
In this paper, our main objective was to investigate the central pattern generator (CPG) neural network model for quadruped gait with time delay. First, we computed the normal form of the model on the center manifold, the bifurcation direction, and stability conditions of the bifurcating periodic solutions. Second, we applied the CPG model for quadruped gait to obtain reference models for goat's diagonal trotting gait on the flat ground and walking gait on the 18 degree slope through the trust region inversion algorithm. Finally, we performed numerical simulations to support theoretical analysis.
Citation: Liqin Liu, Chunrui Zhang. A neural network model for goat gait[J]. Mathematical Biosciences and Engineering, 2024, 21(8): 6898-6914. doi: 10.3934/mbe.2024302
[1] | Liqin Liu, Chunrui Zhang . Dynamic properties of VDP-CPG model in rhythmic movement with delay. Mathematical Biosciences and Engineering, 2020, 17(4): 3190-3202. doi: 10.3934/mbe.2020181 |
[2] | Xiaoguang Liu, Yubo Wu, Meng Chen, Tie Liang, Fei Han, Xiuling Liu . A double-channel multiscale depthwise separable convolutional neural network for abnormal gait recognition. Mathematical Biosciences and Engineering, 2023, 20(5): 8049-8067. doi: 10.3934/mbe.2023349 |
[3] | Xiaoguang Liu, Meng Chen, Tie Liang, Cunguang Lou, Hongrui Wang, Xiuling Liu . A lightweight double-channel depthwise separable convolutional neural network for multimodal fusion gait recognition. Mathematical Biosciences and Engineering, 2022, 19(2): 1195-1212. doi: 10.3934/mbe.2022055 |
[4] | Zhangjie Wu, Minming Gu . A novel attention-guided ECA-CNN architecture for sEMG-based gait classification. Mathematical Biosciences and Engineering, 2023, 20(4): 7140-7153. doi: 10.3934/mbe.2023308 |
[5] | Wajid Aziz, Lal Hussain, Ishtiaq Rasool Khan, Jalal S. Alowibdi, Monagi H. Alkinani . Machine learning based classification of normal, slow and fast walking by extracting multimodal features from stride interval time series. Mathematical Biosciences and Engineering, 2021, 18(1): 495-517. doi: 10.3934/mbe.2021027 |
[6] | Ting Yao, Farong Gao, Qizhong Zhang, Yuliang Ma . Multi-feature gait recognition with DNN based on sEMG signals. Mathematical Biosciences and Engineering, 2021, 18(4): 3521-3542. doi: 10.3934/mbe.2021177 |
[7] | Yubo Sun, Yuanyuan Cheng, Yugen You, Yue Wang, Zhizhong Zhu, Yang Yu, Jianda Han, Jialing Wu, Ningbo Yu . A novel plantar pressure analysis method to signify gait dynamics in Parkinson's disease. Mathematical Biosciences and Engineering, 2023, 20(8): 13474-13490. doi: 10.3934/mbe.2023601 |
[8] | Leandro Donisi, Giuseppe Cesarelli, Pietro Balbi, Vincenzo Provitera, Bernardo Lanzillo, Armando Coccia, Giovanni D'Addio . Positive impact of short-term gait rehabilitation in Parkinson patients: a combined approach based on statistics and machine learning. Mathematical Biosciences and Engineering, 2021, 18(5): 6995-7009. doi: 10.3934/mbe.2021348 |
[9] | Xuecheng Weng, Chang Mei, Farong Gao, Xudong Wu, Qizhong Zhang, Guangyu Liu . A gait stability evaluation method based on wearable acceleration sensors. Mathematical Biosciences and Engineering, 2023, 20(11): 20002-20024. doi: 10.3934/mbe.2023886 |
[10] | Zhen Liu, Pande Zhang, Ming Yan, Yimin Xie, Guangzhi Huang . Additive manufacturing of specific ankle-foot orthoses for persons after stroke: A preliminary study based on gait analysis data. Mathematical Biosciences and Engineering, 2019, 16(6): 8134-8143. doi: 10.3934/mbe.2019410 |
In this paper, our main objective was to investigate the central pattern generator (CPG) neural network model for quadruped gait with time delay. First, we computed the normal form of the model on the center manifold, the bifurcation direction, and stability conditions of the bifurcating periodic solutions. Second, we applied the CPG model for quadruped gait to obtain reference models for goat's diagonal trotting gait on the flat ground and walking gait on the 18 degree slope through the trust region inversion algorithm. Finally, we performed numerical simulations to support theoretical analysis.
Animal gaits are a rhythmic behavior controlled by central pattern generator (CPG), which generates rhythmic signals to the muscle groups of animals, controls their own rhythmic movements, and produces various gaits in animal locomotion. From a biological perspective, CPG is a distributed vibration network composed of neurons in the central nervous system of animals. Human beings get inspiration from the organizational structure, movement mechanism, and behavior of biological bodies, and constantly learn and imitate the characteristics and functions of certain organisms, thereby improving their adaptability to nature and making contributions to the development of science and technology. Therefore, biologists and mathematicians have established some CPG biological and mathematical models by imitating the neural network patterns of animal CPG. Golubitsky et al. [1] has investigated a composition pattern of CPG neurons in quadrupedal gaits and obtained a new primary gait ''jump''. The animal gait network proposed by Buono et al. [2] can generate six primary gaits: walk, trot, pace, bound, pronk, and jump. Furthermore, based on the primary gait, the phase patterns of transverse gallop and rotary gallop in quadruped secondary gait were provided [3]. The present authors have studied CPG neural network models for the primary gait of quadrupeds [4,5].
Animal gait CPG neural networks have wide applications. Inspired by animal movements, in the gait planning of biomimetic robots, a CPG model is set for the robot system by imitating the CPG network of animals in nature. By utilizing the coupling between oscillators in the CPG model, periodic oscillation signals are generated to achieve stable rhythmic gait behavior of the robot. For instance, the CPG network can be used to construct gait networks for hexapod robots gait networks [6,7] and quadruped robots gait networks [8,9,10]. In article [11], a method for designing a robot system controller was proposed using a CPG control network, and in article [12], the robot gait was produced based on animal gait. In [13], a CPG animation model was constructed using the CPG network system.
In quadruped mammals, goats have smaller bodies and stronger bones in their limbs, and their hindlimbs are stronger than their forelimbs specifically. Goats have a brisk and agile gait when walking, and they can walk freely, flexibly, and quickly on the ground, slopes, steep walls and uneven surfaces. Moreover, they can also walk long distances and have good road adaptability and obstacle crossing ability. Goats are less restricted by the environment and, therefore, goat gaits have certain research value. In literature [14], the author used goats as the research object and conducted kinematic analysis on the multimode gait of goats on conventional and unconventional roads. In work [15], the author analyzed the gait of goats from the perspective of structural bionics. At present, there is little research on mathematical models of goat gaits to our knowledge. In this paper, we deal with this problem and propose the CPG neural network models of goat gaits from a mathematical perspective. We study the following model given in [5]
{˙x1(t)=ax1(t)+btanh(x1(t))+dtanh(x7(t))+ctanh(x2(t−τ)),˙x2(t)=ax2(t)+btanh(x2(t))+dtanh(x8(t))+ctanh(x1(t−τ)),˙x3(t)=ax3(t)+btanh(x3(t))+dtanh(x1(t))+ctanh(x4(t−τ)),˙x4(t)=ax4(t)+btanh(x4(t))+dtanh(x2(t))+ctanh(x3(t−τ)),˙x5(t)=ax5(t)+btanh(x5(t))+dtanh(x3(t))+ctanh(x6(t−τ)),˙x6(t)=ax6(t)+btanh(x6(t))+dtanh(x4(t))+ctanh(x5(t−τ)),˙x7(t)=ax7(t)+btanh(x7(t))+dtanh(x5(t))+ctanh(x8(t−τ)),˙x8(t)=ax8(t)+btanh(x8(t))+dtanh(x6(t))+ctanh(x7(t−τ)), | (1.1) |
where a,b,c,d are constants and τ≥0 is the time delay. The state variable (x1(t),x2(t),⋯,x8(t)) is the output signal from the CPG to the legs, where x1(t) and x7(t) are the output signals to the left hind leg, x3(t) and x5(t) are the output signals to the left fore leg, x2(t) and x8(t) are the output signals to the right hind leg, and x4(t) and x6(t) are the output signals to the right fore leg. The network diagram of system (1.1) is shown in Figure 1.
In reference [5], the authors give the conditions (H1):|c|>|a+b+d|if(a+b)d>0, (H2):|c|>|a+b−d|if(a+b)d<0, and critical values τkj(j=1,2,3,4,5,6,7,8) (see Table 1 in [5]) for the Hopf bifurcations of the model (1.1) at the zero equilibrium.
The initial value | Iterations | error f | Numerical solution |
6.1 | 11 | 11.6591 | 6.4300 |
6.3 | 9 | 11.6591 | 6.4300 |
6.5 | 5 | 11.6591 | 6.4300 |
7.2 | 9 | 11.6591 | 6.4300 |
The main objective of this paper is to provide further the bifurcation direction and stability conditions of the bifurcating periodic solutions of the model (1.1). Based on the model (1.1), the CPG neural network models of goat gaits are given by using the parameter inversion algorithm.
In this section, we use the method in reference [16] to calculate the normal forms of Hopf bifurcations on the center manifold of the Eq (1.1) at the zero equilibrium. Normalizing the delay τ by the time-scaling t↦tτ, Eq (1.1) can be rewritten as a functional differential equation in C=C([−1,0],R8),
{˙u1(t)=τ(au1(t)+btanh(u1(t))+dtanh(u7(t))+ctanh(u2(t−1))),˙u2(t)=τ(au2(t)+btanh(u2(t))+dtanh(u8(t))+ctanh(u1(t−1))),˙u3(t)=τ(au3(t)+btanh(u3(t))+dtanh(u1(t))+ctanh(u4(t−1))),˙u4(t)=τ(au4(t)+btanh(u4(t))+dtanh(u2(t))+ctanh(u3(t−1))),˙u5(t)=τ(au5(t)+btanh(u5(t))+dtanh(u3(t))+ctanh(u6(t−1))),˙u6(t)=τ(au6(t)+btanh(u6(t))+dtanh(u4(t))+ctanh(u5(t−1))),˙u7(t)=τ(au7(t)+btanh(u7(t))+dtanh(u5(t))+ctanh(u8(t−1))),˙u8(t)=τ(au8(t)+btanh(u8(t))+dtanh(u6(t))+ctanh(u7(t−1))). | (2.1) |
Denoting U=(u1,u2,u3,u4,u5,u6,u7,u8)T, Ut=U(t+θ),Ut∈C, where θ∈[−1,0]. Let τ=τ02j+μ(j=1,2,3,4), for μ is a bifurcation parameter, then Eq (2.1) can further be written as the following form:
˙U(t)=L(0)Ut+F(Ut,μ), | (2.2) |
where
F(Ut,μ)=(L(μ)−L(0))Ut+F0(Ut,μ), |
L(μ)φ=(τ02j+μ)((a+b)(φ1(0))+dφ7(0)+cφ2(−1)(a+b)(φ2(0))+dφ8(0)+cφ1(−1)(a+b)(φ3(0))+dφ1(0)+cφ4(−1)(a+b)(φ4(0))+dφ2(0)+cφ3(−1)(a+b)(φ5(0))+dφ3(0)+cφ6(−1)(a+b)(φ6(0))+dφ4(0)+cφ5(−1)(a+b)(φ7(0))+dφ5(0)+cφ8(−1)(a+b)(φ8(0))+dφ6(0)+cφ7(−1)), |
and
F0(Ut,μ)=(τ02j+μ)(−b3φ31(0)−d3φ37(0)−c3φ32(−1)−b3φ32(0)−d3φ38(0)−c3φ31(−1)−b3φ33(0)−d3φ31(0)−c3φ34(−1)−b3φ34(0)−d3φ32(0)−c3φ33(−1)−b3φ35(0)−d3φ33(0)−c3φ36(−1)−b3φ36(0)−d3φ34(0)−c3φ35(−1)−b3φ37(0)−d3φ35(0)−c3φ38(−1)−b3φ38(0)−d3φ36(0)−c3φ37(−1))+h.o.t., |
where φ=(φ1,φ2,φ3,φ4,φ5,φ6,φ7,φ8)T∈C. By the Riesz representation theorem, there is η(θ,μ), such that L(μ)φ=∫0−1dη(θ,μ)φ(θ), and we select η(θ,μ)=(τ02j+μ)(Aδ(θ)+Bδ(θ+1)), where
A=(B000D0D0B0000D0B0000D0B0),B=(C00000C00000C00000C0), |
B0=(a+b00a+b),D0=(d00d),C0=(0cc0). |
The Taylor expansion of F(Ut,μ) is denoted as
F(Ut,μ)=12!F2(φ,μ)+13!F3(φ,μ)+h.o.t. |
with
F2(φ,μ)=μ((a+b)(φ1(0))+dφ7(0)+cφ2(−1)(a+b)(φ2(0))+dφ8(0)+cφ1(−1)(a+b)(φ3(0))+dφ1(0)+cφ4(−1)(a+b)(φ4(0))+dφ2(0)+cφ3(−1)(a+b)(φ5(0))+dφ3(0)+cφ6(−1)(a+b)(φ6(0))+dφ4(0)+cφ5(−1)(a+b)(φ7(0))+dφ5(0)+cφ8(−1)(a+b)(φ8(0))+dφ6(0)+cφ7(−1)), |
F3(φ,μ)=τ02j(−b3φ31(0)−d3φ37(0)−c3φ32(−1)−b3φ32(0)−d3φ38(0)−c3φ31(−1)−b3φ33(0)−d3φ31(0)−c3φ34(−1)−b3φ34(0)−d3φ32(0)−c3φ33(−1)−b3φ35(0)−d3φ33(0)−c3φ36(−1)−b3φ36(0)−d3φ34(0)−c3φ35(−1)−b3φ37(0)−d3φ35(0)−c3φ38(−1)−b3φ38(0)−d3φ36(0)−c3φ37(−1)). |
Expanding space C to BC={φ:[−1,0]→C8|φ is continuous on [-1, 0) and limθ→0−φ(θ)∈C8}. The element of BC can be expressed as υ=φ+X0ν,φ∈C,ν∈C8, and
X0(θ)={0,−1≤θ<0I,θ=0 |
with I as the identity matrix.
For ϕ∈C1=C1([−1,0],C8), we define
A0ϕ={˙ϕ,−1≤θ<0∫0−1dη(θ,0)ϕ(θ),θ=0. | (2.3) |
For ψ∈C1∗=C1([−1,0],C8∗), the adjoint operator of A0 is
A∗0ψ={−˙ψ,−1≤s<0,∫0−1ψ(−s)dη(s,0),s=0, | (2.4) |
and a bilinear inner product is the following:
<ψ,ϕ>=ˉψ(0)ϕ(0)−∫0−1∫θ0ˉψ(ξ−θ)dη(θ,0)ϕ(ξ)dξ. | (2.5) |
Let Φ(θ) and Ψ(s) be the eigenvectors corresponding to the eigenvalues iωjτ02j and −iωjτ02j, respectively. Note that Φ(θ)=(ϕ(θ),ˉϕ(θ)), Ψ(s)=(ˉψ(s),ψ(s))T with ˙Φ=Φ(θ)J,˙Ψ(s)=−JΨ(s),<Ψ(s),Φ(θ)>=I, and J=diag(iωjτ02j,−iωjτ02j). It can be easily calculated from (2.3) and (2.4) that
ϕ(θ)=(1,−1,1,−1,1,−1,1,−1)Teiωjτ02jθ, |
ψ(s)=D(−1,1,−1,1,−1,1,−1,1)eiωjτ02js. |
Hence, we obtain D=18(−1+ceiωjτ02j) from <ϕ(s),ϕ(θ)>=1.
Let P be a vector space expanded by ϕ(θ) and ˉϕ(θ), P∗ be a vector space expanded by ϕ(s) and ˉϕ(s), then C can be decomposed as C=P⊕Q, where Q={ϕ∈C:<ψ,ϕ>=0,∀ψ∈P∗}. Define mapping Π:BC→P as Π(φ+X0ν)=Φ[(ψ,ϕ)+ψ(0)ν] and Q1=kerπ⋂C1.
Using the decomposition of U=Φx+y with x=(x1,x2)T,y=(y1,y2,y3,y4,y5,y6,y7,y8)T, system (2.1) can be decomposed as
{˙x=Jx+Ψ(0)F(Φx+y,μ),˙y=AQ1y+(I−Π)X0F(Φx+y,μ). | (2.6) |
We have the Taylor expansion as follows:
{˙x=Jx+12!f12(x,y,μ)+13!f13(x,y,μ)+h.o.t.,˙y=AQ1y+12!f22(x,y,μ)+13!f23(x,y,μ)+h.o.t., | (2.7) |
where
f12(x,y,μ)=Ψ(0)F2(Φx+y,μ),f13(x,y,μ)=Ψ(0)F3(Φx+y,μ), |
f22(x,y,μ)=(I−Π)X0F2(Φx+y,μ),f23(x,y,μ)=(I−Π)X0F3(Φx+y,μ). |
Normal form of Eq (2.7) on the center manifold at the origin is given by
˙x=Jx+12!g12(x,0,μ)+13!g13(x,0,μ)+h.o.t., | (2.8) |
where
g12(x,0,μ)=Projker(M12)f12(x,0,μ), |
g13(x,0,μ)=Projker(M13)˜f13(x,0,μ), |
ker(M12)=span{(μx10),(0μx2)}, |
ker(M13)=span{(μ2x10),(x21x20),(0μ2x2),(0x1x22)}, |
˜f13(x,0,μ)=f13(x,0,μ)+32[(Dxf12)U12(x,μ)+[(Dyf12)h]U22(x,μ), |
U12(x,μ)μ=0=(M12)−1ProjIm(M12)f12(x,0,0),M22U22(x,μ)=f22(x,0,μ). |
After calculating, we obtain
12!g12(x,0,μ)=(ˉa1μx1a1μx2), |
where a1=8D(−a−b−d+ceiωjτ02j).
In the following, we can compute the cubic terms 13!g13(x,0,μ) as
13!g13(x,0,μ)=13!Projker(M13)˜f13(x,0,μ)=13!ProjS˜f13(x,0,0)+O(μ2|x|)=13!ProjSf13(x,0,μ)+14ProjS[(Dxf12)(x,0,0)U12(x,0)+(Dyf12)(x,0,0)U22(x,0)]+O(μ2|x|), |
with
S=span{(x21x20),(0x1x22)}. |
Since f12(x,0,0)=(0,0)T,f12(x,y,0)=(0,0)T, we obtain U12(x,0)=(0,0)T,(Dyf12)(x,0,0)=(0,0)T,
thus
13!g13(x,0,μ)=13!ProjSf13(x,0,0)+O(μ2|x|)=(ˉa2x21x2a2x1x22)+O(μ2|x|), |
where a2=−8τ02jD(−b+ceiωiτ02j−d).
Then, the normal form (2.8) can be written as
{˙x1=iωiτ02jx1+ˉa1μx1+ˉa2x21x2+h.o.t.,˙x2=−iωiτ02jx2+a1μx2+a2x1x22+h.o.t.. | (2.9) |
By transforming the variables x1=w1+iw2, x2=w1−iw2 and w1=rcosξ, w2=rsinξ, Eq (2.9) can be written as
{˙r=k1μr+k2r3+h.o.t,˙ξ=−ωjτ02j+h.o.t, | (2.10) |
where k1=Re(a1),k2=Re(a2).
According to [16], we obtain the following results.
Theorem 2.1 When k2≠0, then
1) If k2<0, then the bifurcation periodic solutions of system (1.1) near τ02j(j=1,2,3,4) are stable; if k2>0, then the bifurcation periodic solutions of system (1.1) are unstable.
2) If k1k2<0, then Hopf bifurcations are supercritical; if k1k2>0, then Hopf bifurcations are subcritical.
In this section, we will provide two numerical examples to validate our theoretical analysis.
Example 1. We consider system (1.1) with a=−4.5,b=1,c=6.5,d=−2, which satisfies the conditions (H1). Using the algorithm in [5], we obtain ω2=7.4772,ω3=6.325,τ04=0.2861,τ06=0.2852. From ω2=7.4772,τ04=0.2861, we get k1=Re(a1)=0.4179>0, k2=Re(a2)=−0.2349<0,k1k2=−0.0982<0. According to Theorem 2.1, the bifurcation periodic solution of system (1.1) is stable and supercritical at τ=0.4>τ04=0.2861, as shown in Figure 2. This periodic solution corresponds to the walking gait of quadrupeds.
From ω3=6.325,τ06=0.2852, we know that k1=Re(a1)=0.6486,k2=Re(a2)=−0.2544<0,k1k2=−0.1650<0. According to Theorem 2.1, the bifurcation periodic solution of system (1.1) is stable and supercritical at τ=0.4>τ06=0.2852 (see Figure 3), and this periodic solution corresponds to the trotting gait of quadrupeds.
Example 2. In system (1.1), we select parameters a=−3,b=0.1,c=4,d=1, which satisfies (H2). By calculation, it is obtained that w1=3.5119,τ02=0.5869,k1=Re(a1)=0.5957>0, k2=Re(a2)=−0.5951<0,k1k2=−0.3545<0. From Theorem 2.1, the bifurcation periodic solution of system (1.1) is stable and supercritical at τ=0.7>τ02=0.5869, as shown in Figure 4. This periodic solution corresponds to the pacing gait of quadrupeds.
In this section. we employ model (1.1) as the inversion model, and use the trust region algorithm [17] to give the goat's diagonal trotting model on flat ground and the walking gait model on the ground with a slope of 18 degrees.
By analyzing the spatiotemporal characteristic diagram of the goat's diagonal trot gait (Figure 3.2(b) in [14]), we obtain that the two legs on the diagonal of the goat are mostly in a supported or airborne state when trotting diagonally on flat ground, and the four legs airborne and single legs support the states only account for a small part of the whole gait cycle and can be ignored. So, we assume that the support state and airborne state of the two legs on the diagonal each account for half of the whole cycle. In the diagonal trotting joint angle change curve of goats (Figures 3 and 4 in [14]), an average of 20 points are extracted from the pastern joint angles and wrist joint angles curves of the two front legs, as well as the toe joint angle and tarsal joint angle curves of the two hind legs, respectively (within one cycle). Next, we translate these points to the vicinity of the origin and convert them into radians as the true values ˆxi(tl)(i=1,2,⋯,8;l=1,2,⋯,20) of CPG oscillators of the corresponding legs of goats. Finally, we averagely select 20 values of periodic solutions x1(t) to x8(t) within one cycle of model (1.1)(a=−2,b=1,d=−2,τ=0.25) as the theoretical values xi(tl,c)(i=1,2,⋯,8;l=1,2,⋯,20) of CPG oscillators, respectively. Using the trust region algorithm [17], the parameter c=6.4300 is obtained, as shown in Table 1. The error in Table 1 is f=128∑i=120∑l=1(ˆxi(tl)−xi(tl,c))2.
Next, in order to better describe the effect of the simulation, we calculated the synchronization error between the theoretical value and the real value. That is, we calculated the point-by-point difference between the 20 pairs of theoretical values and true values of the corresponding joint angles of each leg, as shown in Figures 5 and 6.
As can be seen from Figures 5 and 6, the synchronization error ranges of the pastern joint angle and wrist joint angle of the front leg are [-0.5, 0.6] and [-0.85, 0.08], and the synchronization error ranges of the toe joint angle and the tarsal joint angle of the back leg are [-0.4, 0.95] and [-0.6, 0.95].
For parameters a=−2,b=1,c=6.4300,d=−2,τ=0.25, the critical value τ06=0.2227 of trotting gait and k1=Re(a1)=1>0,k2=Re(a2)=−0.2227<0,k1k2=−0.2227<0 are calculated. According to Theorem 2.1, the bifurcation periodic solution of system (1.1) is stable and supercritical at τ=0.25>τ06=0.2227 (see Figure 7).
Therefore, the model (1.1) (a=−2,b=1,c=6.4300,d=−2,τ=0.25) provides a reference model for the goat's diagonal trot gait on flat ground.
According to the analysis of the spatiotemporal state diagram of each leg of a goat walking on an 18 degree slope (Figures 3–8 in [15]), it is found that during one gait cycle, each leg of the goat takes turns to be suspended while the other three legs are in a supported state, and they took almost the same amount of time. Assuming that within a gait cycle, each leg of the goat is in a suspended state while the other three legs are in a supported state for the same amount of time.
Similar to Section 4.1, we take 20 data from the joint angle curve of each leg in the Figures 3–9 [15] (the curve of the wrist joint angle α of each leg and the curve of the angle β between the thighs of each leg and the forward direction) as the true values ˆxi(tl)(i=1,2,⋯,8;l=1,2,⋯,20) of CPG oscillators, then we select 20 values (within one cycle) from the periodic solutions x1(t) to x8(t) of the model (1.1)(a=−2,b=1,c=6.5,d=−2) as the theoretical values xi(tl,τ)(i=1,2,⋯,8;l=1,2,⋯,20) of CPG oscillators, and the parameter τ=0.2450 is obtained by the trust region algorithm, as shown in Table 2. The precision ε in Table 2 is the degree of accuracy that the gradient of the objective function f=128∑i=120∑l=1(ˆxi(tl)−xi(tl,τ))2 needs to achieve when the objective function f reaches its optimal solution in the trust region algorithm. Similar to Section 4.1, synchronization error diagrams between theoretical values and true values are shown in Figures 8 and 9.
The initial value of τ | Iterations | precision ε | Numerical solution |
0.25 | 5 | 1.85 | 0.2450 |
0.28 | 8 | 1.85 | 0.2450 |
0.39 | 16 | 1.85 | 0.2450 |
0.4 | 13 | 1.85 | 0.2450 |
From the two figures, we see that the synchronization error ranges of the wrist angle of left hind leg, right hind leg, left front leg, and right front leg are [-1.2, 1.2], [-0.9, 0.95], [-1.2, 0.9], and [-0.9, 1.45]. The synchronization error ranges of the angle between the forward direction of thigh movement and the left hind leg, right hind leg, left front leg, and right front leg are, respectively, [-0.95, 0.9], [-1.45, 0.6], [-1.3, 1.5], and [-1.6, -0.6].
For parameters a=−2,b=1,c=6.5,d=−2, the critical values τ04=0.2048 of the walking gait and k1=Re(a1)=0.8232>0,k2=Re(a2)=−0.1867<0,k1k2=−0.1537<0 are calculated. According to Theorem 2.1, the bifurcation periodic solution of system (1.1) is stable and supercritical at τ=0.2450>τ04=0.2048, as shown in Figure 10.
Figure 10 indicates that the periodic solution obtained by the system (1.1) corresponds to a stable walking gait when the parameters a=−2,b=1,c=6.5,d=−2,τ=0.2450. Therefore, model (1.1)(a=−2,b=1,c=6.5,d=−2) provides a reference model for the CPG walking gait model of goats walking uphill on an 18 degree slope when the time delay parameter τ=0.2450.
We employed the normal form theory according to Faria and Magalh˜aes to derive the normal form for Hopf bifurcation of a quadruped gait CPG model on the center manifold, and the bifurcation direction and stability of bifurcating periodic solution at the origin are analyzed. The quadruped gait CPG model was used as an inversion model, and goat gait models were constructed using the trust region inversion algorithm. The feasibility of using the CPG model as a goat gait model was verified through simulations.
The neural network model presented in this article can predict the goat gait to contribute to artificial intelligence. The weakness of this paper is that the synchronization errors of some points are slightly larger when using our mathematical model to simulate goat gait, especially walking gait. We will modify this model further in the future.
The authors declare they have not used Artificial Intelligence (AI) tools in the creation of this article.
This research is supported by the Fundamental Research Funds for the Central Universities, (No.2572022DJ08).
The authors declare there is no conflict of interest.
[1] |
M. Golubitsky, I. Stewart, P. Buono, J. Collins, Symmetry in locomotor central pattern generators and animal gaits, Nature, 401 (1999), 693–695. https://doi.org/10.1038/44416 doi: 10.1038/44416
![]() |
[2] |
P. Buono, M. Golubitsky, Models of central pattern generators for quadruped locomotion. Ⅰ. Primary gaits, J. Math. Biol., 42 (2001), 291–326. https://doi.org/10.1007/s002850000058 doi: 10.1007/s002850000058
![]() |
[3] |
P. Buono, Models of central pattern generators for quadruped locomotion. Ⅱ.Secondary gaits, J. Math. Biol., 42 (2001), 327–346. https://doi.org/10.1007/s002850000073 doi: 10.1007/s002850000073
![]() |
[4] |
L. Q. Liu, C. R. Zhang, Dynamic properties of VDP-CPG model in rhythmic movement with delay, Math. Biosci. Eng., 17 (2020), 3190–3202. https://doi.org/10.3934/mbe.2020181 doi: 10.3934/mbe.2020181
![]() |
[5] |
L. Q. Liu, X. X. Liu, C. R. Zhang, Realization of neural network for gait characterization of quadruped locomotion, J. Appl. Anal. Comput., 12 (2022), 455–463. https://doi.org/10.11948/20210005 doi: 10.11948/20210005
![]() |
[6] |
B. Strohmer, P. Manoonpong, L. B. Larsen, Flexible spiking CPGs for online manipulation during hexapod walking, Front. Neurorobotics, 14 (2020), 1–12. https://doi.org/10.3389/fnbot.2020.00041 doi: 10.3389/fnbot.2020.00041
![]() |
[7] |
C. Bal, Neural coupled central pattern generator based smooth gait transition of a biomimetic hexapod robot, Neurocomputing, 420 (2021), 210–226. https://doi.org/10.1016/j.neucom.2020.07.114 doi: 10.1016/j.neucom.2020.07.114
![]() |
[8] |
Y. Son, T. Kamano, T. Yasuno, T. Suzuki, H. Harada, Generation of adaptive gait patterns for quadruped robot with CPG network including motor dynamic model, Electr. Eng. Jpn., 155 (2006), 35–43. https://doi.org/10.1002/eej.20225 doi: 10.1002/eej.20225
![]() |
[9] |
T. T. Duc, I. M. Koo, Y. H. Lee, H. Moon, S. Park, J. C. Koo, et al., Central pattern generator based reflexive control of quadruped walking robots using a recurrent neural network, Robot. Auton. Syst., 62 (2014), 1497–1516. https://doi.org/10.1016/j.robot.2014.05.011 doi: 10.1016/j.robot.2014.05.011
![]() |
[10] |
J. Q. Zhang, F. Gao, X. L. Han, X. B. Chen, X. Y. Han, Trot gait design and CPG method for a quadruped robot, J. Bionic. Eng., 11 (2014), 18–25. https://doi.org/10.1016/S1672-6529(14)60016-0 doi: 10.1016/S1672-6529(14)60016-0
![]() |
[11] |
J. X. Zhao, T. Iwasaki, CPG control for harmonic motion of assistive robot with human motor control identification, IEEE Trans. Control Syst. Technol., 28 (2020), 1323–1336. https://doi.org/10.1109/TCST.2019.2910160 doi: 10.1109/TCST.2019.2910160
![]() |
[12] | H. Suzuki, H. Nishi, Animal gait generation based on human feeling for quadrupedal robot, Int. J. Innovative Comput., Inf. Control, 4 (2008), 3341–3348. |
[13] |
Z. Bhatti, Oscillator driven central pattern generator (CPG) system for procedural animation of quadruped locomotion, Multimedia Tools Appl., 78 (2019), 30485–30502. https://doi.org/10.1007/s11042-019-7641-1 doi: 10.1007/s11042-019-7641-1
![]() |
[14] | Y. J. Xu, Analysis and Simulation of Kinematic Characteristics of Goat's Multi-mode Gait, Masters Thesis, Jilin University, 2021. https://doi.org/10.27162/d.cnki.gjlin.2021.005610 |
[15] | G. Y. Zhang, Research on Bionic Goat Mechanism on Sloping Fields, Masters Thesis, Henan University of Science and Technolog, 2011. |
[16] |
T. Faria, L. T. Magalhães, Normal forms for retarded functional differential equation with parameters and applications to hopf bifurcation, J. Differ. Equations, 122 (1995), 181–200. https://doi.org/10.1006/jdeq.1995.1144 doi: 10.1006/jdeq.1995.1144
![]() |
[17] |
T. Min, Y. Cheng, M. L. Gu, H. H. You, Parameter estimation of nonlinear dynamic system and sensitivity, Comput. Eng. Appl., 49 (2013), 47–49. https://doi.org/10.3778/j.issn.1002-8331.1110-0488 doi: 10.3778/j.issn.1002-8331.1110-0488
![]() |
The initial value | Iterations | error f | Numerical solution |
6.1 | 11 | 11.6591 | 6.4300 |
6.3 | 9 | 11.6591 | 6.4300 |
6.5 | 5 | 11.6591 | 6.4300 |
7.2 | 9 | 11.6591 | 6.4300 |
The initial value of τ | Iterations | precision ε | Numerical solution |
0.25 | 5 | 1.85 | 0.2450 |
0.28 | 8 | 1.85 | 0.2450 |
0.39 | 16 | 1.85 | 0.2450 |
0.4 | 13 | 1.85 | 0.2450 |