used to construct a thalamus module of the robotic arm control model based on the initial torque instruction, and use the thalamus module to adjust the initial torque instruction to obtain the final torque instruction corresponding to the initial torque instruction; a third construction module, which is used to construct a brainstem module of the robotic arm control model using the final torque instruction to obtain the feedforward torque of the target robotic arm; a fourth construction module, which is used to construct a spinal cord module of the robotic arm control model based on pulse coding and the feedforward torque, and construct the robotic arm control model in combination with the spinal cord module, the cerebellum module, the thalamus module and the brainstem module, so as to output the torque control instruction of the target robotic arm using the robotic arm control model.

[0013] Optionally, in one embodiment of the present application, the first building module includes: a first building unit, used to construct an initial cerebellum module of the robotic arm control model based on the pulse neural network and non-excited pulse neurons; a first generation unit, used to train the initial cerebellum module through the oblique circular trajectory, planar figure eight trajectory, planar circular trajectory, different gravity compensation modes and MSE of the target robotic arm to obtain the cerebellum module.

[0014] Optionally, in one embodiment of the present application, the second construction module includes: a second construction unit, used to construct an initial thalamus module of the robotic arm control model based on a fully connected pulse neural network and neurons; a second generation unit, used to train the initial thalamus module through a first loss function based on the initial torque instruction to obtain the thalamus module.

[0015] Optionally, in one embodiment of the present application, the third construction module includes: a third construction unit, used to establish the first part of the brainstem module based on the pulse neural network, the pulse neuron and the second loss function; a third generation unit, used to construct the second part of the brainstem module based on the single-layer neural network and the final torque instruction, so as to obtain the brainstem module using the first part and the second part.

[0016] Optionally, in one embodiment of the present application, the fourth building module includes: a fourth generation unit, used to obtain the speed information pulse string and the position information pulse string of the target robotic arm based on the planned trajectory and the actual trajectory through nerve fibers and the pulse coding; a fourth construction unit, which constructs the spinal cord module according to the speed information pulse string, the position information pulse string, the feedforward torque and neurons.

[0017] The fourth aspect of the present application provides a brain-like control device for a robotic arm based on a pulse neural network, which is applied to the model application stage, wherein the device includes: an acquisition module, used to acquire control instructions for a robotic arm in a working state; an input module, used to input the control instructions into a pre-built robotic arm control model to output a torque control instruction for the robotic arm, wherein the pre-built robotic arm control model is composed of at least one of a cerebellum module, a thalamus module, a brainstem module and a spinal cord module; an output module, used to control the robotic arm to output a corresponding torque based on the torque control instruction.

[0018] The fifth aspect of the present application provides an electronic device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the brain-like control method of a robotic arm based on a pulse neural network as described in the above embodiment.

[0019] The sixth aspect of the present application provides a computer-readable storage medium, which stores a computer program. When the program is executed by a processor, it implements the above-mentioned brain-like control method of a robotic arm based on a pulse neural network.

[0020] The seventh aspect of the present application provides a computer program product, including a computer program, which, when executed, implements the above-mentioned brain-like control method of a robotic arm based on a pulse neural network.

[0021] The embodiment of the present application can construct a cerebellum module based on a pulse neural network, a thalamus module based on an initial torque instruction, a brainstem module using a final torque instruction, a spinal cord module based on a feedforward torque, and then a mechanical arm control model is constructed by combining the spinal cord module, the cerebellum module, the thalamus module and the brainstem module, so as to output the torque control instruction of the target mechanical arm using the mechanical arm control model, and solve the problem of the mechanical arm control algorithm not being versatile by constructing a mechanical arm control model, so that it has the ability to cope with complex loads. Thus, the problem in the related art that in complex working scenes, due to the influence of the size and mass of the operating object, the working requirements of the mechanical arm when performing different tasks may be significantly different, thereby damaging the versatility of the adaptive control algorithm based on ANN, and the algorithm based on SNN has low control accuracy, poor dynamic performance during the control process, relies on the dynamic model for gravity compensation, and cannot adapt well to the control tasks under various complex load conditions.

[0022] Additional aspects and advantages of the present application will be given in part in the description below, and in part will become apparent from the description below, or will be learned through the practice of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] The above and / or additional aspects and advantages of the present application will become apparent and easily understood from the following description of the embodiments in conjunction with the accompanying drawings, in which:

[0024] Figure 1 A flowchart of a brain-like control method for a robotic arm based on a spiking neural network according to an embodiment of the present application;

[0025] Figure 2 A schematic diagram of a framework of a robotic arm control model constructed according to an embodiment of the present application;

[0026] Figure 3 A schematic block diagram of a brain-like control device for a robotic arm based on a pulse neural network according to an embodiment of the present application;

[0027] Figure 4 This is a flowchart of a brain-like control method for a robotic arm based on a pulse neural network according to another embodiment of the present application.

[0028] Figure 5 A schematic diagram of a curve showing the variation of joint position errors over time during the operation of a robot arm according to an embodiment of the present application;

[0029] Figure 6 A schematic diagram of a curve showing a change in weight provided by a thalamus module according to an embodiment of the present application as a function of load mass;

[0030] Figure 7 The k of each joint in the working process of the robot arm provided according to one embodiment of the present application p and k d Schematic diagram of the curve changing with time;

[0031] Figure 8 A schematic diagram of a curve showing changes in joint errors with load mass in three trajectory tracking tasks provided by an embodiment of the present application, using a PD algorithm with an accurate dynamics model and an embodiment of the present application;

[0032] Fig. 9 A schematic diagram of a curve showing changes in joint error with load mass in three trajectory tracking tasks provided according to an embodiment of the present application, using a PD algorithm without an accurate dynamics model and an embodiment of the present application;

[0033] Fig.10 A schematic block diagram of trajectory curves of the workspace and joint space in a three-dimensional figure eight trajectory tracking task with a 2kg load provided according to an embodiment of the present application;

[0034] Fig.11 A schematic block diagram of a brain-like control device for a robotic arm based on a pulse neural network according to another embodiment of the present application;

[0035] Fig.12 It is a schematic diagram of the structure of an electronic device provided according to an embodiment of the present application. DETAILED DESCRIPTION

[0036] Embodiments of the present application are described in detail below, and examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present application, and should not be construed as limiting the present application.

[0037] The following describes the brain-like control method and device of a manipulator based on a pulse neural network in an embodiment of the present application with reference to the accompanying drawings. In view of the fact that in complex working scenes mentioned in the above background technology, due to the influence of the size and mass of the operating object, the working requirements of the manipulator when performing different tasks may have significant differences, thereby damaging the versatility of the adaptive control algorithm based on ANN, and the control accuracy of the algorithm based on SNN is not high, the dynamic performance is poor during the control process, and it relies on the dynamic model for gravity compensation, and cannot adapt well to the control tasks under various complex load conditions. The present application provides a brain-like control method of a manipulator based on a pulse neural network, in which a cerebellum module can be constructed based on a pulse neural network, a thalamus module can be constructed based on an initial torque instruction, a brainstem module can be constructed using a final torque instruction, and a spinal cord module can be constructed according to a feedforward torque, and then a manipulator control model can be constructed by combining the spinal cord module, the cerebellum module, the thalamus module and the brainstem module, so that the manipulator control model is used to output the torque control instruction of the target manipulator, and the manipulator control algorithm is not widely used by constructing a manipulator control model, so that it has the ability to cope with complex loads. As a result, the problem in the related technology is solved that in complex work scenarios, due to the influence of the size and mass of the operation object, the working requirements of the robot arm when performing different tasks may have significantly different, thereby damaging the versatility of the ANN-based adaptive control algorithm, while the SNN-based algorithm has low control accuracy, poor dynamic performance during the control process, relies on the dynamic model for gravity compensation, and cannot adapt well to control tasks under various complex load conditions.

[0038] Before introducing the method proposed in the embodiment of the present application, the relevant concepts involved in the embodiment of the present application are first explained.

[0039] Spiking Neurons and Spiking Neural Networks

[0040] In SNN, neurons implement information flow through pulse trains, which can be represented by, but not limited to:

[0041] S(t)=∑ f δ(t f -t0),

[0042] Where f = 1, 2, ... is the pulse number, δ(·) is the Dirac function, t f is the pulse emission time, and t0 is the initial time. The input current signal generated by a series of pulses acting on the neuron can be expressed as, but not limited to:

[0043]

[0044] Among them, τ s is the time constant

[0045] In the related art, many spiking neuron models are proposed, among which the LIF (Leaky Integrate and Fire) model can not only describe the dynamic process of cell membrane potential accumulation and leakage, but also maintain a high computational efficiency. Therefore, the embodiment of the present application focuses on the LIF model. The neuron input current causes the change of the neuron membrane potential u, and the relationship satisfied can be expressed as but not limited to:

[0046]

[0047] Among them, τ m is the time constant describing the potential leakage, τ m =RC, where R is the cell membrane resistance, C is the cell membrane capacitance, u reset is the potential of the neuron after reset, i0(t) is the external current driving the neuron state, i j (t) is the input current of the jth synapse, w j is the strength of the jth synapse, and u(t) is the potential value of the cell membrane at the current moment.

[0048] Furthermore, in the embodiment of the present application, when the cell membrane potential exceeds the excitation threshold u fire After that, the neuron sends a pulse, and the relationship it satisfies can be expressed as but not limited to:

[0049] s(t)=Θ(u(t)-u fire ),

[0050] Where s(t) is the pulse train function and Θ(x) is the excitation function. The expression of Θ(x) can be, but is not limited to, expressed as:

[0051]

[0052] Then the cell membrane potential is reset to u reset .

[0053] Furthermore, the neural network constructed using pulse neurons in the embodiment of the present application is called SNN, and the important issue of the SNN in the embodiment of the present application is how to train. Since the pulses between neurons are not differentiable, directly calculating the gradient may cause network instability. In the related technology, ANN can be used for training based on back propagation. During back propagation, the gradient of the approximate function is used instead of the gradient of the pulse. This method can easily migrate the relevant framework of ANN to SNN for network training.

[0054] In addition, the application of SNN also faces the problem of encoding and decoding. There are two main common encoding methods: one is Poisson encoding; the other is constant current input encoding. Among them, Poisson encoding is to encode the input data into a pulse sequence whose distribution of the number of times of emission conforms to the Poisson process; constant current input encoding is to design a network of fully connected pulse neurons as an encoder. There are two main decoding methods: one is to use the pulse emission frequency of the neurons in the network output layer within the time window; the other is to set the threshold of the neurons in the network output layer to infinity so that they do not emit pulses, and use the membrane potential of the neurons at the last moment as the output value.

[0055] Specifically, Figure 1 The present invention provides a flowchart of a brain-like control method of a robotic arm based on a pulse neural network according to an embodiment of the present application.

[0056] like Figure 1 As shown, the brain-like control method of the robotic arm based on the pulse neural network is applied to the model building stage, wherein the method includes the following steps:

[0057] In step S101, a cerebellum module of a robot arm control model is constructed based on a pulse neural network, so as to obtain an initial torque command of a target robot arm using the cerebellum module.

[0058] Through the above analysis, it can be seen that the embodiment of the present application can construct a cerebellum module of the robotic arm control model based on the pulse neural network, and then use the cerebellum module to obtain the initial torque command of the target robotic arm.

[0059] Optionally, in one embodiment of the present application, a cerebellar module of a robotic arm control model is constructed based on a pulse neural network, including: an initial cerebellar module of a robotic arm control model is constructed based on a pulse neural network and non-excited pulse neurons; the initial cerebellar module is trained through the oblique circular trajectory, planar figure eight trajectory, planar circular trajectory, different gravity compensation modes and MSE of the target robotic arm to obtain a cerebellar module.

[0060] As a possible implementation method, the embodiment of the present application can construct an initial cerebellum module of the robotic arm control model based on the pulse neural network and non-excited pulse neurons, and train the initial cerebellum module through the oblique circular trajectory, planar figure eight trajectory, planar circular trajectory, different gravity compensation modes and MSE of the target robotic arm, so as to obtain the cerebellum module.

[0061] Exemplarily, in the embodiment of the present application, under the condition determined by the mechanical arm dynamics model, the gravity compensation torque is actually a nonlinear mapping from joint position to joint torque. In the field of machine learning, the task of fitting observed data into a curve through supervised learning is called a regression task, and there are already a large number of studies and mature methods for regression tasks in related fields. Among them, the training data used for training in the embodiment of the present application can be generated by the mechanical arm dynamics model, and the demand for gravity compensation can be met through the regression task when the training data can cover the workspace.

[0062] Furthermore, if Figure 2 As shown, the initial cerebellum module constructed in the embodiment of the present application is shown in the dotted box of the initial cerebellum module. It is a four-layer recurrent pulse neural network. The actual position of the joint It is connected to the 300 pulse neurons of the second layer in a fully connected manner. Then, the embodiment of the present application is connected to the 200 neurons of the third layer in a similar fully connected manner. Among them, in the embodiment of the present application, the neurons of the third layer not only receive the information of the neurons of the second layer, but also receive the information of the second layer at the previous moment. Finally, the output of the third layer is output to the 140 non-excited pulse neurons of the fourth layer in a fully connected manner. Among them, in the embodiment of the present application, the neurons of the fourth layer are averaged in groups of 20, thereby outputting 7 torques.

[0063] Furthermore, the embodiment of the present application may be, but is not limited to, trained using oblique circular trajectories, planar figure eight trajectories, and planar circular trajectories, and using MSE as the loss function, which is not specifically limited in the present application. The expression of the oblique circular trajectory may be, but is not limited to, expressed as:

[0064]

[0065] The expression of the plane figure eight trajectory can be expressed as but not limited to:

[0066]

[0067] The expression of the plane circular trajectory can be, but is not limited to, expressed as:

[0068]

[0069] Among them, R c =R e =0.14, T c =3s, T e =4s, θ0=30°, and each trajectory was trained for 50 cycles.

[0070] It should be noted that the embodiment of the present application notes that although humans do not need an accurate dynamic model when operating with their arms, they can correct their movement patterns by sensing the weight of objects. Therefore, the embodiment of the present application constructs two gravity compensation modes, light (no load) and heavy (3kg load), in the cerebellum module and trains them separately to obtain two sets of SNNs. During the working process, these two sets of SNNs can receive the actual joint position q a , and give the initial torque command, which can be but not limited to: τ1,

[0071] In step S102, a thalamus module of the robot arm control model is constructed based on the initial torque command, and the thalamus module is used to adjust the initial torque command to obtain a final torque command corresponding to the initial torque command.

[0072] In some embodiments, the embodiments of the present application can construct a thalamus module of the robotic arm control model based on the initial torque command, and then adjust the initial torque command to obtain the final torque command.

[0073] Optionally, in one embodiment of the present application, a thalamus module of a robotic arm control model is constructed based on an initial torque instruction, including: constructing an initial thalamus module of a robotic arm control model based on a fully connected pulse neural network and neurons; based on the initial torque instruction, training the initial thalamus module through a first loss function to obtain a thalamus module.

[0074] In some embodiments, the embodiments of the present application can construct an initial thalamus module based on a fully connected network and neurons, and train the initial thalamus module according to an initial torque instruction and a first loss function to obtain a thalamus module.

[0075] Exemplarily, the embodiment of the present application can use the force sensor data at the end of the robotic arm to enable the initial thalamus module to "feel" the weight of an object like a human, thereby assigning a weight to the initial torque instructions given by the two gravity compensation modes of the cerebellum module through reinforcement learning to cope with loads of different masses and shapes. Among them, in the embodiment of the present application, the initial thalamus module can be a three-layer fully connected network, and the input of the 6-dimensional force sensor is fully connected in a hidden layer of 5 neurons and an output layer of 2 neurons to obtain two weights. Among them, the initial thalamus module constructed in the embodiment of the present application can be as follows Figure 2 shown.

[0076] Furthermore, the embodiment of the present application trains the initial thalamus module based on the initial torque instruction through the first loss function to obtain the thalamus module. The expression of the first loss function can be but is not limited to:

[0077]

[0078] Among them, q d,j ,q a,j , They are the planned joint position, actual joint position, planned joint velocity and actual joint velocity, j=1,…,7.

[0079] In step S103, the brainstem module of the robot control model is constructed using the final torque command to obtain the feedforward torque of the target robot.

[0080] In the actual execution process, the embodiment of the present application can use the final torque instruction to construct a brainstem module to obtain the feedforward torque of the target robotic arm.

[0081] Optionally, in one embodiment of the present application, a brainstem module of a robotic arm control model is constructed using the final torque instruction, including: establishing a first part of the brainstem module based on a pulse neural network, a pulse neuron and a second loss function; constructing a second part of the brainstem module based on a single-layer neural network and the final torque instruction, so as to obtain the brainstem module using the first part and the second part.

[0082] In some embodiments, the brainstem module of the present application embodiment can be functionally divided into two independent parts, as shown in the schematic diagram. Figure 2 As shown, the first part can be regarded as a three-layer spiking neural network trained by reinforcement learning. In the embodiment of the present application, the input of the brainstem module is a vector consisting of the planned position, speed and actual position and speed of each joint. This input is fully connected to the hidden layer of 10 spiking neurons, which then output a 14-element vector in a fully connected manner. Represents the k of each joint p Parameters and k d Parameters. During training, the second loss function can be used as the loss function, and its expression can be, but is not limited to, expressed as:

[0083]

[0084] Among them, q d,j ,q a,j , τ j They are the planned joint position, actual joint position, planned joint velocity, actual joint velocity and output torque of the spinal module, j=1,…,7.

[0085] It can be seen that the second loss function of the embodiment of the present application mainly includes three items, which respectively represent minimizing the position error, minimizing the speed error and reducing the torque as much as possible.

[0086] The second part can be regarded as a simple single-layer neural network, which consists of 7 non-exciting pulse neurons, which receive the torque from the thalamus module, that is, the charge of the cell membrane potential at each moment can be expressed as but not limited to:

[0087] δu(t)=w1τ1+w2τ2,

[0088] At this time, in the embodiment of the present application, the corresponding cell membrane potential can be expressed as but not limited to:

[0089]

[0090] Where T is the time constant.

[0091] Then, the embodiment of the present application can realize the smoothing and filtering of the final torque command through the timing characteristics of the cell membrane potential itself, and take τ g (t) = u(t) is output to the spinal cord module as the feedforward torque at each moment.

[0092] Optionally, in one embodiment of the present application, pulse coding and feedforward torque are used to construct a spinal cord module of a robotic arm control model, including: obtaining a speed information pulse string and a position information pulse string of a target robotic arm through nerve fibers and pulse coding based on a planned trajectory and an actual trajectory; and constructing a spinal cord module according to the speed information pulse string, the position information pulse string, the feedforward torque and neurons.

[0093] In some embodiments, the trajectory planned by the embodiment of the present application and the actual body feedback can be encoded into a series of discrete pulse sequences after entering the spinal module. Each joint is independent of each other, and the planned trajectory q of each joint d,j ,q a,j ,j=1,…,7 plays a role in activating input nerve fibers, such as Figure 2 The actual body feedback is shown as j=1,…,7 plays an inhibitory role, such as Figure 2 As shown by the dotted line in the middle, this process is mathematically equivalent to the difference between the two. The output generated by the input nerve fiber then enters the Poisson encoding link to obtain the speed information pulse string and position information pulse string of the target robot arm, where the expression of the position information pulse string can be, but is not limited to, expressed as:

[0094]

[0095] The expression of the speed information pulse string can be, but is not limited to, expressed as:

[0096]

[0097] Among them, u i~U(0,1), [s q,j,i ], i = 1, ..., 100 is the pulse train of position information, [s dq,j,i ], i=1,…,100 is the pulse train of speed information, Δq j , j=1,…,7 are the position error limit range and speed error limit range of each joint respectively.

[0098] It can be seen that each input nerve fiber of the embodiment of the present application is connected to a group of 100 modules, wherein, in the embodiment of the present application, 50 modules are responsible for processing positive numbers, and 50 modules are responsible for processing negative numbers. This structure corresponds to the characteristic of human muscle acting in pairs, i.e., the characteristics of agonist-antagonist muscle. Each module will generate a random number uniformly distributed in the range of 0-1. When the absolute value of the input nerve fiber output is greater than this random number, the encoding result is 1 or -1 (depending on whether the agonist or antagonist muscle is encoded), otherwise it is 0.

[0099] Furthermore, the embodiment of the present application can obtain two sets of discrete pulses, a speed information pulse train and a position information pulse train, and respectively pass k p Parameters and k d The adjustment is used to stimulate the activity of non-excited neurons, that is, to obtain the charge of the cell membrane potential at the current moment, which can be expressed as but not limited to:

[0100]

[0101] Among them, τ g (t) = [τ g,j (t)], j = 1,…,7 is the feedforward torque of the brainstem module filter output.

[0102] Then, the corresponding cell membrane potential at the current moment can be obtained, which can be expressed as but not limited to:

[0103]

[0104] Where T is the time constant, u j (t-1) is the cell membrane potential at the previous moment.

[0105] Furthermore, in the embodiment of the present application, τ(t)=[u j (t)], j = 1, ..., 7 as torque control instructions are output to the actuator in real time. In this form, the output torque at the current moment combines the information at this moment with the information in the past period of time. Therefore, compared with ANN, it is more suitable for time-continuous tasks such as robot control. In theory, it has a better biological basis.

[0106] In step S104, pulse coding and feedforward torque construct a spinal cord module of the robotic arm control model, and the spinal cord module, cerebellum module, thalamus module and brainstem module are combined to construct a robotic arm control model, so as to use the robotic arm control model to output torque control instructions for the target robotic arm.

[0107] As a possible implementation method, the embodiment of the present application can construct a robotic arm control model based on the spinal cord module, cerebellum module, thalamus module and brainstem module, and then use the robotic arm control model to output the torque control instructions of the target robotic arm.

[0108] For example, the schematic diagram of the robot arm control model constructed in the embodiment of the present application can be as follows: Figure 2 As shown, the robot arm control model includes four modules: cerebellum module, thalamus module, hypothalamus module and brainstem module.

[0109] Among them, the cerebellum module uses a data-driven method to learn several different dynamic modes of the robotic arm in advance, and provides corresponding initial torque instructions based on the current motion information during the working process.

[0110] The thalamus module uses the feedback from the force sensor at the end of the robotic arm to dynamically learn and adjust the initial torque command from the cerebellum module during the working process to obtain the final torque command.

[0111] The brainstem module integrates the final torque command from the thalamus module into the feedforward torque and dynamically learns to adjust the proportional parameter k of the feedback control loop during operation. p and k d .

[0112] The spinal module implements functions similar to those of a PD (Proportional Derivative) controller, integrating the feedback and planning quantities to obtain feedback torque and combining it with the feedforward torque to provide real-time torque control instructions for the robot arm.

[0113] In addition, it should be noted that different control cycles are introduced between different modules in the embodiment of the present application. According to the hierarchy from low to high of the spinal cord-brainstem and thalamus-cerebellum, their control cycles are also set from short to long to T1, T2, and T3, as shown in the single-line frame, double-line frame, and triple-line frame.

[0114] According to the brain-like control method of a manipulator based on a pulse neural network proposed in the embodiment of the present application, a cerebellum module can be constructed based on a pulse neural network, a thalamus module can be constructed based on an initial torque instruction, a brainstem module can be constructed using a final torque instruction, a spinal cord module can be constructed based on a feedforward torque, and then a manipulator control model can be constructed by combining the spinal cord module, the cerebellum module, the thalamus module and the brainstem module, so that the manipulator control model can be used to output the torque control instruction of the target manipulator, and the manipulator control model can be constructed to solve the problem that the manipulator control algorithm is not very versatile, so that it has the ability to cope with complex loads. Thus, the problem in the related art that in complex working scenes, due to the influence of the size and mass of the operating object, the working requirements of the manipulator when performing different tasks may be significantly different, thereby damaging the versatility of the adaptive control algorithm based on ANN, and the control accuracy of the algorithm based on SNN is not high, the dynamic performance is poor during the control process, and it relies on the dynamic model for gravity compensation, and cannot adapt well to the control tasks under various complex load conditions.

[0115] Next, a brain-like control device for a robotic arm based on a pulse neural network proposed in accordance with an embodiment of the present application will be described with reference to the accompanying drawings.

[0116] Figure 3 A block diagram of a brain-like control device for a robotic arm based on a pulse neural network provided according to an embodiment of the present application.

[0117] like Figure 3 As shown, the brain-like control device 30 for a robotic arm based on a pulse neural network is applied to a model building stage, wherein the device includes: a first building module 301, a second building module 302, a third building module 303 and a fourth building module 304.

[0118] Among them, the first construction module 301 is used to construct a cerebellum module of the robotic arm control model based on the pulse neural network, so as to use the cerebellum module to obtain the initial torque command of the target robotic arm.

[0119] The second construction module 302 is used to construct a thalamus module of the robot arm control model based on the initial torque command, and use the thalamus module to adjust the initial torque command to obtain a final torque command corresponding to the initial torque command.

[0120] The third construction module 303 is used to construct a brainstem module of the robot control model using the final torque command to obtain the feedforward torque of the target robot.

[0121] The fourth construction module 304 is used to construct the spinal cord module of the robotic arm control model using pulse coding and feedforward torque, and to construct the robotic arm control model in combination with the spinal cord module, cerebellum module, thalamus module and brainstem module, so as to use the robotic arm control model to output the torque control instruction of the target robotic arm.

[0122] Optionally, in one embodiment of the present application, the first construction module 301 includes: a first construction unit and a first generation unit.

[0123] Among them, the first construction unit is used to construct an initial cerebellum module of the robotic arm control model based on the spiking neural network and non-excited spiking neurons.

[0124] The first generation unit is used to train the initial cerebellum module through the oblique circle trajectory, the planar figure eight trajectory, the planar circular trajectory, different gravity compensation modes and MSE of the target robot arm to obtain the cerebellum module.

[0125] Optionally, in one embodiment of the present application, the second construction module 302 includes: a second construction unit and a second generation unit.

[0126] Among them, the second construction unit is used to construct an initial thalamus module of the robotic arm control model based on a fully connected pulse neural network and neurons.

[0127] The second generating unit is used to train the initial thalamus module through the first loss function based on the initial torque instruction to obtain the thalamus module.

[0128] Optionally, in one embodiment of the present application, the third construction module 303 includes: a third construction unit and a third generation unit.

[0129] Among them, the third construction unit is used to establish the first part of the brainstem module based on the pulse neural network, the pulse neuron and the second loss function.

[0130] The third generating unit is used to construct the second part of the brainstem module based on the single-layer neural network and the final torque instruction, so as to obtain the brainstem module by using the first part and the second part.

[0131] Optionally, in one embodiment of the present application, the fourth construction module 304 includes: a fourth generating unit and a fourth construction unit.

[0132] Among them, the fourth generation unit is used to obtain the speed information pulse string and position information pulse of the target robotic arm through nerve fiber and pulse coding based on the planned trajectory and the actual trajectory.

[0133] The fourth construction unit is used to construct a spinal cord module according to the speed information pulse string, the position information pulse string, the feedforward torque and the neurons.

[0134] It should be noted that the aforementioned explanation of the embodiment of the brain-like control method for a robotic arm based on a pulse neural network is also applicable to the brain-like control device for a robotic arm based on a pulse neural network in this embodiment, and will not be repeated here.

[0135] According to the brain-like control device for a manipulator based on a pulse neural network proposed in the embodiment of the present application, a cerebellum module can be constructed based on a pulse neural network, a thalamus module can be constructed based on an initial torque instruction, a brainstem module can be constructed using a final torque instruction, a spinal cord module can be constructed based on a feedforward torque, and then a manipulator control model can be constructed by combining the spinal cord module, the cerebellum module, the thalamus module, and the brainstem module, so that the manipulator control model can be used to output the torque control instruction of the target manipulator, and the manipulator control model can be constructed to solve the problem that the manipulator control algorithm is not very versatile, so that it has the ability to cope with complex loads. Thus, the problem in the related art that in complex working scenes, due to the influence of the size and mass of the operating object, the working requirements of the manipulator may be significantly different when performing different tasks, thereby damaging the versatility of the adaptive control algorithm based on ANN, and the control accuracy of the algorithm based on SNN is not high, the dynamic performance is poor during the control process, and it relies on the dynamic model for gravity compensation, and cannot adapt well to the control tasks under various complex load conditions.

[0136] The above embodiment describes the model building phase, and the following describes an embodiment of the test application phase.

[0137] Figure 4 This is a flowchart of a brain-like control method for a robotic arm based on a pulse neural network according to another embodiment of the present application.

[0138] like Figure 4 As shown, the brain-like control method of the robotic arm based on the pulse neural network is applied in the model application stage, wherein the method comprises the following steps:

[0139] In step S401, a control instruction of a robot arm in working state is obtained.

[0140] In step S402, a control instruction is input into a pre-built robot arm control model to output a torque control instruction of the robot arm, wherein the pre-built robot arm control model is composed of at least one of a cerebellum module, a thalamus module, a brainstem module and a spinal cord module.

[0141] In step S403, the robot arm is controlled to output a corresponding torque based on the torque control instruction.

[0142] In some embodiments, the embodiments of the present application can input the acquired control instructions of the robotic arm in a working state into a pre-built robotic arm control model, and then output the torque control instructions of the robotic arm, and obtain the corresponding torque of the robotic arm by controlling the torque control instructions.

[0143] Exemplarily, the present application embodiment will conduct simulation experiments to verify the superiority of the pre-built robot arm control model. Among them, in the present application embodiment, all simulation experiments are carried out in an XX workstation equipped with an XX graphics card and an XX CPU, the simulation environment is pybullet, the simulation language is python, the deep learning framework is pytorch, the pulse neural network framework is SpikingJelly, and the simulation control model of the robot arm is XX of XX company.

[0144] In the trajectory tracking task of the embodiment of the present application, the joint position will produce errors, and the curve of the root mean square error changing with time is as follows: Figure 5 As shown, the expression of the root mean square error can be, but is not limited to, expressed as:

[0145]

[0146] Depend on Figure 5 It can be seen that at the beginning, the embodiment of the present application has a load of 1.5 kg at the end of the robot arm, and it converges to a stable working state after about 2 seconds; after running for 12 seconds, an additional load of 1 kg is applied to the end, and similarly, it only takes about 2 seconds to converge; after running for another 12 seconds, the additional load is removed, and the algorithm still converges within 2 seconds. It can be seen that the embodiment of the present application has good dynamic performance and can quickly respond to changes in the mass of the end load.

[0147] Figure 6 What is shown is the curve of the two weights generated by the thalamus module changing with the end load. It can be seen that the greater the end load mass of the embodiment of the present application, the greater the torque weight assigned to the heavy object mode, and the smaller the torque weight assigned to the light object mode. This further indicates that the effect of the thalamus module adjusting the torque weights of different modes through reinforcement learning has certain bionic characteristics.

[0148] Furthermore, physiological studies have shown that in the human arm, the stiffness and damping of the joints are inversely correlated. When the joint starts to move, its stiffness decreases, while when it is stationary, the stiffness increases; when approaching the target, the damping decreases, and when starting to move, the damping increases. Figure 7 (a) shows the k of a part of the joint during the working process p and k d The changing curve during the movement, Figure 7 (b) shows the k of another part of the joint during the working process p and k d The changing curves during the movement are shown as solid and dotted lines. Figure 7It can be seen that the two parameters of each joint basically show opposite trends, indicating that the effect of the brainstem module adjusting joint stiffness and damping through reinforcement learning has certain bionic characteristics. In addition, the embodiment of the present application also noted that this feature is not particularly obvious in some joints near the bottom, which may be because the joints near the bottom are affected by the coupled motion of other joints. Figure 6 and Figure 7 The embodiments of the present application are not only bionic in structure, but also have certain bionic properties in function.

[0149] Figure 8 and Fig. 9 The following are comparison curves of the joint error of the embodiment of the present application and the PD algorithm with an accurate dynamic model as the load mass changes in three trajectory tracking tasks, and the comparison curves of the joint error of the PD algorithm without an accurate dynamic model as the load mass changes, where the dotted line represents the PD algorithm. At this time, in the embodiment of the present application, the root mean square error of the joint position of the trajectory tracking task can be expressed as, but not limited to:

[0150]

[0151] Where N is the simulation duration of a trajectory.

[0152] In addition, if not otherwise specified, in all experiments involving the PD algorithm in the embodiments of the present application, the parameters may be, but are not limited to, expressed as: k p =[200,100,600,100,200,100,100],k d =[20,30,40,3,10,10,3]. Further, the embodiment of the present application assumes that the PD algorithm can accurately obtain the dynamic model of the end load through identification, so as to use the accurate gravity compensation torque. In this case, the result is as follows: Figure 8 As shown. Figure 8 It can be seen that when a 1kg load is mounted at the end of the embodiment of the present application, the PD algorithm in the first two trajectory tracking tasks can obtain a smaller trajectory tracking error than the embodiment of the present application; however, in the third trajectory tracking task, the trajectory tracking error of the PD algorithm is much larger than that of the embodiment of the present application, about 2 times. As the mass of the load continues to increase, the trajectory tracking error of the PD algorithm increases almost linearly, while the trajectory tracking error of the embodiment of the present application can be basically stable, which is better than the trajectory tracking error of the PD algorithm. However, for certain tasks, such as when the load is an elastomer or a liquid, it is difficult or impossible for the embodiment of the present application to obtain an accurate dynamic model through identification, and accordingly, it is impossible to use the model to provide accurate gravity compensation torque for the PD algorithm. At this time, the embodiment of the present application assumes that no matter what the mass of the load is, only the model of a 1kg load can be used to provide gravity compensation for the PD algorithm. The results are as follows. Fig. 9As shown. Fig. 9 It can be seen that as the load mass increases, the gap between the actual model and the preset model becomes larger and larger, and the control accuracy of the PD algorithm decreases significantly. However, since the embodiment of the present application only requires input data from the force sensor and does not require an accurate dynamic model, it can maintain the stability of the tracking accuracy performance under loads of various masses. Overall, the embodiment of the present application is more adaptable and has higher tracking accuracy than the PD algorithm.

[0153] In addition, in the embodiment of the present application, Table 1 shows in more detail the tracking errors of the PD algorithm with an accurate dynamic model, CBMC-V2 and the embodiment of the present application under three trajectories. For the plane circular trajectory and the oblique circular trajectory, it can be seen from Table 1 that although the error of the PD algorithm is about 10% smaller than that of the embodiment of the present application at 1kg, under other mass load conditions, the embodiment of the present application has a tracking accuracy improvement of about 14%-60% on the two trajectories. As for the three-dimensional figure eight trajectory, due to its stronger dynamic characteristics, the PD algorithm with the same set of parameters performs worse than the first two trajectories on this trajectory, while the embodiment of the present application can have a tracking accuracy improvement of more than 50% under various mass loads. On three trajectories and various loads of different masses, compared with CBMC-V2, the tracking accuracy of the embodiment of the present application has been significantly improved, ranging from 24% to 80%. Among them, Table 1 is a comparison table of joint errors of the PD algorithm with an accurate dynamic model, CBMC-V2 and the embodiment of the present application under different mass load conditions in three trajectory tracking tasks provided according to an embodiment of the present application.

[0154] Table 1

[0155]

[0156]

[0157] Fig.10 What is shown is the trajectory of the workspace and joint space of the PD algorithm with an accurate dynamic model and the embodiment of the present application on a three-dimensional figure eight trajectory under the condition of a 2kg load. Fig.10 It can be intuitively seen that the embodiment of the present application has a smaller error and higher tracking accuracy. In addition, the embodiment of the present application also found that the reason for the poor effect of the PD algorithm is not that the parameters of a certain joint are particularly inappropriate, but that each joint has a certain error, which shows that due to the coupling of each joint, the PD algorithm requires a lot of experiments and engineering experience to adjust the parameters, and such a set of parameters can often only work for certain tasks. The embodiment of the present application does not have these problems because it can dynamically adjust the network parameters, which further proves the versatility of the embodiment of the present application. Among them, Fig.10(a) is a three-dimensional view of a workspace trajectory curve in a three-dimensional figure eight trajectory tracking task with a 2 kg load provided according to an embodiment of the present application; Fig.10 (b) is a plane decomposition view of a workspace trajectory curve in a three-dimensional figure eight trajectory tracking task with a 2 kg load provided by an embodiment of the present application; Fig.10 (c) is a two-dimensional view of the joint space trajectory curve in the three-dimensional figure eight trajectory tracking task with a 2kg load provided according to an embodiment of the present application.

[0158] According to the brain-like control method of a robotic arm based on a pulse neural network proposed in the embodiment of the present application, the control command of the robotic arm in a working state can be input into a pre-built robotic arm control model to output the torque control command of the robotic arm, and then the robotic arm outputs the corresponding torque. Through the pre-built robotic arm control model, the torque control command of the robotic arm can be accurately processed and converted, thereby generating more accurate torque control commands, so that the robotic arm can maintain flexibility and adaptability when facing a complex or dynamic working environment, and improve overall work efficiency. Thus, the problem in the related art that in a complex working scene, due to the influence of the size and mass of the operating object, the working requirements of the robotic arm when performing different tasks may be significantly different, thereby damaging the versatility of the adaptive control algorithm based on ANN, and the algorithm based on SNN has low control accuracy, poor dynamic performance during the control process, relies on the dynamic model for gravity compensation, and cannot adapt well to control tasks under various complex load conditions.

[0159] Next, a brain-like control device for a robotic arm based on a pulse neural network proposed in accordance with an embodiment of the present application will be described with reference to the accompanying drawings.

[0160] Fig.11 This is a block diagram of a brain-like control device for a robotic arm based on a pulse neural network according to another embodiment of the present application.

[0161] like Fig.11 As shown, the brain-like control device 110 for a robotic arm based on a pulse neural network is applied in the model application stage, wherein the device 110 includes: an acquisition module 1101, an input module 1102 and an output module 1103.

[0162] The acquisition module 1101 is used to acquire control instructions of the robot arm in working state.

[0163] The input module 1102 is used to input the control instruction into the pre-built robot arm control model to output the torque control instruction of the robot arm, wherein the pre-built robot arm control model is composed of at least one of a cerebellum module, a thalamus module, a brainstem module and a spinal cord module.

[0164] The output module 1103 is used to control the robot arm to output a corresponding torque based on the torque control instruction.

[0165] It should be noted that the aforementioned explanation of the embodiment of the brain-like control method for a robotic arm based on a pulse neural network is also applicable to the brain-like control device for a robotic arm based on a pulse neural network in this embodiment, and will not be repeated here.

[0166] According to the brain-like control device for a manipulator based on a pulse neural network proposed in the embodiment of the present application, the control command of the manipulator in a working state can be input into a pre-built manipulator control model to output the torque control command of the manipulator, and then the manipulator outputs the corresponding torque. Through the pre-built manipulator control model, the torque control command of the manipulator can be accurately processed and converted, thereby generating more accurate torque control commands, so that the manipulator can maintain flexibility and adaptability when facing a complex or dynamic working environment, and improve overall work efficiency. Thus, the problem in the related art that in a complex working scene, due to the influence of the size and mass of the operating object, the working requirements of the manipulator when performing different tasks may be significantly different, thereby damaging the versatility of the adaptive control algorithm based on ANN, and the control accuracy of the algorithm based on SNN is not high, the dynamic performance is poor during the control process, and it relies on the dynamic model for gravity compensation, and cannot adapt well to the control tasks under various complex load conditions.

[0167] Fig.12 This is a schematic diagram of the structure of an electronic device provided according to an embodiment of the present application. The electronic device may include:

[0168] A memory 1201 , a processor 1202 , and a computer program stored in the memory 1201 and executable on the processor 1202 .

[0169] When the processor 1202 executes the program, the brain-like control method of the robotic arm based on the pulse neural network provided in the above embodiment is implemented.

[0170] Furthermore, the electronic device further comprises:

[0171] The communication interface 1203 is used for communication between the memory 1201 and the processor 1202 .

[0172] The memory 1201 is used to store computer programs that can be executed on the processor 1202 .

[0173] The memory 1201 may include a high-speed RAM memory, and may also include a non-volatile memory (non-volatile memory), such as at least one disk memory.

[0174] If the memory 1201, the processor 1202 and the communication interface 1203 are implemented independently, the communication interface 1203, the memory 1201 and the processor 1202 can be connected to each other through a bus and communicate with each other. The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus. The bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Fig.12 Only one thick line is used in the diagram, but this does not mean that there is only one bus or only one type of bus.

[0175] Optionally, in a specific implementation, if the memory 1201, the processor 1202 and the communication interface 1203 are integrated on a chip, the memory 1201, the processor 1202 and the communication interface 1203 can communicate with each other through an internal interface.

[0176] The processor 1202 may be a central processing unit (CPU), or an application specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of the present application.

[0177] An embodiment of the present application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-mentioned brain-like control method of a robotic arm based on a pulse neural network.

[0178] An embodiment of the present application also provides a computer program product, including a computer program, which, when executed, implements the above-mentioned brain-like control method of a robotic arm based on a pulse neural network.

[0179] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or N embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine the different embodiments or examples described in this specification and the features of the different embodiments or examples, without contradiction.

[0180] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include at least one of the features. In the description of this application, "N" means at least two, such as two, three, etc., unless otherwise clearly and specifically defined.

[0181] Any process or method description in a flowchart or otherwise described herein may be understood to represent a module, fragment or portion of code comprising one or N executable instructions for implementing the steps of a custom logical function or process, and the scope of the preferred embodiments of the present application includes alternative implementations in which functions may not be performed in the order shown or discussed, including performing functions in a substantially simultaneous manner or in reverse order depending on the functions involved, which should be understood by technicians in the technical field to which the embodiments of the present application belong.

[0182] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by an instruction execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute instructions), or in combination with these instruction execution systems, devices or apparatuses. For the purpose of this specification, "computer-readable medium" can be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution system, device or apparatus, or in combination with these instruction execution systems, devices or apparatuses. More specific examples of computer-readable media (a non-exhaustive list) include the following: an electrical connection with one or N wirings (electronic devices), a portable computer disk box (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), a fiber optic device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be paper or other suitable medium on which the program is printed, since the program may be obtained electronically by optically scanning the paper or other medium and then editing, interpreting or processing in other suitable ways as necessary and then storing it in a computer memory.

[0183] It should be understood that the various parts of the present application can be implemented by hardware, software, firmware or a combination thereof. In the above embodiment, the N steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. If implemented by hardware, as in another embodiment, it can be implemented by any one or a combination of multiple of the following technologies known in the art: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0184] A person skilled in the art may understand that all or part of the steps in the method for implementing the above-mentioned embodiment may be completed by instructing related hardware through a program, and the program may be stored in a computer-readable storage medium, which, when executed, includes one or a combination of the steps of the method embodiment.

[0185] In addition, each functional unit in each embodiment of the present application may be integrated into a processing module, or each unit may exist physically separately, or two or more units may be integrated into one module. The above-mentioned integrated module may be implemented in the form of hardware or in the form of a software functional module. If the integrated module is implemented in the form of a software functional module and sold or used as an independent product, it may also be stored in a computer-readable storage medium.

[0186] The storage medium mentioned above may be a read-only memory, a magnetic disk or an optical disk, etc. Although the embodiments of the present application have been shown and described above, it can be understood that the above embodiments are exemplary and cannot be understood as limiting the present application. A person of ordinary skill in the art may change, modify, replace and modify the above embodiments within the scope of the present application.

Claims

1. A brain-like control method for a robotic arm based on a pulse neural network, characterized in that: Applied to the model building stage, wherein the method comprises the following steps: A cerebellum module of a robot arm control model is constructed based on a pulse neural network, so as to obtain an initial torque command of a target robot arm by using the cerebellum module; Building a thalamus module of the robot arm control model based on the initial torque command, and adjusting the initial torque command using the thalamus module to obtain a final torque command corresponding to the initial torque command; Using the final torque command to construct a brainstem module of the manipulator control model to obtain a feedforward torque of the target manipulator; A spinal cord module of the robotic arm control model is constructed based on pulse coding and the feedforward torque, and the robotic arm control model is constructed in combination with the spinal cord module, the cerebellum module, the thalamus module and the brainstem module, so as to utilize the robotic arm control model to output torque control instructions of the target robotic arm.

2. The method according to claim 1, characterized in that The cerebellum module for constructing a robotic arm control model based on a spiking neural network includes: Constructing an initial cerebellum module of the robotic arm control model based on the spiking neural network and non-spiking neurons; The initial cerebellum module is trained through the oblique circular trajectory, planar figure eight trajectory, planar circular trajectory, different gravity compensation modes and mean square error (MSE) of the target robotic arm to obtain the cerebellum module.

3. The method according to claim 1, characterized in that The thalamus module for constructing the robotic arm control model based on the initial torque instruction includes: Constructing an initial thalamus module of the robotic arm control model based on a fully connected spiking neural network and neurons; Based on the initial torque instruction, the initial thalamus module is trained by a first loss function to obtain the thalamus module.

4. The method according to claim 1, characterized in that: The brainstem module for constructing the robotic arm control model using the final torque instruction includes: Establishing a first part of the brainstem module based on a spiking neural network, spiking neurons, and a second loss function; The second part of the brainstem module is constructed based on the single-layer neural network and the final torque instruction, so as to obtain the brainstem module using the first part and the second part.

5. The method according to claim 1, characterized in that The spinal cord module for constructing the robotic arm control model based on pulse coding and the feedforward torque comprises: Based on the planned trajectory and the actual trajectory, the speed information pulse string and the position information pulse string of the target manipulator are obtained through the nerve fibers and the pulse coding; The spinal cord module is constructed according to the speed information pulse train, the position information pulse train, the feedforward torque and neurons.

6. A brain-like control method for a robotic arm based on a pulse neural network, characterized in that: Applied to the model application stage, wherein the method comprises the following steps: Obtain control instructions for the robotic arm in working state; Inputting the control instruction into a pre-built robot arm control model to output a torque control instruction of the robot arm, wherein the pre-built robot arm control model is composed of at least one of a cerebellum module, a thalamus module, a brainstem module, and a spinal cord module; The mechanical arm is controlled to output a corresponding torque based on the torque control instruction.

7. A brain-like control device for a robotic arm based on a pulse neural network, characterized in that: Applied to the model building stage, wherein the device comprises: A first construction module is used to construct a cerebellum module of a robot arm control model based on a pulse neural network, so as to obtain an initial torque command of a target robot arm by using the cerebellum module; A second construction module is used to construct a thalamus module of the manipulator control model based on the initial torque command, and use the thalamus module to adjust the initial torque command to obtain a final torque command corresponding to the initial torque command; A third construction module is used to construct a brainstem module of the manipulator control model using the final torque command to obtain a feedforward torque of the target manipulator; The fourth construction module is used to construct a spinal cord module of the robotic arm control model based on pulse coding and the feedforward torque, and to construct the robotic arm control model in combination with the spinal cord module, the cerebellum module, the thalamus module and the brainstem module, so as to use the robotic arm control model to output the torque control instruction of the target robotic arm.

8. The device according to claim 7, characterized in that The first building block comprises: A first construction unit is used to construct an initial cerebellum module of the robotic arm control model according to the spiking neural network and non-spiking neurons; The first generating unit is used to train the initial cerebellum module through the oblique circular trajectory, the planar figure eight trajectory, the planar circular trajectory, different gravity compensation modes and MSE of the target robotic arm to obtain the cerebellum module.

9. The device according to claim 7, characterized in that The second building block comprises: A second construction unit is used to construct an initial thalamus module of the robotic arm control model based on a fully connected spiking neural network and neurons; The second generating unit is used to train the initial thalamus module based on the initial torque instruction through a first loss function to obtain the thalamus module.

10. The device according to claim 7, characterized in that The third building block comprises: A third construction unit, configured to establish a first part of the brainstem module based on a spiking neural network, spiking neurons, and a second loss function; The third generating unit is used to construct the second part of the brainstem module based on the single-layer neural network and the final torque instruction, so as to obtain the brainstem module by using the first part and the second part.

11. The device according to claim 7, characterized in that The fourth building block comprises: A fourth generating unit, configured to obtain a velocity information pulse string and a position information pulse string of the target manipulator through nerve fibers and the pulse coding based on the planned trajectory and the actual trajectory; The fourth construction unit is used to construct the spinal cord module according to the speed information pulse string, the position information pulse string, the feedforward torque and neurons.

12. A brain-like control device for a robotic arm based on a pulse neural network, characterized in that: Applied to the model application stage, wherein the device comprises: An acquisition module, used for acquiring control instructions of a robotic arm in a working state; An input module, used for inputting the control instruction into a pre-built robot arm control model to output a torque control instruction of the robot arm, wherein the pre-built robot arm control model is composed of at least one of a cerebellum module, a thalamus module, a brainstem module and a spinal cord module; An output module is used to control the mechanical arm to output a corresponding torque based on the torque control instruction.

13. An electronic device, characterized in that: include: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the brain-like control method for a robotic arm based on a pulse neural network as described in any one of claims 1 to 5 or claim 6.

14. A computer-readable storage medium having a computer program stored thereon, characterized in that: The program is executed by a processor to implement a brain-like control method for a robotic arm based on a pulse neural network as described in any one of claims 1 to 5 or claim 6.

15. A computer program product, characterized in that It includes a computer program which, when executed, is used to implement the brain-like control method of a robotic arm based on a pulse neural network as described in any one of claims 1 to 5 or claim 6.

Images