CN103203746B - Biped robot CPG net control topological structure construction method - Google Patents

Abstract

The present invention relates to a kind of biped robot CPG net control topological structure construction method, comprise the following steps: CPG net control is divided into the health net control (body for controlling hip joint? network) part and the leg net control (leg for controlling leg joint? network) part, the symmetry of left and right leg joint control signal and the rotational phase relation of left and right leg control in the process of walking to achieve biped robot; The mode of being of coupled connections in CPG net control between neuron elements is optimized, reduces the complexity of CPG net control; Parameter tuning is carried out to CPG net control, constructs optimal network topological structure.Compared with prior art, the present invention has the advantages such as complexity is low, net control is rational in infrastructure.

Description

Biped robot CPG net control topological structure construction method

Technical field

The present invention relates to a kind of information processing method of network topology structure, especially relate to a kind of biped robot CPG net control topological structure construction method.

Background technology

Travelling control be two foot and anthropomorphic robot investigation and application field in a key technology.Traditional method adopts the artificial planning based on model, and robot is moved by the movement locus preset.Along with robot is applied to unstructured moving grids gradually, the conventional method based on robot model and walking environmental modeling hinders the practical application of robot.Along with the further investigation of people to biped walking essence and the development of Neuscience, the control method based on Neuscience is applied in the travelling control of biped robot gradually.Control thinking based on central pattern generator (cpg) (CPG, central pattern generator) is a Typical Representative in this direction.

Biologist thinks, the motion control neutral net of animal, centered by CPG, accepts from high level neural regulation and control order, and from the feedback information of somatic sensor.CPG is the local oscillation network be made up of neuron, can produce stable phase place interlocked relationship, and excite body region of interest to produce rhythmic movement by self-oscillation by mutual suppression the between neuron.The high level regulation and control of brain and environmental feedback can play regulating action to the rhythmic movement of animal, make the motion of animal have adaptability.Control method based on CPG has the interpretation on biology, causes interest widely at engineering circles thus recently, and starts CPG mechanism to carry out Engineering Modeling, is applied in the motion control of all kinds of robot.Try hard to by neural for biological motion control mechanism is combined with the bionic moving mechanism of robot, improve the exercise performance of robot, promote the practicalization of robot in various actual environment.

The basic ideas controlled based on CPG are: first carry out Engineering Modeling to CPG, design the function that can produce stable oscillation stationary vibration output signal, as the controller of the robot free degree, the control of multiple free degree generally realizes with the CPG network that multiple CPG unit is formed, change the output mode that topology of networks can change oscillator signal, thus realize different motor patterns.At present, for the system of this complexity of biped robot, CPG model does not have very strong practicality, and current research also more rests on the analog simulation stage, or is only control the rhythmic movement in some joint of biped robot.The application of CPG mechanism in motion planning and robot control mainly adopts joint space (joint space) control method, and the control framework of its entirety as shown in Figure 1.In the controls, CPG mixed-media network modules mixed-media is the nucleus module producing joint control signal, and the reasonability of its design of network topology structure and Availability are to the quality of control effects.At present when research and engineer applied, generally the free degree of CPG unit according to robot is distributed, be assigned to the joint space of robot one by one, utilize the intercoupling motor pattern producing and expect between multiple CPG unit.This control method is applied to biped robot and also there is no successful real experiment at present, difficult point is just the method for designing that CPG design of network topology structure is also ununified, the complexity of network structure adds the difficulty of parameter tuning, is difficult to produce the joint control signal expected.On the other hand, because the biped robot free degree is more, the CPG number of unit formed required for controller is also relatively many.How determining rational Topology connection and model parameter between CPG unit, is a complicated optimization problem.Generally need to adopt evolution algorithm to be optimized control system, but owing to relating to topological structure optimizing and Model Parameter Optimization simultaneously, adopt traditional optimized algorithm to be difficult to obtain a good result.Meanwhile, this method is optimized consuming time longer, robot entity realizes also have certain difficulty.Therefore each free degree CPG unit simply being distributed to robot that can not be simple, rely on evolutionary computation and carry out optimizing, premised on network topology structure framework that must be reasonable in design, then be aided with connected mode and the model parameter that evolutionary computation optimizes CPG further.

Summary of the invention

Object of the present invention be exactly in order to overcome above-mentioned prior art exist defect and the biped robot CPG net control topological structure construction method that a kind of complexity is low, net control is rational in infrastructure is provided.

Object of the present invention can be achieved through the following technical solutions:

A kind of biped robot CPG net control topological structure construction method, comprises the following steps:

1) CPG net control being divided into for health net control (bodynetwork) part of control hip joint and leg net control (leg network) part for controlling leg joint, being convenient to control the phase relation of the symmetry of biped robot left and right leg control signal and robot left and right leg in the process of walking;

2) coupled modes in CPG network between neuron elements are optimized, simplify the connected mode between neuron;

3) carry out parameter tuning to CPG net control, construct optimal network topological structure.

CPG net control in step 1) adopts neural oscillator model, and the mathematic(al) representation of this model is:

$T_r{\stackrel{\·}{u}}_i^{\{e,f\}}={-u}_i^{\{e,f\}}+w_{\mathrm{fe}}{r}_i^{\{f,e\}}-\mathrm{\β}{v}_i^{\{e,f\}}+s_0+{\mathrm{Feed}}_i^{\{e,f\}}+\underset{j=1}{\overset{n}{\mathrm{\Σ}}}{w}_{\mathrm{ij}}{r}_j^{\{e,f\}}$

$T_a{\stackrel{\·}{v}}_i^{\{e,f\}}=-v_i^{\{e,f\}}+r_i^{\{e,f\}}$

$r_i^{\{e,f\}}=\max (u_i^{\{e,f\}},0)$

$r_i={-u}_i^{\left\{e\right\}}+u_i^{\left\{f\right\}}$

Wherein, i represents i-th CPG unit, and e represents musculus flexor neuron, and f represents extensor neuron, u _ifor neuronic internal state, v _ifor neuron is from holddown, for neuronic output, T _rand T _abe respectively rise time and adaptation time constant, w _fefor neuronic mutual rejection coefficient, β is neuronic from rejection coefficient, s ₀represent the periodic swinging signal that Motion Control Network exports, Feed _ifor feedback input signal, w _ijfor the connection weight between neuron j and neuron i, r _ibe the output of i-th CPG unit, in order to oscillation-damped device output signal is the impact of null part, the design adopts the neuronic status items of musculus flexor neuron, extensor (Kimura model utilizes neuronic output item in the output of linear synthesis oscillator with the output of linear synthesis oscillator).

In step 1) when building CPG network topology structure, health net control (body network) is designed in other joints of the hip joint of biped robot and leg respectively and leg net control (leg network) controls, object is while the phase relation ensureing the symmetry of biped robot left and right leg control signal and robot left and right leg in the process of walking, the complexity of attenuating system, reduces the network parameter needing to optimize.

Step 2) in concrete optimization method be: only consider that mutual rejection coefficient between musculus flexor neuron and between extensor neuron is to obtain the phase relation between joint freedom degrees.

Step 3) specifically comprises the following steps:

A) single-parameter analysis method is adopted to obtain the effect tendency of single model parameter for joint control signal of CPG net control;

B) according to each model parameter, also multi-target evolution computational methods are assisted to the effect tendency of joint control signal, obtain and robot can be made in the optimal joint control signal of level walking;

C) according to the walking effect of reality, to step B) in model parameter under optimal joint control signal finely tune.

Compared with prior art, the present invention is based on the Space Control Method of the biped robot of CPG, propose a kind of new CPG net control construction method, and devise more rationally effective model parameter and to adjust strategy.The design improves traditional net control construction method to a certain extent, and its complexity is reduced, and net control structure is more reasonable.

Accompanying drawing explanation

Fig. 1 is the control system integral frame of Linkspace control method;

Fig. 2 is the joint freedom degrees distribution map of biped robot;

Fig. 3 is the CPG network structure of health net control (body network) part for controlling hip joint;

Fig. 4 is the oscillator signal that health net control (body network) part exports;

Fig. 5 is the CPG net control topological structure proposed by the present invention;

Fig. 6 is the mode that is of coupled connections between musculus flexor neuron and extensor neuron.

Detailed description of the invention

Below in conjunction with the drawings and specific embodiments, the present invention is described in detail.

Embodiment

A kind of biped robot CPG net control topological structure construction method, comprises the following steps:

Step 1: the free degree distribution schematic diagram shown in Fig. 2 being two sufficient anthropomorphic robot legs of current modal investigation and application.For the biped robot travelling control of the type, the control of the hip joint free degree to left and right two leg phase place is the most important.Therefore CPG net control is divided into the health net control for controlling hip joint (body network) part and leg net control (leg network) part for controlling leg joint by the design, realizes the phase relation of the symmetry of biped robot left and right leg control signal and robot left and right leg in the process of walking.For health net control (body network) part, as shown in Figure 3, connect and compose by the bidirectional couple of four neuron elements four frees degree that health net control (body network) controls biped robot hip joint.Connect adopting the mutual suppression of equal weight between four neuron elements of health net control (body network) part (suppressing weight to be set to-1), phase relation as shown in Figure 4 can be obtained, the HipPitch phase difference of left and right leg is 180 °, meet the anti-phase relation of left and right hip joint fore-and-aft direction free degree control signal, the desirable phase difference relation of 90 ° of two frees degree in same joint can be obtained simultaneously.

Herein, CPG net control adopts neural oscillator model, and the mathematic(al) representation of this model is:

$T_r{\stackrel{\·}{u}}_i^{\{e,f\}}={-u}_i^{\{e,f\}}+w_{\mathrm{fe}}{r}_i^{\{f,e\}}-\mathrm{\β}{v}_i^{\{e,f\}}+s_0+{\mathrm{Feed}}_i^{\{e,f\}}+\underset{j=1}{\overset{n}{\mathrm{\Σ}}}{w}_{\mathrm{ij}}{r}_j^{\{e,f\}}$

$T_a{\stackrel{\·}{v}}_i^{\{e,f\}}=-v_i^{\{e,f\}}+r_i^{\{e,f\}}$

$r_i^{\{e,f\}}=\max (u_i^{\{e,f\}},0)$

$r_i={-u}_i^{\left\{e\right\}}+u_i^{\left\{f\right\}}$

Wherein, i represents i-th CPG unit, and e represents musculus flexor neuron, and f represents extensor neuron, u _ifor neuronic internal state, v _ifor neuron is from holddown, for neuronic output, T _rand T _abe respectively rise time and adaptation time constant, w _fefor neuronic mutual rejection coefficient, β is neuronic from rejection coefficient, s ₀represent the periodic swinging signal that Motion Control Network exports, Feed _ifor feedback input signal, w _ijfor the connection weight between neuron j and neuron i, r _ibe the output of i-th CPG unit, in order to oscillation-damped device output signal is the impact of null part, the design adopts the neuronic status items of musculus flexor neuron, extensor (Kimura model utilizes neuronic output item in the output of linear synthesis oscillator with the output of linear synthesis oscillator).

Step 2: be optimized the mode of being of coupled connections in CPG net control between neuron, reduces the complexity of CPG network.Connected mode between traditional neuron is musculus flexor neuron and the coupling of extensor neuron omnidirectional, for two neuron elements, needs the modulation of consideration four connection weights.For the biped robot of multiple free degree, if connected mode traditionally, need the Connecting quantity substantial amounts optimized, be difficult to obtain suitable optimum results.The method that is of coupled connections as shown in Figure 6 is adopted: consider that the mutual rejection coefficient between musculus flexor neuron and between extensor neuron just can obtain the phase relation of expectation, therefore in order to reduce the complexity of whole topological structure in the present invention.

Step 3: parameter tuning is carried out to CPG net control based on evolutionary computation, optimize model parameter and the Topology connection mode of CPG net control, specifically can be subdivided into the following steps:

First, single-parameter analysis method is adopted to obtain the single model parameter (T of CPG net control _r, T _a, β, s ₀and w _fe) for the effect tendency of joint control signal.Owing to being intercouple between parameter, there is not linear relationship to the impact exported in parameter, one of them model parameter selected, as measured parameter, only changes the value of this measured parameter, by judging that CPG outputs signal whether stable oscillation stationary vibration, estimate the span of measured parameter; In the scope estimated, regulate measured parameter continuously, measure export can the quantitative relationship of quantization signifying amount (amplitude, frequency, phase place etc.) and measured parameter;

Then, the joint control signal of Neng Shi robot level walking is obtained based on evolutionary computation method.Based on the Linkspace control method of CPG due to kinematics model and the computation of inverse-kinematics not based on biped robot, so the stability of robot ambulation by optimizing the parameter of CPG network, can only obtain the joint control signal that can realize each free degree coordination of robot.In order to improve the walking stability of robot, the design considers that in parameter optimization target the stability margin (using ZMP to the stability margin of beeline as pedestrian system supporting convex polygon border) when utilizing robot ambulation improves the stability of robot ambulation.The design utilizes multi-objective Evolutionary Algorithm, the NSGA-II(Non-Dominated Sorting Genetic Algorithm-II based on Kalyanmoy Deb) carry out the design of multi-objective genetic algorithm.The target optimized obtains realizing the optimum Topology connection mode of robot level walking and the model parameter of CPG, and the design of fitness function is considered to be designed to feasibility and the stability of robot ambulation:

${\mathrm{fitness}}_1=1/\sqrt{{(x_{\mathrm{end}}-x_0)}^2}$

fitness ₂＝1/D _s

Wherein, x ₀the initial position (x direction) of robot, x _endrepresent the position (x direction) during robot stopping walking.First aim function is the straight line moving in order to realize robot, therefore only considers robot advance (x direction) direction travel distance when optimizing.Second optimization aim considers the stability of walking, D _sfor the stability margin of robot ambulation.

Finally, according to the walking effect of biped robot reality, to step B) in model parameter under optimal joint control signal finely tune, it is optimized.Finally, according to the robot ambulation experiment effect of reality, at completing steps A) and step B) basis on carry out structure adjusting parameter in conjunction with the experience of designer.This parameter tuning method fully combines the intelligence of computational intelligence and the mankind, allows designer according to the comprehensive Output rusults of optimized algorithm, progressively can deepen the understanding to controller and controlled device, and to obtain new design philosophy and inspiration.

The present invention is based on the Space Control Method of the biped robot of CPG, propose a kind of new CPG net control construction method, and devise more rationally effective model parameter and to adjust strategy, improve traditional net control construction method to a certain extent, its complexity is reduced, and net control structure is more reasonable.

Claims (3)

1. a biped robot CPG net control topological structure construction method, is characterized in that, comprise the following steps:

1) CPG net control be divided into the part of the body network for controlling hip joint and be used for controlling the leg network part in leg joint, ensureing the phase relation of the symmetry of biped robot left and right leg control signal and robot left and right leg in the process of walking;

2) mode of being of coupled connections between CPG network introflexion muscular nerve unit and extensor neuron is optimized, reduces the complexity of CPG net control;

3) parameter tuning is carried out to CPG net control, constructs optimal network topological structure, specifically comprise the following steps:

A) single-parameter analysis method is adopted to obtain the effect tendency of single model parameter for joint control signal of CPG net control;

B) according to each model parameter, also multi-target evolution computational methods are assisted to the effect tendency of joint control signal, obtain and robot can be made in the optimal joint control signal of level walking;

C) according to the walking effect of reality, to step B) in model parameter under optimal joint control signal finely tune;

Step 2) in concrete optimization method be: only consider that mutual rejection coefficient between musculus flexor neuron and between extensor neuron is to obtain the phase relation between joint freedom degrees.

2. a kind of biped robot CPG net control topological structure construction method according to claim 1, is characterized in that, step 1) in CPG network adopt neural oscillator model, the mathematic(al) representation of this model is:

$T_r{\stackrel{\·}{u}}_i^{\left\{e,f\right\}}=-u_i^{\left\{e,f\right\}}+w_{fe}{r}_i^{\left\{f,e\right\}}-{\mathrm{\βv}}_i^{\left\{e,f\right\}}+s_0+{\mathrm{Feed}}_i^{\left\{e,f\right\}}+\underset{j=1}{\overset{n}{\Σ}}{w}_{ij}{r}_j^{\left\{e,f\right\}}$

$T_a{\stackrel{\·}{v}}_i^{\left\{e,f\right\}}=-v_i^{\left\{e,f\right\}}+r_i^{\left\{e,f\right\}}$

$r_i^{\left\{e,f\right\}}=max(u_i^{\left\{e,f\right\}},0)$

$r_i=-u_i^{\left\{e\right\}}+u_i^{\left\{f\right\}}$

Wherein, i represents i-th CPG unit, and e represents musculus flexor neuron, and f represents extensor neuron, u _ifor neuronic internal state, v _ifor neuron is from holddown, r _i ^{{ e, f}}for neuronic output, T _rand T _abe respectively rise time and adaptation time constant, w _fefor neuronic mutual rejection coefficient, β is neuronic from rejection coefficient, s ₀represent the periodic swinging signal that Motion Control Network exports, Feed _ifor feedback input signal, w _ijfor the connection weight between neuron j and neuron i, r _ibe the output of i-th CPG unit, in order to oscillation-damped device output signal is the impact of null part, adopt musculus flexor neuron, the neuronic status items of extensor the output of linear synthesis oscillator.

3. a kind of biped robot CPG net control topological structure construction method according to claim 1, it is characterized in that: the hip joint of biped robot and leg joint are designed respectively body network part and legnetwork partly controls, while the phase relation ensureing the symmetry of biped robot left and right leg control signal and robot left and right leg in the process of walking, the complexity of attenuating system, reduces the network parameter needing to optimize.