CN106914901B - Layered construction method of bionic robot control network - Google Patents

Layered construction method of bionic robot control network Download PDF

Info

Publication number
CN106914901B
CN106914901B CN201710323731.3A CN201710323731A CN106914901B CN 106914901 B CN106914901 B CN 106914901B CN 201710323731 A CN201710323731 A CN 201710323731A CN 106914901 B CN106914901 B CN 106914901B
Authority
CN
China
Prior art keywords
neuron
output
control network
oscillator
motion control
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN201710323731.3A
Other languages
Chinese (zh)
Other versions
CN106914901A (en
Inventor
刘强
赵琳琳
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Huaihai Institute of Techology
Original Assignee
Huaihai Institute of Techology
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Huaihai Institute of Techology filed Critical Huaihai Institute of Techology
Priority to CN201710323731.3A priority Critical patent/CN106914901B/en
Publication of CN106914901A publication Critical patent/CN106914901A/en
Application granted granted Critical
Publication of CN106914901B publication Critical patent/CN106914901B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • B—PERFORMING OPERATIONS; TRANSPORTING
    • B25—HAND TOOLS; PORTABLE POWER-DRIVEN TOOLS; MANIPULATORS
    • B25J—MANIPULATORS; CHAMBERS PROVIDED WITH MANIPULATION DEVICES
    • B25J9/00—Programme-controlled manipulators
    • B25J9/16—Programme controls
    • B25J9/1602—Programme controls characterised by the control system, structure, architecture
    • B25J9/161—Hardware, e.g. neural networks, fuzzy logic, interfaces, processor

Landscapes

  • Engineering & Computer Science (AREA)
  • Automation & Control Theory (AREA)
  • Physics & Mathematics (AREA)
  • Artificial Intelligence (AREA)
  • Evolutionary Computation (AREA)
  • Fuzzy Systems (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Robotics (AREA)
  • Mechanical Engineering (AREA)
  • Manipulator (AREA)

Abstract

A layered construction method of a bionic robot control network comprises the following steps: according to the speed and the characteristics of the movement gait, the movement gait of the bionic robot is grouped or classified; constructing a certain layer of motion control network of the bionic robot according to a certain group or class of motion gaits; combining all layers of motion control networks together to form the whole bionic robot motion control network; the upper nerve center is used for generating control signals, so that the corresponding part of the whole motion control network is activated, inactivated and saturated to generate various motion gaits of the bionic robot. The invention provides a method for constructing a bionic robot control network in a layered mode and then combining the bionic robot control network to construct the whole bionic robot control network, and solves the problem of constructing the whole bionic robot control network. The method can enable the bionic robot to generate more abundant and flexible movement gaits which accord with the characteristics of animal movement, and particularly can generate rhythmic movement and non-rhythmic movement.

Description

Layered construction method of bionic robot control network
Technical Field
The invention belongs to the technical field of robot control, and relates to a method for constructing a layered robot motion control neural network.
Background
Central Pattern Generators (CPG) are an important component of the neuromotor control neural network that can produce rhythmic output without feedback of external sensory information. In order to utilize the central mode generator to control the bionic robot. Many researchers have conducted research on the construction of a biomimetic robot motion control neural network. These central pattern generator construction methods can be broadly divided into three categories: an oscillator CPG, a biological neuron CPG and a connection CPG. The oscillator CPG is widely applied to the control of the bionic robot due to the characteristics of simple structure, small operand and the like, but the existing oscillators have some defects in different degrees, so that the construction of a motion control neural network of the bionic robot is difficult. Among the oscillators, an oscillator which is composed of two integrator-missing neurons with fatigue characteristics and is formed by mutual suppression and proposed by Matsuoka becomes one of oscillators widely applied in the construction of a bionic robot motion control neural network. However, the oscillator still has more problems in the aspect of the construction of the motion control neural network, such as the inability to include excitatory connections, the difficult construction of the network topology, and the like. The invention discloses a novel method for designing a bionic robot motion control network (NMN), namely a novel neuron oscillator-201510507029.3 and a construction method of a bionic robot motion control neural network-201510632804.8. however, the method has a good effect on the construction of a certain gait control network (or a certain type of gait control network), and is very difficult to realize the construction of the bionic robot overall control network of various complex motion gaits, such as various rhythmic motion gaits and non-rhythmic motion gaits.
Disclosure of Invention
The invention aims to provide a layered construction method of a bionic robot control network aiming at the defects of the prior art, and the method can enable the bionic robot to generate motion gaits which are richer, more flexible and accord with the characteristics of animal motion.
The technical problem to be solved by the present invention is achieved by the following technical means. The invention relates to a layered construction method of a bionic robot control network, which is characterized by comprising the following steps:
(1) according to the speed and the characteristics of the movement gait, the movement gait of the bionic robot is grouped or classified;
(2) constructing a certain layer of motion control network of the bionic robot according to a certain group or class of motion gaits;
(3) combining all layers of motion control networks together to form the whole bionic robot motion control network; the upper nerve center is used for generating control signals, so that the corresponding part of the whole motion control network is activated, inactivated and saturated to generate various motion gaits of the bionic robot.
The invention relates to a layered construction method of a bionic robot control network, which adopts the further preferable technical scheme that in the step (2), the constructed motion control network layer is divided into a rhythm motion control network layer and a non-rhythm motion control network layer.
The invention relates to a layered construction method of a bionic robot control network, which further adopts the preferable technical scheme that a rhythmic motion control network layer and a non-rhythmic motion control network layer are formed by a motion control network layer generating certain motion gait.
The invention further discloses a layered construction method of a bionic robot control network, which adopts the preferable technical scheme that in the step (3), a synaptic connection relation between motion control networks is established according to needs.
The layered construction method of the bionic robot control network further adopts the preferable technical scheme that in the step (3), the center is a brainstem.
The invention relates to a layered construction method of a bionic robot control network, which further adopts the preferable technical scheme that the construction method of the motion control network of each layer comprises the following steps:
the method is based on a novel neuron oscillator;
the novel neuron oscillator is characterized in that a neuron model is established, and then two neurons are connected with each other through inhibitory synapses to form an oscillator model; the neuron model is formed by adding output saturation and self-excitability characteristics on the basis of a leaky integrator neuron model with fatigue characteristics; the output of the neuron model is represented by a nonlinear function, the nonlinear function satisfies that when x is larger than or equal to theta, the output has a saturation characteristic, and when x is smaller than theta, the neuron does not output;
the method for constructing the motion control neural network is characterized in that a novel neuron oscillator is used as the motion control neural network of a robot joint, the output of one neuron is used as a flexor control signal of the robot joint, the output of the other neuron is used as an extensor control signal of the joint, and then the connection relation between the robot joint oscillators is established by utilizing the connection relation of inhibition and excitability according to the motion relation between joints of the robot and the topological structure characteristics of an actual bionic biological neural control loop.
Based on the construction method of the bionic robot motion control network layer, not only a rhythm motion control network layer but also a non-rhythm motion control network layer can be constructed, and the difference lies in that the parameters of the selected motion control network layer are different.
The neuron model adopts one of the following two differential equation sets:
Figure BDA0001290149450000021
Figure BDA0001290149450000022
wherein x is the membrane potential of the neuron; y is the output of the neuron; s is all external inputs received by the neuron; a is the connection weight of self-excitability feedback received by the neuron, and a is more than 0; tau is a time constant related to the membrane potential of the neuron, and tau is more than 0; time constant related to fatigue process of the neuron, gamma is more than 0; x' is a variable reflecting the degree of neuronal fatigue; b is fatigue strength of neuron, b is more than 0; theta is the output threshold of the neuron,
Figure BDA0001290149450000023
is an upper bound of the neuron output, an
Figure BDA0001290149450000024
Epsilon and sigma are constant coefficients, epsilon is more than 0 and sigma is more than 0; λ is the saturation coefficient of the neuron output.
Two neurons are mutually inhibited, each neuron having a self-excitatory link; the concrete model is as follows:
Figure BDA0001290149450000025
Figure BDA0001290149450000031
in the formula, xiIs the membrane potential of the ith neuron; y isiIs the output of the ith neuron; siAn external input received for an ith neuron; a isij(j ∈ {1,2}, j ≠ i) is the connection weight between neurons, aij＜0；aiiConnection weights for self-excitatory feedback received by the ith neuron; tau isiTime constant, τ, associated with membrane potential of ith neuroni＞0；γiWhen associated with i-th neuron fatigue processConstant of between, gammai＞0；x′iA variable reflecting the degree of fatigue of the ith neuron; biIs the fatigue strength of the ith neuron, bi＞0；θiIs the output threshold of the ith neuron,
Figure BDA0001290149450000032
is an upper bound of the output of the ith neuron, an
Figure BDA0001290149450000033
εiAnd σiIs a constant coefficient of ∈i> 0 and σi＞0；λiIs the saturation coefficient of the ith neuron output;
the oscillator equilibrium state
Figure BDA0001290149450000034
Satisfy the requirement of
Figure BDA0001290149450000035
The oscillator can generate either an oscillating output or a non-oscillating output, the external input s being at this timeiThe value range of (i ═ 1,2) is as follows:
Figure BDA0001290149450000036
when the oscillator generates oscillation output, the parameters of the oscillator meet the following conditions:
Figure BDA0001290149450000037
Figure BDA0001290149450000038
when the oscillator generates non-oscillation output, the parameters of the oscillator meet the following conditions:
(1)
Figure BDA0001290149450000039
(2)
Figure BDA00012901494500000310
(3)
Figure BDA00012901494500000311
or
(1)
Figure BDA00012901494500000312
(2)
Figure BDA00012901494500000313
(3)σi≥1，(i＝1，2)
According to the conditions of the oscillation output and the non-oscillation output of the oscillator, the oscillation output and the non-oscillation output of the oscillator can be adjusted by adjusting the self-excitation coefficient aiiAnd fatigue coefficient biTo perform a handover;
the oscillator satisfies s when the input is inputi＜εiθiWhen (i ═ 1,2), the output of the oscillator is inactive, when
Figure BDA0001290149450000041
The output of the oscillator is saturated;
wherein the oscillation frequency and response speed of the oscillator are determined by a time constant tau associated with the membrane potential of the neuroniI-1, 2 and a time constant γ associated with the process of neuronal fatigueiI is adjusted to 1, 2;
wherein the saturated output and the non-activated output of the oscillator and the magnitude of the rhythmic output and the non-rhythmic output of the oscillator can be input by the external input s of the oscillatoriI is adjusted to 1, 2.
The model of the bionic robot motion control neural network formed by the mutual connection of the oscillators is as follows:
Figure BDA0001290149450000042
Figure BDA0001290149450000043
wherein n is the number of neurons; x is the number ofiIs the membrane potential of the ith neuron; y isiIs the output of the ith neuron; siAn external input received for an ith neuron; a isij(j ∈ {1, …, n }, j ≠ i) is the connection weight between neurons, aij> 0 denotes an excitatory linkage, aij< 0 indicates inhibitory attachment; a isiiConnection weights for self-excitatory feedback received by the ith neuron; tau isiTime constant, τ, associated with membrane potential of ith neuroni＞0；γiTime constant, gamma, associated with the i-th neuron fatigue processi＞0；x′iA variable reflecting the degree of fatigue of the ith neuron; biIs the fatigue strength of the ith neuron, bi＞0；θiIs the output threshold of the ith neuron,
Figure BDA0001290149450000044
is an upper bound of the output of the ith neuron, an
Figure BDA0001290149450000045
εiAnd σiIs a constant coefficient of ∈i> 0 and σi＞0；λiIs the saturation coefficient of the ith neuron output;
the bionic robot motion control neural network is used for controlling the neural network when the equilibrium state of a part of neurons exists in the neural network
Figure BDA0001290149450000046
Satisfy the requirement of
Figure BDA0001290149450000047
Namely, it is
Figure BDA0001290149450000048
When the neuron is in the first state, the neuron can generate oscillation output and non-oscillation output, and the oscillation output and the non-oscillation output are equal to each otherExternal input s corresponding to the state of balancei，i∈Λ3The value range is as follows:
Figure BDA0001290149450000049
wherein omegai＝{j|aij＞0，j∈Λ3，j≠i}；
Figure BDA00012901494500000410
This part of the neuron, i.e. i ∈ Λ3The conditions that need to be satisfied for the neurons to generate the oscillation output are as follows:
Figure BDA00012901494500000411
Figure BDA00012901494500000412
this part of the neuron, i.e. i ∈ Λ3The conditions that need to be satisfied for the neuron to produce a non-oscillating output are as follows:
(1)
Figure BDA0001290149450000051
(2)
Figure BDA0001290149450000052
(3)
Figure BDA0001290149450000053
or
(1)
Figure BDA0001290149450000054
(2)
Figure BDA0001290149450000055
(3)σi≥1，(i∈Λ3)
This part of the neurons, i.e. i ∈ Λ, is conditioned on the oscillating and non-oscillating outputs mentioned above3Can be switched between oscillatory and non-oscillatory outputs by varying the self-excitation coefficient aiiAnd fatigue coefficient biIs implemented, and the output of the part of the neuron follows the external input si，i∈Λ3Is changed;
this part of the neuron, i.e. i ∈ Λ3The oscillation frequency and response speed of the neuron can be controlled by a time constant tau associated with the membrane potential of the neuroni，i∈Λ3Time constant gamma associated with neuronal fatigue processesi，i∈Λ3Carrying out adjustment;
the bionic robot motion control neural network is used for controlling the balance state of a part of neurons
Figure BDA0001290149450000056
Satisfy the requirement of
Figure BDA0001290149450000057
Then, the part of neurons can be collected
Figure BDA0001290149450000058
Indicating and they are inactive, their external input si，i∈A1The value range is as follows:
Figure BDA0001290149450000059
the bionic robot motion control neural network is used for controlling the balance state of a part of neurons
Figure BDA00012901494500000510
Satisfy the requirement of
Figure BDA00012901494500000511
When, this part of neuronsCan use the set
Figure BDA00012901494500000512
Represent and they are saturated, with its external input si，i∈Λ2The value range is as follows:
Figure BDA00012901494500000513
the derivation of the above conclusion is as follows:
for convenience of presentation, assume
Figure BDA00012901494500000514
Figure BDA00012901494500000515
Wherein the content of the first and second substances,
Figure BDA00012901494500000516
the equilibrium state of the neural network (1) is controlled for the movement.
Order to
Figure BDA00012901494500000517
And
Figure BDA00012901494500000518
the equilibrium state of the motion control neural network (1) can be calculated
Figure BDA00012901494500000519
Comprises the following steps:
Figure BDA00012901494500000520
Figure BDA0001290149450000061
Figure BDA0001290149450000062
when the neural network (1) is in an equilibrium state
Figure BDA0001290149450000063
Figure BDA0001290149450000064
And is
Figure BDA0001290149450000065
Time, external input siThe range of (i ∈ {1, …, n }) can be calculated in the following manner.
Figure BDA0001290149450000066
Figure BDA0001290149450000067
Figure BDA0001290149450000068
Wherein omegai＝{j|aij＞0，j∈Λ3，j≠i}；
Figure BDA0001290149450000069
Putting the motion control neural network (1) in equilibrium
Figure BDA00012901494500000610
Figure BDA00012901494500000611
For convenience of presentation, let Λ be1＝{1，…，l}、Λ2＝{l+1，…，m}、Λ3The system that can be linearized is given by m +1, …, n:
Figure BDA00012901494500000612
wherein the content of the first and second substances,
Figure BDA00012901494500000613
Figure BDA00012901494500000614
Figure BDA0001290149450000071
Figure BDA0001290149450000072
Figure BDA0001290149450000073
Figure BDA0001290149450000074
since the matrix trace is equal to the sum of the matrix characteristic roots. From the matrix A, when Λ is observed3When the value is not equal to phi, the sufficient condition that the linearization system (5) has a positive real part characteristic root is as follows:
Figure BDA0001290149450000075
according to the Lyapunov theorem, the equilibrium state of the neural network (1) is known for any initial state when equation (6) is satisfied
Figure BDA0001290149450000076
Are all unstable and the corresponding inputs are equations (2) - (4).
Since the solution of the motion control neural network (1) is uniquely present and bounded, it can be derived that the motion control neural network (1) has an oscillating output when the motion control neural network (1) satisfies equation (6).
According to Gerschgorin Circle Theorem, the linearized model (5) can be easily analyzed when the parameters of the motion control neural network (1) meet
(1)
Figure BDA0001290149450000081
(2)
Figure BDA0001290149450000082
(3)
Figure BDA0001290149450000083
Or
(1)
Figure BDA0001290149450000084
(2)
Figure BDA0001290149450000085
(3)σi≥1，(i∈Λ3)
The linearized system of the motor control neural network (1) is asymptotically stable.
According to the first principle of Lyapunov, for any initial state, the motion control neural network (1) is in a state of equilibrium
Figure BDA0001290149450000086
And
Figure BDA0001290149450000087
is asymptotically stable, i.e., the output of the motion control neural network (1) is non-oscillating, and the corresponding inputs are equations (2) - (4).
The invention provides a method for constructing a bionic robot control network in a layered mode and then combining the bionic robot control network to construct the whole bionic robot control network, and solves the problem of constructing the whole bionic robot control network. Compared with the prior art, the construction method of the layered motion control network of the bionic robot can enable the bionic robot to generate more abundant and flexible motion gaits which accord with the characteristics of animal motion, and particularly can generate rhythmic motion and non-rhythmic motion.
Drawings
Figure 1 is a schematic structural view of a salamander robot;
fig. 2 is a schematic structural diagram of a low-frequency walking control network of a salamander robot, wherein the high-frequency swimming control network is the same as a body control network part of the low-frequency walking control network, and the thin circles are extensor neurons and the thick circles are flexor neurons;
figure 3 is a gait simulation result diagram of forward and backward walking of the salamander robot and switching of the salamander robot (A), except a transition oscillator, output curve diagrams of all body oscillators (B), a motion simulation diagram of a forward walking period of the salamander robot entity model (C), except a knee oscillator, output curve diagrams of all limb oscillators (D), and a motion simulation diagram of a backward walking period of the salamander robot entity model (D);
figure 4 is a gait simulation result diagram of forward walking and swimming of the salamander robot and switching of the salamander robot (A), except a transition oscillator, output curve diagrams of all body oscillators (B), a motion simulation diagram of a forward walking cycle of the salamander robot entity model (C), except a knee oscillator, output curve diagrams of all forward walking network limb oscillators (D), and a motion simulation diagram of a swimming cycle of the salamander robot entity model;
figure 5 when the left and right asymmetric input of the body network is gradually increased, the simulation graph (A) of the swimming turning motion of the salamander robot is input into the asymmetric external input signal graph (B) of the body network of the neck and trunk, except the transition oscillator, all the output graphs (C) of the body oscillators are the simulation graph of the swimming turning motion of the salamander robot physical model;
fig. 6 simulation diagrams (a) and (B) of forward walking turning gait of an salamander robot are input to a body network and a right forelimb swinging oscillator (C) when an asymmetric input signal is input to the body network and the right forelimb swinging oscillator, all remaining body oscillator output diagrams (D) except a transition oscillator when an asymmetric input signal is input to the body network and the right forelimb swinging oscillator, all remaining forward walking network limb oscillator output diagrams (E) when an asymmetric input signal is input only to the right forelimb swinging oscillator, except a knee oscillator, a forward walking turning simulation diagram (F) of the salamander robot entity model when an asymmetric input signal is input only to the body network, a forward walking turning simulation diagram (G) of the salamander robot entity model when an asymmetric input signal is input only to the body network and the right forelimb swinging oscillator, a forward walking and turning simulation diagram of the salamander robot entity model;
figure 7 is a schematic structural view of a salamander robot;
figure 8 is a schematic diagram of a salamander robot rhythm control network structure, the high frequency control network is the same as the body control network part of the low frequency control network, the thin circles are extensor neurons, and the thick circles are flexor neurons;
figure 9 is a schematic diagram of a non-rhythm control network structure of a salamander robot, wherein the thin circles are extensor neurons and the thick circles are flexor neurons.
FIG. 10 is a schematic diagram of the network coordination control of rhythms and non-rhythms;
figure 11 is a diagram of simulation results of walking motion of the salamander robot;
figure 12 is a diagram of simulation results of salamander robot walking and swimming and gait switching.
Detailed Description
The present invention will be described in detail below with reference to the accompanying drawings and specific embodiments.
The specific implementation mode is as follows:
embodiment 1, a hierarchical construction method of a bionic robot control network, comprising the following steps:
1) according to the characteristics of speed and movement gait, the movement gait of the bionic robot is grouped or classified;
2) constructing a certain layer of motion control network of the bionic robot according to a certain group (or class) of motion gaits; these motion control network layers may be divided into a rhythmic motion control network layer and a non-rhythmic motion control network layer; the rhythmic motion and non-rhythmic motion control network layer can be formed by a motion control network which generates certain motion gait;
3) combining the networks together, and if necessary, properly establishing a synaptic connection relationship between the networks to form the whole bionic robot motion control network; the upper nerve center (such as brainstem) is used for generating control signals, so that the corresponding part of the whole motion control network is activated, inactivated and saturated to generate various motion gaits of the bionic robot.
Embodiment 2, a layered construction method experiment of a bionic robot motion control network:
the layering method of the bionic robot motion control network is different from that of different bionic robots. The following describes the layering process of the motion control network for different animal bionic robots:
(1) layered construction of motion control network of finfish bionic robot
Research shows that the finfish mainly adopts pectoral fins and the like to carry out rhythm propulsion movement when moving at low speed, and the swimming speed of the finfish is gradually increased along with the increase of the flapping frequency of the pectoral fins; when the swimming speed of the fin fish increases to a certain degree, the pectoral fin keeps still against the body, and the fin fish mainly adopts the rhythmic swinging motion of the body and the tail fin to carry out high-speed swimming motion; in the motional states such as rest and starting, when the posture of the radial fin fish is kept and adjusted and the radial fin fish is started, the motion of the radial fin fish is also subjected to more non-rhythmic motion. Therefore, the motion control network of the finfish bionic robot can be divided into a low-speed and high-speed rhythm swimming motion control network layer and a non-rhythm motion control network layer, so that the motion control network of the whole bionic robot is formed.
(2) Hierarchical construction of motion control network of salamander bionic robot
Research shows that the salamander mainly moves at low speed through the mutual coordination of the body and limbs when moving on the land; when the salamander goes from the land to the water, the limbs lean backwards to keep still against the body and move at a high speed through the swinging motion of the body; the movement of the salamander has more non-rhythmic movements (such as predation, monitoring, turning in place and the like) besides rhythmic movements, and in addition, the salamander also needs to keep a certain rigidity of the body to resist the action of gravity during the rhythmic walking process on the land. Therefore, the motion control network of the salamander can be divided into a low-speed walking rhythm motion control network layer, a high-speed swimming rhythm motion control network layer and a non-rhythm motion control network layer to form the motion control network of the whole salamander bionic robot.
(3) Layered construction of quadruped bionic robot motion control network
The gait of the quadruped, such as a horse, a dog, a cat and the like, also changes gradually along with the increase of the movement speed, and the gait of the quadruped can be divided into the following steps along with the movement speed from low to high: walking, scouting, jogging and spriting. In each gait, relative motion phases of four limbs of the animal are different, motion control networks with different structures are required to be constructed for realization, and the relative motion phases are respectively used as rhythm control network layers of the quadruped bionic robot. The non-rhythmic motion control network layer of the bionic robot is established by considering the non-rhythmic motion of the quadruped animals and the action of resisting gravity. And combining all the motion control network layers together to form the motion control network of the whole bionic robot.
2. Hierarchical construction of salamander bionic robot rhythm control network
Salamanders are used as a representative species for the evolution of organisms from water to land, and the construction of a bionic robot motion control network related to the salamanders is important and representative in the construction of various bionic robot motion control networks. Therefore, the layered construction process of the bionic robot motion control network according to the invention is described below by the construction of the salamander bionic robot motion control network.
According to the structure of the salamander robot (as shown in figure 1), a control network with two mutually-inhibited neurons is constructed as an oscillator for joint control, wherein the output of one neuron is used as an extensor control signal, and the output of the other neuron is used as a flexor control signal. Considering the characteristics that the swinging speed of the body of the salamander can be obviously increased after the salamander is converted into swimming from land walking, the rhythm control network of the salamander robot is divided into a high-frequency swimming control network and a low-frequency land walking control network to respectively carry out swimming and walking control of the salamander robot. Wherein the low frequency control network is shown in fig. 2, and the high frequency control network is the same as the body control network portion of the low frequency control network.
(1) Low frequency walk control network construction
As can be seen from fig. 2, the low-frequency control network of the salamander robot can be divided into two parts, namely a body control network and a limb control network. The body control network of the salamander robot comprises eleven oscillators, wherein each body joint is controlled by one oscillator. In addition, in order to coordinate the relationship between the body control network and the limb control network, two transitional oscillators are added to the shoulder and the crotch of the body control network. The salamander robot rhythm control network uses descending dominant excitatory connections to establish the connection relationship between body oscillators, as shown in figure 2. Control network of salamander robot four limbs is similar, and every leg has three control oscillator, and the back and forth swinging motion of leg is being controlled to the swing oscillator, lifts up and falls down of leg oscillator control leg, and the knee oscillator is controlling the swinging motion of knee joint. When the salamander walks forward, the four legs of the salamander are in diagonal synchronization, and the swinging motion between the two forelimbs (or the two hind limbs) is in opposite phase. In order to coordinate the movements of the fore and aft limbs by using a body network to realize a limb diagonal synchronous movement gait, the flexor neurons of the body shoulder transition oscillator are connected to the flexor neurons of the right forelimb swing oscillator through inhibitory projections and to the flexor neurons of the left forelimb swing oscillator through excitatory synapses. The connection between extensor neurons and the forelimbs of the shoulder transition oscillator and the connection between flexor neurons and the forelimbs are symmetrical about the longitudinal section of the body of the salamander. The connection between the front limb swing oscillator and the body oscillator functions to move the left and right front limbs in anti-phase with each other. Since the body of the salamander presents a standing wave when walking forward, extensor neurons that swing the forelimb oscillators are connected to ipsilateral neurons of the trunk oscillator in proximity thereto through excitatory synapses. The connection between the back limb swinging oscillator and the body crotch transition oscillator is established by utilizing the standing wave characteristic of the salamander body and considering the diagonal synchronization of the salamander limb movement and the opposite phase movement between the left back limb and the right back limb, the connection relation is similar to the connection relation between the front limb swinging oscillator and the shoulder transition oscillator, and only the synaptic characteristic is correspondingly changed. The extensor neurons of the hind limb oscillatory oscillators are connected to ipsilateral neurons proximal to the caudal oscillators via excitatory synapses, and the flexor neurons are connected to ipsilateral neurons proximal to the trunk oscillators via excitatory synapses. Because the leg needs to be lifted to reduce the influence of ground friction on the forward swing of the leg in the forward swing process of the salamander leg, the leg needs to be pressed down to increase the grip when swinging backwards, and therefore, the output of the flexor neurons of the limb swing oscillator is connected to the flexor neurons of the leg lifting oscillator through excitatory synapses.
In addition to walking forward, the salamander can also walk backward, and the motion of its limbs is diagonal synchronous. Therefore, similar to walking forward, the salamander robot adds a swinging oscillator and a leg-lifting oscillator to each limb. In order to make extensor neurons of the limb swing oscillator activate before flexor neurons, excitatory synapses from ipsilateral neurons of the body shoulder transition oscillator and inhibitory synapses from contralateral neurons of the body shoulder transition oscillator are connected to extensor neurons of the backward walking network shoulder joint swing oscillator. Excitatory synapses from contralateral neurons of a body crotch transition oscillator and inhibitory synapses from ipsilateral neurons of the body crotch transition oscillator are connected to extensor neurons of a walk-back network crotch joint swing oscillator. In addition, for the walk-back network of each limb, the extensor neurons of the oscillatory oscillator of each limb are connected to the flexor neurons of the leg-raising oscillator by unidirectional excitatory synapses. In the backward walking process, when the limbs swing backward, the limbs are lifted up simultaneously; instead, the limb grabs the ground, thereby enabling backward walking. Similar to the control of the brain stem of the salamander on the spinal cord control network, various rhythm movement gaits can be generated by planning the external input signal of the salamander robot control network.
(2) High frequency swimming motion control network
When swimming at high frequency, the salamander generates propulsive force through the rhythm swing motion of health, swimming forward. Therefore, the high-frequency swimming motion control network of the salamander mainly comprises a body control network part, the structure of which is the same as that of the low-frequency walking motion control network. The low-frequency walking motion control network layer and the high-frequency swimming motion control network layer are combined together to form the whole rhythm motion control network of the robot.
(3) Gait generation and control of salamander bionic robot
By utilizing the motion control network, the salamander bionic robot can generate forward walking, backward walking, swimming, turning motion and mutual conversion. When walking forward, the high-frequency swimming network and the backward walking network of four limbs are inactivated, and the salamander bionic robot walks forward under the control of the low-frequency walking network. When the forward walking network of the limbs is inactivated and the backward walking network is activated, the salamander robot walks backwards. When the body part of the low-frequency walking network is inactivated, the backward walking network of the limbs is inactivated, extensor neurons of the forward walking network are saturated and flexor neurons are inactivated, so that the limbs of the salamander robot lean against the body backwards, and the body generates rhythm swinging to realize swimming movement. In swimming motion and walking motion process, when asymmetric input is applied to the body network or the four limbs network of salamander bionic robot, salamander robot alright in order to produce corresponding turning motion.
For the rhythmic motion control network of the salamander robot, the parameters of the oscillators of all low-frequency control networks are assumed to be the same, and the parameters of the oscillators are selected as follows: tau isi＝0.04、γi＝0.044、εi＝2、σi＝0.1、aii＝2.12、bi＝0.08、θi＝-0.2、
Figure BDA0001290149450000111
And aii+1＝-0.05、ai+1i-0.03(i ═ 1, 3, …, 45). For the connection weight between oscillators, the excitability weight between the body oscillators is selected to be 0.03, the bodyThe weight of excitatory connection between the body oscillator and the limb oscillator is 0.1, the weight of inhibitory connection between the body oscillator and the limb oscillator is-0.1, and the weight of excitatory connection between the limb oscillator is 0.1. Except for tauiAnd gammaiIn addition, the oscillator parameters of the high frequency swim control network are the same as the oscillator parameters of the low frequency control network. For high frequency control networks, τ is seti0.02 and γi＝0.022。
According to the design step (2), activation, inactivation and saturation output of the salamander robot rhythm control network can be realized by changing external input of the control network. For swim, forward-walk, and backward-walk rhythm control networks, when only neuron i is in inactive or saturated output, while the other neurons are in rhythmic output, the inactive external input s of neuron iiAnd saturated external input siThe lower bounds of (a) are shown in tables 1,2 and 3.
Based on the parameters of the motion control network and the generation method of the motion gait, the external input of the motion control network of the salamander robot during walking forwards, walking backwards and swimming is selected as shown in tables 4 and 5. The external input of the turning gait motion is added with corresponding offset on the basis of the external input of the corresponding forward walking and swimming gaits. Under the action of the parameters and external input, the simulation results of forward walking, backward walking, swimming, turning and mutual conversion of the salamander robot are shown in figures 3-6.
TABLE 1 external input s for inactivation of Forward Walking rhythm control network neurons iiUpper bound and saturated external input siLower bound
Figure BDA0001290149450000112
Figure BDA0001290149450000121
TABLE 2 external inputs s for inactivation of Backward-Walking rhythm control network neurons iiUpper bound and saturated external inputsiLower bound
Figure BDA0001290149450000122
TABLE 3 external inputs s for inactivation of neural elements i of the swim rhythm control networkiUpper bound and saturated external input siLower bound
Figure BDA0001290149450000123
Figure BDA0001290149450000131
TABLE 4 external inputs during Forward and Backward gait
Figure BDA0001290149450000132
TABLE 5 external input during Forward Walking and swimming gait
Figure BDA0001290149450000133
As can be seen from fig. 3-6, the salamander robot can produce the motion gait of walking forward, walking backward, swimming, turning and interconversion and so on, and they are consistent with the motion gait of the actual salamander.
3. Salamander robot motion control network construction comprising non-rhythmic motion control network
For the above embodiment example 2, in order to keep the salamander bionic robot standing and having a certain rigidity of the body during walking, the body and the limb muscles have a certain rigidity, so that the salamander bionic robot can perform rhythmic walking movement under the control of the above rhythmic movement control network. However, because the inherent rigidity of salamander robot body and limb muscle is not controllable, consequently, the salamander robot can only keep standing, and can not make the salamander robot produce the crooked gait of lying down of four limbs.
To adapt to various environments, the movement of salamanders is very flexible, variable and can be divided into rhythmic and non-rhythmic movements. The salamander is formed by combining rhythmic movement and non-rhythmic movement from lying on the ground to standing up and walking. The process of the salamander standing up is generally regarded as non-rhythmic movement, and the major effect of the salamander nerve control system is to resist the action of gravity to make the salamander stand up. At this time, it is very difficult to control the walking motion of the salamander only with a control neural network that can generate only rhythmic motion. In order to solve the problem, the invention divides the control of the simulated spinal cord control network of the salamander robot into a rhythm control network and a non-rhythm control network. According to the characteristic that the salamander moves with different gaits along with the change of the movement speed, the salamander robot rhythm movement control network is divided into a low-frequency walking control network layer and a high-frequency swimming control network layer, and then the networks of the layers are combined to form the movement control network of the whole robot. Under the control of upper neural center signals, the salamander robot utilizes the motion control network, can realize various motion gaits. The specific design process of each layer network is as follows:
(1) rhythmic motion control network
According to the structure of the salamander robot (as shown in figure 7), a control network with two mutually inhibitory neurons is constructed as an oscillator for joint control, wherein the output of one neuron is used as an extensor control signal, and the output of the other neuron is used as a flexor control signal. Considering the characteristics that the swinging speed of the body of the salamander can be obviously increased after the salamander is converted into swimming from land walking, the rhythm control network of the salamander robot is divided into a high-frequency swimming control network and a low-frequency land walking control network to respectively carry out swimming and walking control of the salamander robot. Wherein the low frequency control network is shown in fig. 2, and the high frequency control network is the same as the body control network portion of the low frequency control network.
As can be seen from fig. 8, the low-frequency control network of the salamander robot can be divided into two parts, namely a body control network and a limb control network. The body control network of the salamander robot comprises eleven oscillators, wherein each body joint is controlled by one oscillator. In addition, in order to coordinate the relationship between the body control network and the limb control network, two transitional oscillators are added to the shoulder and the crotch of the body control network. The salamander robot rhythm control network uses descending dominant excitatory connections to establish the connection relationship between body oscillators, as shown in figure 8. Control network of salamander robot four limbs is similar, and every leg has three control oscillator, and the back and forth swinging motion of leg is being controlled to the swing oscillator, lifts up and falls down of leg oscillator control leg, and the knee oscillator is controlling the swinging motion of knee joint. When the salamander walks forward, the four legs of the salamander are in diagonal synchronization, and the swinging motion between the two forelimbs (or the two hind limbs) is in opposite phase. In order to coordinate the movements of the fore and aft limbs by using a body network to realize a limb diagonal synchronous movement gait, the flexor neurons of the body shoulder transition oscillator are connected to the flexor neurons of the right forelimb swing oscillator through inhibitory projections and to the flexor neurons of the left forelimb swing oscillator through excitatory synapses. The connection between extensor neurons and the forelimbs of the shoulder transition oscillator and the connection between flexor neurons and the forelimbs are symmetrical about the longitudinal section of the body of the salamander. The connection between the front limb swing oscillator and the body oscillator functions to move the left and right front limbs in anti-phase with each other. Since the body of the salamander presents a standing wave when walking forward, extensor neurons that swing the forelimb oscillators are connected to ipsilateral neurons of the trunk oscillator in proximity thereto through excitatory synapses. The connection between the back limb swinging oscillator and the body crotch transition oscillator is established by utilizing the standing wave characteristic of the salamander body and considering the diagonal synchronization of the salamander limb movement and the opposite phase movement between the left back limb and the right back limb, the connection relation is similar to the connection relation between the front limb swinging oscillator and the shoulder transition oscillator, and only the synaptic characteristic is correspondingly changed. The extensor neurons of the hind limb oscillatory oscillators are connected to ipsilateral neurons proximal to the caudal oscillators via excitatory synapses, and the flexor neurons are connected to ipsilateral neurons proximal to the trunk oscillators via excitatory synapses. Because the leg needs to be lifted to reduce the influence of ground friction on the forward swing of the leg in the forward swing process of the salamander leg, the leg needs to be pressed down to increase the grip when swinging backwards, and therefore, the output of the flexor neurons of the limb swing oscillator is connected to the flexor neurons of the leg lifting oscillator through excitatory synapses. In order to improve the walking speed, the salamander sometimes can increase the step length of walking forward through adduction of forelimb knee joint in the process of forelimb backward swing. To this end, the output of the extensor neurons of the forelimb swing oscillator are connected to the flexor neurons of the knee oscillator via excitatory synapses. Similar to the control of the brain stem of the salamander on the spinal cord control network, various rhythm movement gaits can be generated by planning the external input signal of the salamander robot control network.
(2) Non-rhythmic motion control network
Considering that the roles of flexor and extensor of joints are mutually inhibited when salamanders move in a non-rhythm state, two neuron networks which are mutually inhibited are utilized as a non-rhythm control network of the joints. Because the realization of movement gaits is not influenced, the connection relation between joint non-rhythm control networks is not established for a while, only a two-neuron mutual inhibition control network is established for each degree of freedom of each joint of the salamander robot, then, the non-rhythm control network of the salamander robot is controlled through a neural control signal from the brainstem, and meanwhile, the rhythm control network is coordinated to realize various movement gaits, and the specific structure is shown in figure 9.
(3) Coordinated control of rhythmic and non-rhythmic networks
The connection relation between the rhythm control network and the non-rhythm control network is not established for the moment under the condition that the walking and swimming gait of the salamander robot are not influenced. In addition, it is assumed that the outputs of the flexor neurons (or extensor neurons) of the joint control networks of the rhythm control network and the non-rhythm control network are added and then output to the flexors (or extensors) of the corresponding joints, as shown in fig. 10. Under the action of brainstem control signals, the rhythmic motion and non-rhythmic motion control networks cooperate to generate the standing walking motion gait of the salamander robot.
For the rhythmic motion control network of the salamander robot, the parameters of the oscillators of all low-frequency control networks are assumed to be the same, and the parameters of the oscillators are selected as follows: tau isi＝0.04、γi＝0.044、εi＝2、σi＝0.1、aii＝2.12、bi＝0.08、θi＝-0.2、
Figure BDA0001290149450000151
And aii+1＝-0.05、ai+1i-0.03(i ═ 1, 3, …, 45). For the connection weight between the oscillators, the excitability weight between the body oscillators is selected to be 0.03, the excitability connection weight between the body oscillators and the limb oscillators is selected to be 0.1, the inhibitory connection weight between the body oscillators and the limb oscillators is selected to be-0.1, and the excitability connection weight between the limb oscillators is selected to be 0.1. Except for tauiAnd gammaiIn addition, the oscillator parameters of the high frequency swim control network are the same as the oscillator parameters of the low frequency control network. For high frequency control networks, τ is seti0.02 and γi0.022. For the non-rhythm control network, the parameters of the two mutually suppressed neuron control networks of all the joints are assumed to be the same, and the parameters of the two mutually suppressed neuron control networks are selected as follows: tau isi＝0.04、γi＝0.044、εi＝2、σi＝0.1、aii＝0.05、bi＝0.6、θi＝-0.2、
Figure BDA0001290149450000152
And aii+1＝-0.05、ai+1i＝-0.05(i＝1，3，…，41)。
According to the design step (2), activation, inactivation and saturation output of the salamander robot rhythm and non-rhythm control network can be realized by changing external input of the control network. For swimmingAnd a walking rhythm control network, when only neuron i is in inactive or saturated output and the other neurons are in rhythm output, the inactive external input s of neuron iiAnd saturated external input siThe lower bounds of (2) are shown in tables 6 and 7. For a non-rhythm control network, when only neuron i is in an inactive or saturated state, while the other neurons are all active and less than the saturated output, the inactive external input s of neuron ii(i-1, …, 42) has an upper bound of-0.4 and the neuron i is saturated with an external input siThe lower bound of (i ═ 1, …, 42) is 181.2.
TABLE 6 external inputs s for inactivation of the neurons i of the walking rhythm control networkiUpper bound and saturated external input siLower bound
Figure BDA0001290149450000161
TABLE 7 inactive external inputs s for swimming rhythm control network neuronsiUpper bound and saturated external input siLower bound
Figure BDA0001290149450000162
By utilizing a dynamic model of the salamander robot and a designed motion control network, walking and swimming gaits of the salamander robot and the performance of mutual switching between the walking and swimming gaits are simulated.
(1) Walk on land
When the salamander was walked on land, except that the support that utilizes the shank resisted gravity and keeps standing, the health also can keep certain rigidity to the efficiency of reinforcing motion. Therefore, it is assumed that the external input signal generated by the brainstem activates the non-rhythmic motion control network of the entire salamander robot. At the moment, the external input of the salamander robot non-rhythm control network generated by the brainstem is si10, (i 1, …, 42). Then, the external input signal generated by the brainstem is utilized to keep the high-frequency swimming control network inactive, the low-frequency walking rhythm control network is activated, and the robot can produceRhythmic upright forward walking motion in which the external input signal s of the low-frequency walking rhythm control networki(i-1, …, 46) are shown in table 8, and the corresponding simulation results are shown in fig. 11.
TABLE 8 external input to Low frequency Walking network
Figure BDA0001290149450000171
As can be seen from fig. 11, the salamander robot presents a standing wave on its body during walking, and the limbs are diagonal and synchronous. Moreover, the leg can upwards be lifted up when the forward wobbling in-process of four limbs in the walking process, and can push down when the backward wobbling, increases the ground gripping power, and the ability that the salamander robot walked forward is strengthened to the lower arm adduction of front limb simultaneously.
(2) Swimming sport
When the salamander walks to the aquatic from land after, the antigravity effect of its limbs weakens or disappears under the effect of buoyancy. For this reason, after the salamander swims, the non-rhythmic control network of the salamander robot limb control network is assumed to be inactive. For the body non-rhythm control network of the salamander robot, assume that its external input remains the same as when walking on land. After switching from walking to swimming motion, through the external input who sets up salamander robot motion control network, make the high-frequency rhythm motion control network of salamander robot activated, high-frequency rhythm motion network external input s this momenti(i ═ 1, …, 22) was 0.2, and the body control network of the low frequency rhythmic walking network was deactivated. The extensor neurons of the limb swing oscillator, the leg raising oscillator and the flexor neurons of the knee oscillator generate saturated outputs, and the other neurons of the corresponding oscillators are in an inactive state. Under the control of these external input signals, the limbs of salamander robot upwards lift up to tightly lean on the salamander robot health backward, with the resistance that reduces to moving forward, simultaneously, the robot health produces a travelling wave, promotes the salamander robot and moves forward. The simulation results of walking and swimming movements of the salamander robot and their mutual switching are shown in figure 12.
As can be seen from fig. 12, the salamander robot first performs overland walking movement, and switches to swimming movement at 10 seconds. When the salamander walks on land, the motion of health and four limbs is similar with the simulation result of above-mentioned walking on land motion, accords with the motion characteristics of salamander walking on land gait. The body presents a traveling wave when swimming, promotes the forward swimming of salamander robot. Furthermore, the frequency of swimming is higher than that of walking on land, and the limbs are lifted upwards and backwards against the body to reduce resistance. The above simulation results are consistent with the results of studies on the movement morphology of salamanders and the neuromuscular control mechanism.

Claims (1)

1. A layered construction method of a bionic robot control network is characterized by comprising the following steps:
(1) according to the speed and the characteristics of the movement gait, the movement gait of the bionic robot is grouped or classified;
(2) constructing a certain layer of motion control network of the bionic robot according to a certain group or class of motion gaits; the constructed motion control network layer is divided into a rhythm motion control network layer and a non-rhythm motion control network layer; the rhythmic motion and non-rhythmic motion control network layer is composed of a motion control network layer which generates certain or some kind of motion gait;
(3) combining all layers of motion control networks together to form the whole bionic robot motion control network; generating a control signal by using an upper nerve center to activate, inactivate and saturate the corresponding part of the whole motion control network so as to generate various motion gaits of the bionic robot; the center is brainstem;
the construction method of the motion control network of each layer is as follows:
the method is based on a novel neuron oscillator;
the novel neuron oscillator is characterized in that a neuron model is established, and then two neurons are connected with each other through inhibitory synapses to form an oscillator model; the neuron model is formed by adding output saturation and self-excitability characteristics on the basis of a leaky integrator neuron model with fatigue characteristics; the output of the neuron model is represented by a nonlinear function, the nonlinear function satisfies that when x is larger than or equal to theta, the output has a saturation characteristic, and when x is smaller than theta, the neuron does not output;
the method for constructing the motion control neural network is characterized in that a novel neuron oscillator is used as the motion control neural network of a robot joint, the output of one neuron is used as a flexor control signal of the robot joint, the output of the other neuron is used as an extensor control signal of the joint, and then the connection relation between the robot joint oscillators is established by utilizing the connection relation of inhibition and excitability according to the motion relation between joints of the robot and the topological structure characteristics of an actual bionic biological nerve control loop;
based on the construction method of the bionic robot motion control network layer, not only a rhythm motion control network layer but also a non-rhythm motion control network layer can be constructed, and the difference lies in that the parameters of the selected motion control network layer are different;
the neuron model adopts one of the following two differential equation sets:
Figure FDA0002241589130000021
Figure FDA0002241589130000022
wherein x is the membrane potential of the neuron; y is the output of the neuron; s is all external inputs received by the neuron; a is the connection weight of self-excitability feedback received by the neuron, and a is more than 0; tau is a time constant related to the membrane potential of the neuron, and tau is more than 0; time constant related to fatigue process of the neuron, gamma is more than 0; x' is a variable reflecting the degree of neuronal fatigue; b is fatigue strength of neuron, b is more than 0; theta is the output threshold of the neuron,
Figure FDA0002241589130000023
for output of neuronsTo the upper bound, and
Figure FDA0002241589130000024
epsilon and sigma are constant coefficients, epsilon is more than 0 and sigma is more than 0; λ is the saturation coefficient of the neuron output;
two neurons are mutually inhibited, each neuron having a self-excitatory link; the concrete model is as follows:
Figure FDA0002241589130000031
Figure FDA0002241589130000032
in the formula, xiIs the membrane potential of the ith neuron; y isiIs the output of the ith neuron; siAn external input received for an ith neuron; a isij(j ∈ {1,2}, j ≠ i) is the connection weight between neurons, aij＜0；aiiConnection weights for self-excitatory feedback received by the ith neuron; tau isiTime constant, τ, associated with membrane potential of ith neuroni＞0；γiTime constant, gamma, associated with the i-th neuron fatigue processi＞0；x′iA variable reflecting the degree of fatigue of the ith neuron; biIs the fatigue strength of the ith neuron, bi＞0；θiIs the output threshold of the ith neuron,
Figure FDA0002241589130000033
is an upper bound of the output of the ith neuron, an
Figure FDA0002241589130000034
εiAnd σiIs a constant coefficient of ∈i> 0 and σi＞0；λiIs the saturation coefficient of the ith neuron output;
the oscillator equilibrium state
Figure FDA0002241589130000035
Satisfy the requirement of
Figure FDA0002241589130000036
The oscillator can generate either an oscillating output or a non-oscillating output, the external input s being at this timeiThe value range of (i ═ 1,2) is as follows:
Figure FDA0002241589130000037
when the oscillator generates oscillation output, the parameters of the oscillator meet the following conditions:
Figure FDA0002241589130000038
Figure FDA0002241589130000041
when the oscillator generates non-oscillation output, the parameters of the oscillator meet the following conditions:
(1)
Figure FDA0002241589130000042
(2)
Figure FDA0002241589130000043
(3)
Figure FDA0002241589130000044
or
(1)
Figure FDA0002241589130000045
(2)
Figure FDA0002241589130000046
(3)σi≥1,(i＝1,2)
According to the conditions of the oscillation output and the non-oscillation output of the oscillator, the oscillation output and the non-oscillation output of the oscillator can be adjusted by adjusting the self-excitation coefficient aiiAnd fatigue coefficient biTo perform a handover;
the oscillator satisfies s when the input is inputi＜εiθiWhen (i ═ 1,2), the output of the oscillator is inactive, when
Figure FDA0002241589130000047
The output of the oscillator is saturated;
wherein the oscillation frequency and response speed of the oscillator are determined by a time constant tau associated with the membrane potential of the neuroniI-1, 2 and a time constant γ associated with the process of neuronal fatigueiI is adjusted to 1, 2;
wherein the saturated output and the non-activated output of the oscillator and the magnitude of the rhythmic output and the non-rhythmic output of the oscillator can be input by the external input s of the oscillatoriI is adjusted to 1, 2;
the model of the bionic robot motion control neural network formed by the mutual connection of the oscillators is as follows:
Figure FDA0002241589130000051
Figure FDA0002241589130000052
wherein n is the number of neurons; x is the number ofiIs the membrane potential of the ith neuron; y isiIs the output of the ith neuron; siAn external input received for an ith neuron; a isij(j ∈ {1, …, n }, j ≠ i) is the connection weight between neurons, aij> 0 denotes an excitatory linkage, aij< 0 indicates inhibitory attachment; a isiiConnection weights for self-excitatory feedback received by the ith neuron; tau isiAssociated with membrane potential of ith neuronTime constant, τi＞0；γiTime constant, gamma, associated with the i-th neuron fatigue processi＞0；xi' is a variable reflecting the degree of fatigue of the ith neuron; biIs the fatigue strength of the ith neuron, bi＞0；θiIs the output threshold of the ith neuron,
Figure FDA0002241589130000053
is an upper bound of the output of the ith neuron, an
Figure FDA0002241589130000054
εiAnd σiIs a constant coefficient of ∈i> 0 and σi＞0；λiIs the saturation coefficient of the ith neuron output;
the bionic robot motion control neural network is used for controlling the neural network when the equilibrium state of a part of neurons exists in the neural network
Figure FDA0002241589130000055
Satisfy the requirement of
Figure FDA0002241589130000056
Namely, it is
Figure FDA0002241589130000057
The part of neurons can generate an oscillatory output and a non-oscillatory output, the external input s corresponding to the equilibrium statei,i∈Λ3The value range is as follows:
Figure FDA0002241589130000061
wherein omegai＝{j|aij＞0,j∈Λ3,j≠i}；
Figure FDA0002241589130000062
This part of the neuron, i.e. i ∈ Λ3Need to be full to generate oscillation outputThe conditions for the feet were as follows:
Figure FDA0002241589130000063
Figure FDA0002241589130000064
this part of the neuron, i.e. i ∈ Λ3The conditions that need to be satisfied for the neuron to produce a non-oscillating output are as follows:
(1)
Figure FDA0002241589130000065
(2)
Figure FDA0002241589130000066
(3)
Figure FDA0002241589130000067
or
(1)
Figure FDA0002241589130000068
(2)
Figure FDA0002241589130000069
(3)σi≥1,(i∈Λ3)
This part of the neurons, i.e. i ∈ Λ, is conditioned on the oscillating and non-oscillating outputs mentioned above3Can be switched between oscillatory and non-oscillatory outputs by varying the self-excitation coefficient aiiAnd fatigue coefficient biIs implemented, and the output of the part of the neuron follows the external input si,i∈Λ3Is changed;
this part of the neuron, i.e. i ∈ Λ3The oscillation frequency and the response speed of the neuron can be controlled byTime constant tau associated with the membrane potential of this part of the neuroni,i∈Λ3Time constant gamma associated with neuronal fatigue processesi,i∈Λ3Carrying out adjustment;
the bionic robot motion control neural network is used for controlling the balance state of a part of neurons
Figure FDA0002241589130000071
Satisfy the requirement of
Figure FDA0002241589130000072
Then, the part of neurons can be collected
Figure FDA0002241589130000073
Indicating and they are inactive, their external input si,i∈Λ1The value range is as follows:
Figure FDA0002241589130000074
the bionic robot motion control neural network is used for controlling the balance state of a part of neurons
Figure FDA0002241589130000075
Satisfy the requirement of
Figure FDA0002241589130000076
Then, the part of neurons can be collected
Figure FDA0002241589130000077
Represent and they are saturated, with its external input si,i∈Λ2The value range is as follows:
Figure FDA0002241589130000078
CN201710323731.3A 2017-05-09 2017-05-09 Layered construction method of bionic robot control network Active CN106914901B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201710323731.3A CN106914901B (en) 2017-05-09 2017-05-09 Layered construction method of bionic robot control network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201710323731.3A CN106914901B (en) 2017-05-09 2017-05-09 Layered construction method of bionic robot control network

Publications (2)

Publication Number Publication Date
CN106914901A CN106914901A (en) 2017-07-04
CN106914901B true CN106914901B (en) 2020-06-05

Family

ID=59568584

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201710323731.3A Active CN106914901B (en) 2017-05-09 2017-05-09 Layered construction method of bionic robot control network

Country Status (1)

Country Link
CN (1) CN106914901B (en)

Families Citing this family (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107991868B (en) * 2017-11-28 2020-08-04 淮海工学院 Coordination control method for left and right pectoral fins of pharyngeal-jaw mode bionic system
CN108582066B (en) * 2018-03-13 2021-09-03 同济大学 Layered CPG and application thereof in walking control of humanoid robot
CN111037572B (en) * 2019-12-31 2022-05-17 江苏海洋大学 Robot stepping priority control method

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101916071A (en) * 2010-08-04 2010-12-15 中国科学院自动化研究所 CPG feedback control method of biomimetic robot fish movement
CN103376742A (en) * 2012-04-24 2013-10-30 中国科学院合肥物质科学研究院 CPG control system of wall-climbing robot imitating flexible structure of feet of gecko
CN105182754A (en) * 2015-08-18 2015-12-23 淮海工学院 Construction method of biomimetic robot motion control neutral network

Family Cites Families (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP5224506B2 (en) * 2008-03-04 2013-07-03 国立大学法人九州工業大学 Control system, vibration control device, and control signal generation method
CN103203746B (en) * 2012-09-29 2015-10-28 同济大学 Biped robot CPG net control topological structure construction method

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101916071A (en) * 2010-08-04 2010-12-15 中国科学院自动化研究所 CPG feedback control method of biomimetic robot fish movement
CN103376742A (en) * 2012-04-24 2013-10-30 中国科学院合肥物质科学研究院 CPG control system of wall-climbing robot imitating flexible structure of feet of gecko
CN105182754A (en) * 2015-08-18 2015-12-23 淮海工学院 Construction method of biomimetic robot motion control neutral network

Also Published As

Publication number Publication date
CN106914901A (en) 2017-07-04

Similar Documents

Publication Publication Date Title
Sharbafi et al. Bioinspired legged locomotion: models, concepts, control and applications
Ferrell A comparison of three insect-inspired locomotion controllers
CN106914901B (en) Layered construction method of bionic robot control network
Fukuoka et al. Dynamic locomotion of a biomorphic quadruped ‘Tekken’robot using various gaits: walk, trot, free-gait and bound
Klein et al. A physical model of sensorimotor interactions during locomotion
CN105182754B (en) The construction method of bio-robot motion control neutral net
Das et al. Invariant structure in locomotion
Anand et al. A deep reinforcement learning based approach towards generating human walking behavior with a neuromuscular model
Wang et al. Lateral undulation of the flexible spine of sprawling posture vertebrates
CN108582066A (en) A kind of layering CPG and the application in Humanoid Robot Based on Walking control
Gordleeva et al. Control of movement of underwater swimmers: Animals, simulated animates and swimming robots
Liu et al. A new model of the spinal locomotor networks of a salamander and its properties
CN112109085B (en) CPG network topology structure control method of humanoid beaver robot
Liu et al. Locomotion control of quadruped robots based on central pattern generators
Wyffels et al. Realization of a passive compliant robot dog
Zhang et al. Gait transition for a quadrupedal robot by replacing the gait matrix of a central pattern generator model
Urbina-Meléndez et al. Brain-body-task co-adaptation can improve autonomous learning and speed of bipedal walking
Ijspeert Vertebrate locomotion
Pei et al. Adaptive control of a quadruped robot based on Central Pattern Generators
Tong et al. Online optimization of normalized CPGs for a multi-joint robotic fish
CN107991868B (en) Coordination control method for left and right pectoral fins of pharyngeal-jaw mode bionic system
Zharinov et al. CPG as a controller for biomimetic floating robots
Ady et al. Analysis of cane-assisted walking through nonlinear optimization
Ijspeert A neuromechanical investigation of salamander locomotion
Zhang et al. Model the leg cycling movement with neural oscillator

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant