A basal ganglia-inspired dual-level CPG-based gait control method for quadruped robots

Abstract

The present invention discloses a gait control method for a quadruped robot using a dual-level central pattern generator (CPG) inspired by the basal ganglia, comprising: modeling the single legs of the quadruped robot using forward kinematics and inverse kinematics; using a dual-level central pattern generator for each leg, consisting of a rhythm generation part and a pattern formation part, wherein the rhythm generation part is mainly responsible for phase change, and the pattern formation part is mainly responsible for the coordination of each joint; determining the hierarchical structure and coupling topology of the CPG network; and introducing a behavior selection model of the basal ganglia, which can be used to select the optimal gait for different environments and improve environmental adaptability. Compared with traditional control methods, the control of the quadruped robot using the present invention has a high degree of biological rationality, stronger robustness to external interference, good adaptability to terrain, and can flexibly control various movement modes of the quadruped robot.

Description

Technical Field

[0001] The present invention belongs to the field of quadruped robot control, and in particular to a quadruped robot gait control method based on a double-level CPG inspired by basal ganglia. Background Art

[0002] With the rapid development of robotics, intelligent robots are increasingly being used in military, industrial, and civilian applications. Robots are expected to assist humans in many arduous, monotonous tasks, as well as dangerous and unsanitary ones. However, in everyday environments, conventional wheeled robots are unable to adapt to rough and irregular surfaces, and tracked robots are unable to traverse common terrain such as ravines, obstacles, and steps. In comparison, legged robots offer superior mobility, environmental adaptability, and dynamic performance, making them a focus of robotics research.

[0003] For legged robots, the most fundamental issue in achieving their functionality is stable walking. Among existing motion planning methods, model-based approaches require precise modeling of the environment and can achieve high-precision motion in known environments, but their adaptability and perception capabilities in unknown environments are poor. Behavior-based control methods have good adaptability to unknown dynamic environments, but they struggle to account for all situations encountered in complex environments and plan complex, global tasks. Furthermore, neural system model-based approaches are primarily based on the structure and mechanism of the motion control nervous system of higher animals. Drawing on the reflex mechanisms of the nervous system of higher animals, they incorporate external environmental information into the robot's rhythmic motion controller, enabling rhythmic motion to adapt to diverse environments and terrains.

[0004] The central pattern generator (CPG) is a classic control method for spontaneous rhythmic movement based on a neural system model. The CPG is composed of multiple neurons, with excitatory and inhibitory neurons being the two most basic types. The combined action of these neurons creates a stable phase relationship, stimulating rhythmic movement of the limbs and ultimately forming gait. The CPG can generate stable rhythmic signals through self-oscillation without requiring high-level commands. At the same time, high-level commands can be used to regulate the robot's movement through the CPG. Furthermore, its structure is simple and highly adaptable to terrain.

[0005] In recent years, research on central pattern generators, especially those applied to areas such as manipulator control and robot control, has attracted increasing attention. In the paper “Sensorimotor interactions during locomotion: principles derived from biological systems,” Cohen pointed out that sensory-motor interactions are real-time, bidirectional, and occur at almost all levels, such as local muscles, CPGs, and motor centers in the cerebral cortex. In the paper “Self-organized control of biped locomotion by neural oscillators in an unpredictable environment,” Taga et al. used the concept of “global transmission” to explain adaptive rhythmic movement, pointing out that animal movement control systems based on CPGs and reflex functions can achieve self-organizational interactions among the nervous system, biomechanical system, and environment, allowing rhythmic behavior to be transmitted throughout the system in the form of a global limit cycle. Summary of the Invention

[0006] The purpose of the present invention is to address the problems existing in the above-mentioned prior art and provide a basal ganglia-inspired dual-level CPG quadruped robot control method with better adaptability and control effect.

[0007] The technical solution to achieve the purpose of the present invention is: a quadruped robot gait control method based on a two-level CPG inspired by the basal ganglia, the method comprising the following steps:

[0008] Step 1: Construct the kinematic equations of the quadruped robot and verify the accuracy of the established model under the ADAMS simulation platform;

[0009] Step 2, establish a two-layer CPG neural network model;

[0010] Step 3: Construct a behavior selection model of the basal ganglia. According to this model, the quadruped robot's behavior requests flow along their respective channels through the various nuclei of the basal ganglia, including the striatum, the globus pallidus external nucleus (GPe), and the subthalamic nucleus (STN), and finally reach the output nuclei (GPi) and the substantia nigra pars reticulata (SNr).

[0011] Furthermore, the two-layer CPG neural network model in step 2 includes two parts: rhythm generation RG and pattern formation PF;

[0012] The RG part is composed of a semi-central rhythm generator, including the RG neuron RG-E for extension and the RG neuron RG-F for flexion; the PF part includes four PF neurons, which control the four consecutive stages of swing, landing, stance and lift-off respectively;

[0013] Specifically:

[0014] In step 2-1, the membrane potentials of the extensor and flexor neurons and motor neurons in the RG and PF parts are expressed as:

[0015]

[0016] The membrane potential of an interneuron is expressed as:

[0017]

[0018] Where, I Nap is the continuous sodium current, I K is the potassium rectifier current, I Leak is the leakage current, I SynE and I SynI are the synaptic excitatory and synaptic inhibitory currents, respectively, C is a neuronal capacitance, and V is the average membrane voltage of the neuron;

[0019] The ion current formula is as follows:

[0020]

[0021] The voltage-dependent activation variables of potassium delayed rectifier and persistent sodium channels are expressed as:

[0022]

[0023] Voltage-dependent inactivation variable h NaP Calculated by the following differential equation:

[0024]

[0025] in:

[0026]

[0027] f(V i ) represents the average population activity and the output of neuron j, which is defined as a sigmoid function:

[0028]

[0029] Where, E {Na,K,Leak,SynE,SynI} is the corresponding reversal potential, is the maximum conductance of the corresponding ion channel, a jiis the gain of the excitatory synaptic input of neuron j relative to neuron i, b ji is the gain of neuron j relative to the inhibitory input of neuron i, c i is the gain of the tonic descending signal on the neuron, and w {1,2},i is the incoming feedback to neuron i {1,2} Gain; τ hNapmax is a time constant; V 1 / 2 is the half-activation voltage, k is the S-type function gain, V th is the threshold of each neuron; Σ k a ki f(V k )+Σ m β mi f(V m ) shows the connections from other neurons to neuron i and creates an intrinsic mutual coordination;

[0030] The Ia afferent feedback from the extensor motor neurons as a stretch reflex is effective for supporting weight. Specifically, when the leg is weight-bearing, the following feedback signals from the hip extensors are input to the corresponding motor neurons:

[0031]

[0032] Where, F load is the load on each leg measured by the force sensors mounted on its legs, k v is the gain, I norm2 is the driving current of the legs, v norm2 is the driving voltage of the legs;

[0033] In step 2-2, the hierarchical structure and coupling topology of the CPG network are determined. The connection parameters between RG and PF neurons are responsible for forming the basic rhythm, and the connection parameters between PF neurons and motor neurons are responsible for creating the leg trajectory in four consecutive phases. The stiffness parameters and the gain load of the leg load feedback tend to produce more realistic leg trajectories and robust walking on uneven terrain and under perturbations.

[0034] Furthermore, step 2 also includes:

[0035] Step 2-3, based on the connection parameters described in step 2-2 and based on its characteristics, performs parameter adjustment, including the following steps:

[0036] (1) Suspend a quadruped robot and let its legs swing in the air. Manually adjust the parameters of the CPG model so that each leg can produce a loose trajectory and stable oscillation in each of four consecutive phases. Specifically, first determine the connection between RG and PF neurons to reproduce the basic rhythm. Second, focus on exploring the connection between PF neurons and motor neurons to roughly reproduce the coordinated pattern of activity in each phase.

[0037] (2) Place the quadruped robot on the ground and use the adjustment software to adjust the parameters so that the quadruped robot can walk on flat terrain;

[0038] (3) Manually modify parameters so that the motor neuron activation patterns and leg trajectories of the quadruped robot are the same as those of quadruped mammals.

[0039] Compared with the existing technology, the present invention has the following significant advantages: 1) It does not rely on accurate modeling of the environment and can still walk stably in unknown environments; 2) The two-layer CPG network includes a rhythm generation part and a pattern formation part, and its structure is clearer, the division of labor is clearer, and it is easy to control; 3) In response to changes in the environment, the basal ganglia are used to make decisions and select the optimal gait, which has fast speed, strong adaptability and good stability.

[0040] The present invention is further described in detail below with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 Schematic diagram of a simplified model of a single leg of a quadruped robot.

[0042] Figure 2 Detailed schematic diagram of the double-layer CPG used in a single leg.

[0043] Figure 3 Schematic diagram of the interaction between the various nuclei of the direct and indirect pathways of the basal ganglia. DETAILED DESCRIPTION

[0044] In order to make the purpose, technical solutions and advantages of this application more clear, the following further describes this application in detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain this application and are not intended to limit this application.

[0045] It should be noted that if there are descriptions involving "first", "second", etc. in the embodiments of the present invention, the descriptions of "first", "second", etc. are only for descriptive purposes and cannot be understood as indicating or implying their relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include at least one of such features. In addition, the technical solutions between the various embodiments can be combined with each other, but they must be based on the ability of ordinary technicians in this field to implement them. When the combination of technical solutions is contradictory or cannot be implemented, it should be deemed that such a combination of technical solutions does not exist and is not within the scope of protection required by the present invention.

[0046] In one embodiment, a basal ganglia-inspired dual-level CPG quadruped robot gait control method is provided, the method comprising the following steps:

[0047] Step 1: Construct the kinematic equations of the quadruped robot and verify the accuracy of the established model under the ADAMS simulation platform;

[0048] Each leg of the quadruped robot includes a side swing joint, a hip joint and a knee joint, and its external structure and the motor model used are the same.

[0049] ADAMS software can not only visually display the motion trajectory and running curve of the quadruped robot, but also provide an external interface. The algorithm results can be intuitively displayed using the joint simulation solution of ADAMS and MATLAB.

[0050] Step 2: For each leg, a two-layer CPG neural network model is established;

[0051] Step 3: Build a basal ganglia behavior selection model. According to this model, the quadruped robot's behavior requests flow through various nuclei in the basal ganglia along their respective pathways, including the striatum, globus pallidus externus (GPe), and subthalamic nucleus (STN), ultimately reaching the output nuclei (GPi) and substantia nigra pars reticulata (SNr). The GPi / SNr maintain or increase inhibition on unselected pathways while removing inhibition on selected pathways, allowing the selected behavior to be executed via the thalamus.

[0052] Furthermore, in one embodiment, in combination Figure 1 , the kinematic equations of the quadruped robot are constructed as described in step 1, specifically including:

[0053] Construct the forward solution of the kinematic equations of the quadruped robot:

[0054]

[0055] Construct a quadruped robot and inversely solve the kinematic equations:

[0056]

[0057] Where L1 is the length from the knee joint to the foot, L2 is the length from the hip joint to the knee joint, θ1 is the angle between L1 and the ground, θ2 is the angle between L2 and the extension line of L1, x and y are the posture in the x direction and the posture in the y direction respectively, atan2(y,x) is the inverse tangent of x and y, s2 is sinθ2, and c2 is cosθ2.

[0058] Furthermore, in one embodiment, in combination Figure 2 The two-layer CPG neural network model in step 2 includes two parts: rhythm generation RG and pattern formation PF; the rhythm generation part is mainly responsible for phase change, while the pattern formation part is mainly responsible for the coordination of each joint;

[0059] The RG part is composed of a semi-central rhythm generator, including the RG neuron RG-E for extension and the RG neuron RG-F for flexion; the PF part includes four PF neurons, which control the four consecutive stages of swing, landing, stance and lift-off respectively;

[0060] Specifically:

[0061] In step 2-1, the membrane potentials of the extensor and flexor neurons and motor neurons in the RG and PF parts are expressed as:

[0062]

[0063] The membrane potential of an interneuron is expressed as:

[0064]

[0065] Where, I Nap is the continuous sodium current, I K is the potassium rectifier current, I Leak is the leakage current, I SynE and I SynI are the synaptic excitatory and synaptic inhibitory currents, respectively, C is a neuronal capacitance, and V is the average membrane voltage of the neuron;

[0066] The ion current formula is as follows:

[0067]

[0068] The voltage-dependent activation variables of potassium delayed rectifier and persistent sodium channels are expressed as:

[0069]

[0070] Voltage-dependent inactivation variable hNaP Calculated by the following differential equation:

[0071]

[0072] in:

[0073]

[0074] f(V i ) represents the average population activity and the output of neuron j, which is defined as a sigmoid function:

[0075]

[0076] Where, E {Na,K,Leak,SynE,SynI} is the corresponding reversal potential, is the maximum conductance of the corresponding ion channel, a ji is the gain of the excitatory synaptic input of neuron j relative to neuron i, b ji is the gain of neuron j relative to the inhibitory input of neuron i, c i is the gain of the tonic descending signal on the neuron, and w {1,2},i is the incoming feedback to neuron i {1,2} Gain; τ hNapmax is a time constant; V 1 / 2 is the half-activation voltage, k is the S-type function gain, V th is the threshold of each neuron; Σ k a ki f(V k )+∑ m β mi f(V m ) shows the connections from other neurons to neuron i and creates an intrinsic mutual coordination;

[0077] The Ia afferent feedback from the extensor motor neurons as a stretch reflex is effective for supporting weight. Specifically, when the leg is weight-bearing, the following feedback signals from the hip extensors are input to the corresponding motor neurons:

[0078]

[0079] Where, F load is the load on each leg measured by the force sensors mounted on its legs, k v is gain;

[0080] In step 2-2, the hierarchical structure and coupling topology of the CPG network are determined. The connection parameters between RG and PF neurons are responsible for forming the basic rhythm, and the connection parameters between PF neurons and motor neurons are responsible for creating the leg trajectory in four consecutive phases. The stiffness parameters and the gain load of the leg load feedback tend to produce more realistic leg trajectories and robust walking on uneven terrain and under perturbations.

[0081] Furthermore, in one embodiment, step 2 further includes:

[0082] Step 2-3, based on the connection parameters described in step 2-2 and based on its characteristics, performs parameter adjustment, including the following steps:

[0083] (1) Suspend a quadruped robot and let its legs swing in the air. Manually adjust the parameters of the CPG model so that each leg can produce a loose trajectory and stable oscillation in each of four consecutive phases. Specifically, first determine the connection between RG and PF neurons to reproduce the basic rhythm. Second, focus on exploring the connection between PF neurons and motor neurons to roughly reproduce the coordinated pattern of activity in each phase.

[0084] (2) Place the quadruped robot on the ground and use the adjustment software to adjust the parameters so that the quadruped robot can walk on flat terrain;

[0085] (3) Manually modify parameters so that the motor neuron activation patterns and leg trajectories of the quadruped robot are the same as those of quadruped mammals.

[0086] Furthermore, in one embodiment, the behavior selection model of the basal ganglia is constructed in step 3. According to this model, the behavior request of the quadruped robot flows along respective channels through the various nuclei of the basal ganglia, including the striatum, the globus pallidus external nucleus (GPe), and the subthalamic nucleus (STN), and finally reaches the output nuclei (GPi) and the substantia nigra pars reticulata (SNr), specifically including:

[0087] Assuming that a suitable behavior needs to be selected from I behavioral channels, the various nuclei of the basal ganglia are also divided into I channels accordingly. Each channel is represented by a leaky integration neuron. The leaky integration neuron model is:

[0088]

[0089] Where x is the state of the neuron, y is the output of the neuron, k, m, and ε are all model parameters, and H is the step function;

[0090] Basal ganglia behavioral selection models such as Figure 3 As shown, the mathematical models of the various nuclei of the basal ganglia are as follows:

[0091] (1) The mathematical model of striatum D1 is:

[0092]

[0093] Where u i SD1 、a i SD1 、y i SD1 are the input, state, and output of neuron i in striatum D1 channel, w CSD1 is the connection weight from the cortex to the striatum D1, λ is the influence of dopamine neurons, which has an incentive effect on the striatum D1, and ε SD1 Output threshold for striatum D1;

[0094] (2) The mathematical model of striatum D2 is:

[0095]

[0096] Among them, u i SD2 、a i SD2 、y i SD2 are the input, state, and output of neuron i of striatum D2 channel, w CSD2 is the connection weight from the cortex to the striatum D2, λ is the influence of dopamine neurons, which has an inhibitory effect on the striatum D2, and ε SD2 is the striatum D2 output threshold, y i C is the output of the ith channel of the cerebral cortex;

[0097] (3) The mathematical model of the nucleus externa of the globus pallidus is:

[0098]

[0099] Among them, u i GPe 、a i GPe 、y i GPe are the input, state, and output of the neuron i in the nucleus externa of the globus pallidus, respectively, and w SD2GPe is the connection weight from striatum D2 to the nucleus externa pallidus, ε GPe is the output threshold of the globus pallidus external nucleus;

[0100] (4) The mathematical model of the subthalamic nucleus is:

[0101]

[0102] Where ui STN 、a i STN 、y i STN are the input, state, and output of the neuron i in the nucleus externa of the globus pallidus, respectively, and w GPeSTN is the connection weight between the nucleus externus pallidus and the subthalamic nucleus, ε STN is the output threshold of the subthalamic nucleus;

[0103] (5) The mathematical model of the globus pallidus kernel is:

[0104]

[0105] Where u i GPi 、a i GPi 、y i GPi are the input, state, and output of the neuron i in the globus pallidus nucleus, respectively, and w SD1GPi 、w STNGPi are the connection weights from striatum D1 and subthalamic nucleus to globus pallidus nucleus, ε GPi is the output threshold of the globus pallidus kernel.

[0106] In the above connection weights, the w CSD1 、w CSD2 、w STNGPi is positive, indicating an excitatory connection, w SD1GPi 、w SD2GPe 、w GPeSTN A negative value indicates an inhibitory connection.

[0107] In one embodiment, a quadruped robot gait control system based on a dual-level CPG inspired by the basal ganglia is provided, the system comprising:

[0108] The first module is used to construct the kinematic equations of the quadruped robot and verify the accuracy of the established model under the ADAMS simulation platform;

[0109] The second module is used to establish a two-layer CPG neural network model;

[0110] The third module is used to build a behavioral selection model of the basal ganglia. According to this model, the behavioral requests of the quadruped robot flow along their respective channels through the various nuclei of the basal ganglia, including the striatum, the globus pallidus external nucleus (GPe), and the subthalamic nucleus (STN), and finally reach the output nuclei (GPi) and the substantia nigra pars reticulata (SNr).

[0111] Regarding the specific limitations of the basal ganglia-inspired dual-level CPG quadruped robot gait control system, please refer to the limitations of the basal ganglia-inspired dual-level CPG quadruped robot gait control method, which will not be repeated here. Each module in the above-mentioned basal ganglia-inspired dual-level CPG quadruped robot gait control system can be fully or partially implemented by software, hardware, and a combination thereof. Each of the above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to each of the above modules.

[0112] The quadruped robot gait control method of the double-layer CPG inspired by the basal ganglia in the present invention has better adaptability to environmental changes than the single-layer CPG controller, and has a clearer structure, can achieve better control effects, and has strong robustness of the control system. It also uses the basal ganglia to select the gait of the quadruped robot, thereby enhancing its environmental adaptability.

[0113] The basic principles, main features, and advantages of the present invention are shown and described above. Those skilled in the art should understand that the present invention is not limited to the foregoing embodiments. The foregoing embodiments and descriptions are merely illustrative of the principles of the present invention. Various changes and modifications may be made to the present invention without departing from the spirit and scope of the present invention. Such changes and modifications are intended to fall within the scope of the present invention. The scope of protection claimed in the present invention is defined by the appended claims and their equivalents.

Claims

1. A gait control method for a quadruped robot based on a two-level CPG inspired by the basal ganglia, characterized by: The method comprises the following steps: Step 1: Construct the kinematic equations of the quadruped robot and verify the accuracy of the established model under the ADAMS simulation platform; Step 2, establish a two-layer CPG neural network model; Step 3: Construct a behavior selection model of the basal ganglia. According to this model, the quadruped robot's behavior requests flow along their respective pathways through the various nuclei of the basal ganglia, including the striatum, the globus pallidus externus (GPe), and the subthalamic nucleus (STN), and ultimately reach the output nuclei (GPi) and substantia nigra pars reticulata (SNr). The two-layer CPG neural network model in step 2 includes two parts: rhythm generation RG and pattern formation PF; The RG part is composed of a semi-central rhythm generator, including the RG neuron RG-E for extension and the RG neuron RG-F for flexion; the PF part includes four PF neurons, which control the four consecutive stages of swing, landing, stance and lift-off respectively; Specifically include: In step 2-1, the membrane potentials of the extensor and flexor neurons and motor neurons in the RG and PF parts are expressed as: The membrane potential of an interneuron is expressed as: Where, I Nap is the continuous sodium current, I K is the potassium rectifier current, I Leak is the leakage current, I SynE and I SynI are the synaptic excitatory and synaptic inhibitory currents, respectively, C is a neuronal capacitance, and V is the average membrane voltage of the neuron; The ion current formula is as follows: The voltage-dependent activation variables of potassium delayed rectifier and persistent sodium channels are expressed as: Voltage-dependent inactivation variable h NaP Calculated by the following differential equation: in: f(V i ) represents the average population activity and the output of neuron j, which is defined as a sigmoid function: Where, E {Na,K,Leak,SynE,SynI} is the corresponding reversal potential, is the maximum conductance of the corresponding ion channel, a ji is the gain of the excitatory synaptic input of neuron j relative to neuron i, b ji is the gain of neuron j relative to the inhibitory input of neuron i, c i is the gain of the tonic descending signal on the neuron, and w {1,2},i is the incoming feedback to neuron i {1,2} Gain; τ hNapmax is a time constant; V 1 / 2 is the half-activation voltage, k is the S-type function gain, V th is the threshold of each neuron; ∑ k a ki f(V k )+∑ m β mi f(V m ) shows the connections from other neurons to neuron i and creates an intrinsic mutual coordination; The Ia afferent feedback from the extensor motor neurons as a stretch reflex is effective for supporting weight. Specifically, when the leg is weight-bearing, the following feedback signals from the hip extensors are input to the corresponding motor neurons: Where, F load is the load on each leg measured by the force sensors mounted on its legs, k v is the gain, I norm2 is the driving current of the legs, v norm2 is the driving voltage of the legs; In step 2-2, the connection parameters between RG and PF neurons are responsible for forming the basic rhythm, and the connection parameters between PF neurons and motor neurons are responsible for creating the leg trajectory in four consecutive phases.

2. The basal ganglia-inspired dual-level CPG quadruped robot gait control method according to claim 1, characterized in that: The kinematic equations of the quadruped robot are constructed as described in step 1, specifically including: Construct the forward solution of the kinematic equations of the quadruped robot: Construct a quadruped robot and inversely solve the kinematic equations: Where L1 is the length from the knee joint to the foot, L2 is the length from the hip joint to the knee joint, θ1 is the angle between L1 and the ground, θ2 is the angle between L2 and the extension line of L1, x and y are the posture in the x direction and the posture in the y direction respectively, atan2(y,x) is the inverse tangent of x and y, s2 is sinθ2, and c2 is cosθ2.

3. The quadruped robot gait control method of the double-level CPG inspired by basal ganglia according to claim 1 or 2, characterized in that: Each leg of the quadruped robot includes a side swing joint, a hip joint and a knee joint, and its external structure and the motor model used are the same.

4. The basal ganglia-inspired dual-level CPG quadruped robot gait control method according to claim 1, characterized in that: Step 2 also includes: Step 2-3, based on the connection parameters described in step 2-2 and based on its characteristics, performs parameter adjustment, including the following steps: (1) Suspend the quadruped robot so that its legs swing in the air, and manually adjust the parameters of the CPG model so that each leg can produce a loose trajectory and stable oscillation in each of the four consecutive stages; (2) Place the quadruped robot on the ground and use the adjustment software to adjust the parameters so that the quadruped robot can walk on flat terrain; (3) Manually modify parameters so that the motor neuron activation patterns and leg trajectories of the quadruped robot are the same as those of quadruped mammals.

5. The basal ganglia-inspired dual-level CPG quadruped robot gait control method according to claim 4, characterized in that: Step 3 constructs a behavior selection model for the basal ganglia. According to this model, the quadruped robot's behavior requests flow along their respective channels through the various nuclei of the basal ganglia, including the striatum, the globus pallidus external nucleus (GPe), and the subthalamic nucleus (STN), and finally reach the output nuclei (GPi) and the substantia nigra pars reticulata (SNr). Specifically, Assuming that a suitable behavior needs to be selected from I behavioral channels, the various nuclei of the basal ganglia are also divided into I channels accordingly. Each channel is represented by a leaky integration neuron. The leaky integration neuron model is: Where x is the state of the neuron, y is the output of the neuron, k, m, and ε are all model parameters, and H is the step function; For the direct and indirect pathway model, the mathematical models of the various nuclei of the basal ganglia are as follows: (1) The mathematical model of striatum D1 is: Where u i SD1 、a i SD1 、y i SD1 are the input, state, and output of neuron i in striatum D1 channel, w CSD1 is the connection weight from the cortex to the striatum D1, λ is the influence of dopamine neurons, which has an incentive effect on the striatum D1, and ε SD1 Output threshold for striatum D1; (2) The mathematical model of striatum D2 is: Among them, u i SD2 、a i SD2 、y i SD2 are the input, state, and output of neuron i of striatum D2 channel, w CSD2 is the connection weight from the cortex to the striatum D2, λ is the influence of dopamine neurons, which has an inhibitory effect on the striatum D2, and ε SD2 is the striatum D2 output threshold, y i C is the output of the ith channel of the cerebral cortex; (3) The mathematical model of the nucleus externa of the globus pallidus is: Among them, u i GPe 、a i GPe 、y i GPe are the input, state, and output of the neuron i in the nucleus externa of the globus pallidus, respectively, and w SD2GPe is the connection weight from striatum D2 to the nucleus externa pallidus, ε GPe is the output threshold of the globus pallidus external nucleus; (4) The mathematical model of the subthalamic nucleus is: Where u i STN 、a i STN 、y i STN are the input, state, and output of the neuron i in the nucleus externa of the globus pallidus, respectively, and w GPeSTN is the connection weight between the nucleus externus pallidus and the subthalamic nucleus, ε STN is the output threshold of the subthalamic nucleus; (5) The mathematical model of the globus pallidus kernel is: Where u i GPi 、a i GPi 、y i GPi are the input, state, and output of the neuron i in the globus pallidus nucleus, respectively, and w SD1GPi 、w STNGPi are the connection weights from striatum D1 and subthalamic nucleus to globus pallidus nucleus, ε GPi is the output threshold of the globus pallidus kernel.

6. The basal ganglia-inspired dual-level CPG quadruped robot gait control method according to claim 5, characterized in that: The w CSD1 、w CSD2 、w STNGPi is positive, indicating an excitatory connection, w SD1GPi 、w SD2GPe 、w GPeSTN A negative value indicates an inhibitory connection.

7. A quadruped robot gait control system based on a two-level CPG inspired by basal ganglia method according to any one of claims 1 to 6, characterized in that: The system comprises: The first module is used to construct the kinematic equations of the quadruped robot and verify the accuracy of the established model under the ADAMS simulation platform; The second module is used to establish a two-layer CPG neural network model; The third module is used to build a behavioral selection model of the basal ganglia. According to this model, the behavioral requests of the quadruped robot flow along their respective channels through the various nuclei of the basal ganglia, including the striatum, the globus pallidus external nucleus (GPe), and the subthalamic nucleus (STN), and finally reach the output nuclei (GPi) and the substantia nigra pars reticulata (SNr).

Images