1: %%
2: %% proceedings of Shizuoka symposium
3: %%
4: %% ========== documentstyle and stylefiles
5: \documentclass[twocolumn,preprintnumbers,amsmath,amssymb]{revtex4}
6: %\documentclass[preprint,showpacs,preprintnumbers,amsmath,amssymb]{revtex4}
7:
8: \usepackage{graphicx}% Include figure files
9: \usepackage{dcolumn}% Align table columns on decimal point
10: \usepackage{bm}% bold math
11:
12: %\nofiles
13:
14: \voffset=15mm
15: \setlength{\unitlength}{1cm}
16: \renewcommand{\topfraction}{0.99}
17: \renewcommand{\bottomfraction}{0.99}
18: \begin{document}
19:
20: \title{Collective Dynamics of Active Elements: \\
21: Task Allocation and Pheromone Trailing}
22:
23: \author{T. Mizuguchi}
24: \affiliation{Department of Mathematical Sciences, Osaka Prefecture University}%
25: \author{K. Sugawara}%
26: \affiliation{Department of Liberal Arts, Tohoku Gakuin University}%
27: %
28: \author{H. Nishimori}%
29: \affiliation{Department of Mathematical Sciences, Osaka Prefecture University}%
30: \author{T. Tao}%
31: \affiliation{Department of Mathematical Sciences, Osaka Prefecture University}%
32: \author{T. Kazama}%
33: \affiliation{Graduate School of Information Systems, University of Electro-Communications}%
34: \author{H. Nakagawa}%
35: \affiliation{Department of Mathematical Sciences, Osaka Prefecture University}%
36: \author{Y. Hayakawa}%
37: \affiliation{Department of Physics, Tohoku University}%
38: \author{M. Sano}%
39: \affiliation{Department of Physics, University of Tokyo}%
40:
41: \date{\today}% It is always \today, today,
42: % but any date may be explicitly specified
43:
44: \begin{abstract}
45: Collective behavior of active elements
46: inspired by mass of biological organisms is addressed.
47: Especially, two topics are focused on among
48: amazing behaviors performed by colony of ants.
49: First, task allocation phenomena are treated from the viewpoint
50: of proportion regulation of population between different states.
51: Using a dynamical model consisting of elements
52: and external ``stock materials'',
53: adaptability against various disturbances is numerically studied.
54: In addition,
55: a dynamical model for a colony ants interacting via two kind of pheromones is
56: studied, in which simulated ants, as a mass, are shown to make an efficient foraging
57: flexibly varying the foraging tactics according to feeding schedules.
58: Finally,
59: experiments are performed with robots moving in
60: virtual pheromone fields simulated by CG and CCD camera feedback system.
61: Trail formation processes are demonstrated by this
62: multi-robot system.
63: \end{abstract}
64:
65: \maketitle
66: % introduction ------------------------------------------------------------
67:
68:
69: \section{Introduction}
70: Among intriguing phenomena exhibited by biological mass,
71: it is remarkable that various functions or
72: performances of the whole mass
73: exceed simple superposition of each individual's ability.
74: The mass of individuals we consider here is that of
75: slime molds, (social) insects, fish, birds, etc.
76: For example, a kind of cellular slime mold shows
77: regulation of ratio between two different types of cell
78: during its differentiation process
79: \cite{Raper,Bonner,Loomis,Oyama}.
80: About the motion of the group,
81: birds flock, fish school and mosquitoes swarm
82: \cite{Wilson2, Edelstein-Keshet, Partridge, Inoue, Vicsek, Shimoyama}.
83: Social insects exhibit lots of amazing performances
84: like colony formation, age polyethism, group foraging, etc\cite{Wilson}.
85: It is worth noticing that these collective performances
86: do not require special individual(s)
87: which controls the behavior of the whole.
88: Hereditarily homogeneous individuals achieve
89: these collective behaviors
90: by interacting each other through direct sight
91: or chemical materials like c-AMP, pheromones, etc.
92: How does each of them ``feel'' the situation of the whole and
93: how does it determine its behavior?
94: Here, we focus on the task allocation phenomena
95: and the pheromone trail formation processes
96: among various collective behaviors described above.
97:
98: This article is organized as follows.
99: In the next section,
100: from the viewpoint of proportion regulation,
101: multi-task allocation phenomena are treated
102: with a dynamical model consisting of simple
103: elements and external variables.
104: In \S 3, foraging behavior of modeled ants was treated in which
105: ants leave and follow only the local information of pheromone, however
106: they can fulfill the highly efficient foraging
107: flexibly adapting to various feeding conditions.
108: In \S 4, real robots experiments are performed.
109: Using a virtual pheromone system (V-DEAR) which consists of
110: LC projector and CCD camera,
111: both task allocation process and trail formation process
112: are examined by multi-robot system.
113: Summary is given in \S 5.
114:
115: %% task allocation --------------------------------------------------------
116: \section{Task allocation}
117:
118: Division of labor ({\it polyethism}) performed by social insects
119: or cell differentiation observed in the life cycle
120: of {\it Dictyostelium discoideum} exhibit
121: typical proportion regulation phenomena.
122: Namely, individuals (or cells) with the same inheritance
123: divide into two or more groups and they play different roles.
124: The proportion of populations of each group is regulated
125: against external disturbance without special individual.
126: Regulation means a kind of adaptability,
127: i.e., for large disturbance like loss of some part of colony
128: or change environment,
129: individuals know the change of the situation
130: about the whole and change their behavior.
131: As a result of such autonomous process,
132: the proportion of population between the groups is maintained.
133: Several questions naturally arise for these phenomena, i.e.,
134: how do they know about the whole and how do they determine their behavior?
135: In this section, dynamical models for internal states of individuals and
136: external variables of chemical materials
137: are used to reproduce these phenomena
138: and consider their mechanism through numerical simulations.
139:
140: As a previous work, we suggest a dynamical model for
141: proportion regulation phenomena for two states.
142: The model consists of $N$ identical elements with
143: two internal variables,
144: i.e., activator $u_i(t)$ and inhibitor $v_i(t)$,
145: couple globally through their average quantities\cite{GlobalTuring}.
146: The dynamics of the internal variables $u_i(t)$ and $v_i(t)$ is
147: %$$
148: \begin{eqnarray}
149: \dot{u_i} &=& au_i - bv_i - u_i^3 + K_1(\bar{u} - u_i), \\
150: \dot{v_i} &=& cu_i - dv_i + K_2(\bar{v} - v_i),
151: \end{eqnarray}
152: %$$
153: where $\bar{\phantom{u}}$ denotes averaging over population,
154: i.e. , $\bar{u}\equiv \sum u_j / N$
155: and $\bar{v}\equiv \sum v_j/N$.
156: $a,b,c,$ and $d$ are non-negative parameters.
157: $K_1$ and $K_2$ are the non-negative susceptibilities
158: of each variables in the context of fast diffusion limit.
159: The model has following properties under suitable
160: parameter conditions $a-d<0$, $ad-bc<0$ and $K_1=0$:
161: (i) The population ``differentiates'' into two states for $K_2>K_{2c}$.
162: (ii) The population ratio between the two states
163: is regulated within some range which is obtained analytically.
164: (iii) The system maintains the ratio against large
165: disturbance by changing the state of individuals.
166: (iv) State of each element depends only on its initial condition.
167: The model, however, is only available for two states allocation
168: because of its simplicity.
169: In order to treat multi-states case,
170: we adopt another dynamical model in the next subsection.
171:
172: \subsection{Model}
173:
174: \begin{figure}[t]
175: \includegraphics[width=8cm]{fig1.eps}
176: \caption{\label{fig:rules}
177: Examples of rule between multi-states.
178: (a) sequential (b) intermediate, and (c) democratic.
179: Arrows denote the path which the rule allows to change.
180: Note that the paths between the states in rule (a) and (c)
181: are all symmetric while it is not in (b).
182: }
183: \end{figure}
184:
185: \begin{figure}[h]
186: \begin{picture}(7,5)(0,0)
187: \put(0.,0.0){\includegraphics[width=7cm]{fig2.eps}}
188: \put(2.4,1.7){$\uparrow$}\put(2.4,1.2){1}
189: \put(3.4,0.9){$\uparrow$}\put(3.4,0.4){2}
190: \put(4.4,1.9){$\uparrow$}\put(4.4,1.4){3}
191: \end{picture}
192: \caption{\label{fig:potential}
193: Typical shape of potential function $U$ with three states.
194: }
195: \end{figure}
196:
197: Unlike the two states case,
198: multi-states case requires detail setting up the problem,
199: i.e., not only the ratio between the states
200: but also the rule between the states are needed to be assigned.
201: Here, the rule is considered to be
202: determined by the connectivity between the states.
203: Several examples of rule are shown in Fig.\ref{fig:rules}.
204: We assume the sequential and symmetric rule (Fig.\ref{fig:rules}(a)) here,
205: and set up the situation as follows:
206: $N$ identical individuals,
207: $M$-states with proportion of population $p_j$ for $j$-th state
208: with $0 < p_j < 1$ and $\sum p_j = 1$.
209: $M=3$ and $p_j=\{0.5, 0.3, 0.2\}$ are used as typical setting up.
210:
211: Our model consists of $N$ identical elements and $M$ ``stock materials''.
212: The elements represent the individuals and
213: have internal variable $u_i(t)$, $i=1,...,N$.
214: The stock materials correspond the states and
215: their quantities are expressed by external variable $w_j(t)$ , $i=1,...,M$.
216: Each state is expressed by a valley of potential function $U(u)$,
217: \begin{eqnarray}
218: U(u) &=& \sum_j (\alpha_j-w_j(t)) U_j(u) + \alpha_0 U_0(u),\\
219: U_j(u) &=& -{e^{-\beta(u-s_j)} \over(1+e^{-\beta(u-s_j)})^2},\\
220: U_0(u) &=& u^2,
221: \end{eqnarray}
222: where $s_j$ and $\alpha_j$ denote the positions and depths
223: of $j$-th valley, respectively (Fig.\ref{fig:potential}).
224: $\beta$ gives an average depth of the valleys.
225: These quantities are determined by the environment.
226: $U_0$ is used for boundedness of the internal variables
227: of individuals.
228: $w_j(t)$, the quantity of $j$-th stock materials ,
229: changes effective depth of $j$-th potential and
230: obeys following equation
231: \begin{eqnarray}
232: \tau_j \dot{w}_j &=& -w_j + a \sum_i(-U_j(u_i)),
233: \end{eqnarray}
234: where the first term of the R.H.S represents decay process
235: and the second term is production process by the individuals
236: staying in the corresponding state.
237: $\tau_j$ is characteristic time of $j$-th material
238: and $a$ is a production rate.
239: The dynamics of $u_i(t)$, the internal variable of $i$-th element, is
240: \begin{eqnarray}
241: \dot{u}_i &=& - {\partial U(u) \over \partial u} + \eta,
242: \end{eqnarray}
243: where $\eta$ is a white uniform noise with amplitude $\nu$.
244: Individuals are considered to be subjected under the
245: time dependent potential $U(u)$ and the noise $\eta$.
246:
247:
248: \begin{figure}[h]
249: \begin{picture}(8,12.5)(0,0)
250: \put(1.3, 9.2){\frame{\includegraphics[height=2.5cm,width=4.9cm]{fig3a.eps}}}
251: \put(1.3, 6.5){\frame{\includegraphics[height=2.5cm, width=4.9cm]{fig3b.eps}}}
252: \put(1.3, 11.3){\normalsize (a)}
253: \put(1.3, 8.7){\normalsize (b)}
254: \put(3.5, 6.1){\normalsize $u_j$}
255: \put(1.1, 6.1){\normalsize $-1$}
256: \put(5.9, 6.1){\normalsize $1$}
257: \put(0.8, 11.2){\normalsize $j$}
258: \put(6.3, 11.2){\normalsize $U$}
259: \put(0.8, 8.6){\normalsize $j$}
260: \put(6.3, 8.6){\normalsize $U$}
261: \put(5.0, 11.3){\normalsize $t=0$}
262: \put(4.6, 8.7){\normalsize $t=100$}
263: \put(1.2, 0.5){\rotatebox{90}{\includegraphics[width=5cm]{fig3c.eps}}}
264: \put(1.3, 5.1){\normalsize (c)}
265: \put(3.6, 0.1){\normalsize $u_j$}
266: \put(1.2, 0.1){\normalsize $-1$}
267: \put(6.0, 0.1){\normalsize $1$}
268: \put(0.6,2.8){\normalsize $t$}
269: \put(0.9,0.5){\normalsize $0$}
270: \put(0.5,5.3){\normalsize $100$}
271: \put(7.0, 0.5){\normalsize $0$}
272: \put(7.0, 5.3){\normalsize $60$}
273: \end{picture}
274: \caption{\label{fig:noiseless}
275: State of $j$-th element $u_j$ (open rectangle) and
276: the shape of the potential at $t=0$ (a) and $t=100$ (b).
277: The potential is modified by stock materials $w_j$.
278: To display the modification by $w_j$,
279: initial shape of the potential ($w_j=0$) are also plotted in (b).
280: (c) Evolution of distribution function.
281: Color scale denotes number of individuals from $0$ to $60$.
282: }
283: \end{figure}
284:
285: \subsection{Noiseless case}
286:
287: We now show the results of numerical simulations.
288: Following conditions are used in this and the next subsections:
289: $N=100$, $a=3.5$, $\beta=20.0$, $\tau_j=1$,
290: $\alpha_0=0.1$, $\alpha_1=0.3$, $\alpha_2=0.5$, $\alpha_3=0.2$,
291: $s_1=-0.3$, $s_2=0$ and $s_3=0.3$.
292: Typical snapshots and the evolution of the system
293: are displayed in Fig.\ref{fig:noiseless} for the noiseless case $(\nu = 0)$.
294: As an initial condition,
295: the internal variable $u_j(0)$ of each element is randomly distributed
296: and the stock materials $w_j(0) $ are all set be zero (Fig.\ref{fig:noiseless}(a)).
297: Each element feels the potential $U(t)$
298: and gathers toward the valley for each state.
299: Simultaneously, the elements which staying at the $j$-th valley
300: product the corresponding stock materials $w_j$ and
301: change effective depth of the potential.
302: Designed ratio (3:5:2) is approximately achieved and
303: the shape of the potential becomes almost flat
304: at $t=100$ (Fig.\ref{fig:noiseless}(b)).
305: The evolution of distribution function shows
306: the designed ratio is almost established
307: in early stage (Fig.\ref{fig:noiseless}(c)).
308:
309: \begin{figure}
310: \begin{picture}(8,15)(0,0)
311: \put(1.5,10.5){\rotatebox{90}{\includegraphics[height=6cm,width=4cm]{fig4a.eps}}}
312: \put(1.7,14.0){\normalsize (a)}
313: \put(4.1,10.1){\normalsize $u_j$}
314: \put(1.4,10.1){\normalsize $-1$}
315: \put(6.6,10.1){\normalsize $1$}
316: \put(0.9,12.2){\normalsize $t$}
317: \put(1.1,10.5){\normalsize $0$}
318: \put(0.8,14.2){\normalsize $100$}
319: \put(1.5,11.2){\normalsize $\rightarrow$}
320: \put(1.1,11.2){\normalsize $t^*$}
321: \put(1.5,5.5){\rotatebox{90}{\includegraphics[height=6cm,width=4cm]{fig4b.eps}}}
322: \put(1.7,9.0){\normalsize (b)}
323: \put(4.1,5.1){\normalsize $u_j$}
324: \put(1.4,5.1){\normalsize $-1$}
325: \put(6.6,5.1){\normalsize $1$}
326: \put(0.9,7.2){\normalsize $t$}
327: \put(1.1,5.5){\normalsize $0$}
328: \put(0.8,9.2){\normalsize $100$}
329: \put(1.5,6.2){\normalsize $\rightarrow$}
330: \put(1.1,6.2){\normalsize $t^*$}
331: \put(1.5,0.5){\rotatebox{90}{\includegraphics[height=6cm,width=4cm]{fig4c.eps}}}
332: \put(1.7,4.0){\normalsize (c)}
333: \put(4.1,0.1){\normalsize $u_j$}
334: \put(1.4,0.1){\normalsize $-1$}
335: \put(6.6,0.1){\normalsize $1$}
336: \put(0.9,2.2){\normalsize $t$}
337: \put(1.1,0.5){\normalsize $0$}
338: \put(0.8,4.2){\normalsize $100$}
339: \put(1.5,1.2){\normalsize $\rightarrow$}
340: \put(1.1,1.2){\normalsize $t^*$}
341: \end{picture}
342: \caption{\label{fig:noiseless2}
343: Response for the disturbances added at $t^*$.
344: The evolution of distribution function is displayed.
345: (a) Insensitivity for small disturbances.
346: (b) Adaptability for large disturbances.
347: (c) Invalid case against large disturbances.
348: }
349: \end{figure}
350:
351: The response of the system for various type of
352: disturbance is shown in Fig.\ref{fig:noiseless2}.
353: First, a few elements in state 3 are moved to
354: state 2 at $t^*$ as a small disturbance.
355: The system is insensitive and reconstruction
356: of distribution does not occur (Fig.\ref{fig:noiseless2}(a)).
357: Next, large disturbance is given, i.e.,
358: all the elements in state 1 are moved to state 2.
359: In this case, the reconstruction process occurs and
360: the ratio between the states are regulated again
361: (Fig.\ref{fig:noiseless2}(b)).
362: Finally, an inadequate example is presented.
363: When all the elements in state 3 are moved to state 2,
364: undesired two clusters is realized (Fig.\ref{fig:noiseless2}(c)).
365: This is because the system is trapped in local minimum of the potential.
366: This fault can be avoided by the noise term as shown in the next subsection.
367:
368: \begin{figure}
369: \begin{picture}(6,9)(0,0)
370: \put(0.5,5){\rotatebox{90}{\includegraphics[height=5.5cm,width=4cm]{fig5a.eps}}}
371: \put(0.7,8.5){\normalsize (a)}
372: \put(2.9,4.6){\normalsize $u_j$}
373: \put(0.4,4.6){\normalsize $-1$}
374: \put(5.2,4.6){\normalsize $1$} \put(-0.1,6.9){\normalsize $t$}
375: \put(0.1,5.0){\normalsize $0$}
376: \put(-0.2,8.7){\normalsize $100$}
377: \put(0.5,5.7){\normalsize $\rightarrow$}
378: \put(0.1,5.7){\normalsize $t^*$}
379: \put(0.2,0){\reflectbox{\rotatebox{90}{\includegraphics[height=5.4cm,width=4cm]{fig5b.eps}}}}
380: \put(0.7,3.4){\normalsize (b)}
381: \put(2.9,-0.2){\normalsize $u_j$}
382: \put(5.2,-0.2){\normalsize $1$}
383: \put(0.4,-0.2){\normalsize $-1$}
384: \put(-0.1,2.1){\normalsize $t$}
385: \put(0.1,0.2){\normalsize $0$}
386: \put(-0.2,3.6){\normalsize $100$}
387: \put(0.5,0.7){\normalsize $\rightarrow$}
388: \put(0.1,0.7){\normalsize $t^*$}
389: %
390: \end{picture}
391: \caption{\label{fig:noisy}
392: Typical evolution in noisy case.
393: (a) Noise is added from $t^*$.
394: (b) Evolution of randomly selected elements in (a).
395: }
396: \end{figure}
397:
398: \subsection{Noisy case}
399: We now consider noisy case, $\nu \ne 0$.
400: The noise introduced here may have
401: physical, chemical or biological origin,
402: like the thermal noise of the system,
403: the fluctuation of distribution of the chemical substances,
404: diversity of the age of the individuals
405: %, or heterogeneity of the gene
406: .
407: Added noise kicks individuals away from the potential valley
408: and escape from the local minima is expected.
409: Fig.\ref{fig:noisy} displays a typical
410: evolution of the system in noisy case ($\nu=4$) .
411: The distribution becomes broad but the ratio does not change
412: (Fig.\ref{fig:noisy}(a)).
413: Fig.\ref{fig:noisy}(b) shows the evolution of
414: randomly selected elements as samples of behavior of individuals.
415: During the time displayed, most of the elements stay in the same state.
416: A few elements, however, change their internal state by the noise.
417: The larger $\nu$ becomes, the more frequent
418: hopping process between the states occurs.
419: With too strong noise, the distribution function
420: collapses into single peak.
421: On the other hand,
422: the system hardly escapes from the local minima
423: if the noise is too weak.
424: In this sense there is an adequate noise amplitude.
425: Actually, noise added from $t^*$ set the trapped system
426: free from the local minimum as shown in Fig.\ref{fig:recovery},
427: where the initial conditions are set to be trapped case
428: in Fig.\ref{fig:noiseless2}(c).
429:
430: \begin{figure}
431: \begin{picture}(6,5)(0,0)
432: \put(0.5,0.5){\rotatebox{90}{\includegraphics[height=5.5cm,width=4cm]{fig6.eps}}}
433: %\put(0.7,3.5){\normalsize (a)}
434: \put(2.9,0.1){\normalsize $u_j$}
435: \put(0.4,0.1){\normalsize $-1$}
436: \put(5.2,0.1){\normalsize $1$}
437: \put(-0.1,2.2){\normalsize $t$}
438: \put(0.1,0.5){\normalsize $0$}
439: \put(-0.2,4.2){\normalsize $100$}
440: \put(0.5,1.2){\normalsize $\rightarrow$}
441: \put(0.1,1.2){\normalsize $t^*$}
442: \end{picture}
443: \caption{\label{fig:recovery}
444: Recovery from the invalid case shown in Fig.\ref{fig:noiseless2}(c).
445: Noise ($\nu=4$) is added from $t^*$.
446: }
447: \end{figure}
448:
449: \begin{figure}
450: \begin{picture}(6,5)(0,0)
451: \put(0.5,0.5){\rotatebox{90}{\includegraphics[height=5.5cm,width=4cm]{fig7.eps}}}
452: \put(2.9,0.1){\normalsize $u_j$}
453: \put(0.4,0.1){\normalsize $-1$}
454: \put(5.2,0.1){\normalsize $1$}
455: \put(-0.1,2.2){\normalsize $t$}
456: \put(0.1,0.5){\normalsize $0$}
457: \put(-0.2,4.2){\normalsize $100$}
458: \put(0.5,1.2){\normalsize $\rightarrow$}
459: \put(0.1,1.2){\normalsize $t^*$}
460: \end{picture}
461: \caption{\label{fig:changedesign}
462: Ratio design is changed from 3:5:2 to 1:5:4 from $t^*$.
463: Noise ($\nu=4$) are added from $t=0$.
464: }
465: \end{figure}
466:
467: As another example of the adaptability of the system,
468: the environment parameter is changed.
469: The shape of the potential characterized by
470: $(\alpha_1, \alpha_2, \alpha_3)$
471: is changed from $(0.3, 0.5, 0.2)$ to $(0.1, 0.5, 0.4)$
472: at $t^*$ (Fig.\ref{fig:changedesign}).
473: The system adapts to the new environment quickly.
474:
475: \begin{figure}
476: \begin{picture}(6,11)(0,0)
477: \put(0.5,5.5){\rotatebox{90}{\includegraphics[height=5.5cm,width=4cm]{fig8a.eps}}}
478: \put(0.7,9.0){\normalsize (a)}
479: \put(2.9,5.1){\normalsize $u_j$}
480: \put(0.4,5.1){\normalsize $-1$}
481: \put(5.2,5.1){\normalsize $1$}
482: \put(-0.1,7.2){\normalsize $t$}
483: \put(0.1,5.5){\normalsize $0$}
484: \put(-0.4,9.2){\normalsize $1000$}
485: \put(1.6,10.0){\normalsize death}
486: \put(2.0,9.6){\normalsize $\downarrow$}
487: \put(4.2,10.0){\normalsize birth}
488: \put(4.5,9.6){\normalsize $\downarrow$}
489: \put(0.3,0.3){\rotatebox{90}{\includegraphics[height=5.4cm,width=4.3cm]{fig8b.eps}}}
490: \put(0.7,4.0){\normalsize (b)}
491: \put(2.7,0.1){\normalsize $N(t)$}
492: \put(5.3,0.1){\normalsize $0$}
493: \put(0.4,0.1){\normalsize $100$}
494: \put(-0.1,2.4){\normalsize $t$}
495: \put(0.2,0.6){\normalsize $0$}
496: \put(-0.4,4.2){\normalsize $1000$}
497: \end{picture}
498: \caption{\label{fig:open}
499: Example of ``open'' system with ``birth'' and ``death'' process.
500: The time series of distribution function (a) and
501: total number of elements (b) are displayed.
502: }
503: \end{figure}
504:
505: How about the disturbance of changing total number $N$ ?
506: We demonstrate an open system
507: including ``birth'' and ``death'' process
508: as a caricature of "polyethism" observed
509: in colony of social insect.
510: As a birth process of individuals,
511: a new element enters into the system
512: with constant internal value $u_{new} = 0.7$
513: for every five time units.
514: As a death process, elements which exceed
515: the lower threshold $u_{death} = -0.31$ are removed
516: from the system.
517: Fig.\ref{fig:open} shows the typical evolution of
518: distribution function and population of the whole.
519: Both the proportion between the states and
520: the total number are maintained with fluctuations.
521:
522: % pheromone trailing ---------------------------------------------
523: \section{PHEROMONE TRAILING}\label{sec:introduction}
524: So far, various studies have been made on the collective behavior
525: of social insects \cite{Wil,De1,Sug1}.
526: Especially, the trail formation of ants \cite{Wil,De1,Hel,Schw1,Bo,Nishinari,Tao}
527: have called a wide interest as a remarkably synergetic
528: behavior by presumably not highly intelligent individuals.
529: In this section, setting a simple set of rules of which several ingredients
530: are following previous studies \cite{De1,Hel}, we
531: investigate the relation between i)the schedules of feeding which
532: represent the unsteady natural environment, ii)emerging patterns of trails
533: which connect between a nest with food resources, and iii)the foraging efficiency
534: as a group. Our final purpose is to connect the pattern formation of the trail
535: to the group strategy of foraging ants under an unsteadily varying environment.
536:
537:
538: \subsection{Model}
539:
540: The model consists of a colony of N movable agents (we call them 'ants')
541: which are situated on a 2D honeycomb lattice with periodic boundary
542: condition. Each of these ants is located at one site in the lattice
543: and is heading to one of the six nearest sites(Fig.9) to which, the right-hand neighbor or the left-hand, she will move according to the
544: dynamical rule described below.
545: The task for ants is foraging, that is, finding food and carrying it back to a nest and the feeding sites are located at two corners of a regular triangle
546: with edge-length $L$ and the nest as the third corner (fig.10).
547: The amount and the schedule of feeding are
548: controlled through a pair of control parameters.
549: {$T$, $M$} as explained in Fig.10.
550: \begin{figure}
551: %\begin{center}
552: \includegraphics[width=0.8\linewidth]{fig9.eps}
553: %\includegraphics{./eps2/dHWL3.eps}
554: \caption{
555: %\scriptsize
556: Each ant is, at each time step,
557: heading one of the six nearest sites.
558: This heading direction corresponds to the moving direction in the previous
559: time step. The possible moving directions in each time step are:
560: the forward (=facing) direction and its neighboring (right and left)
561: directions. To choose one of them, a stochastic rule with the weight,
562: $P_\alpha = exp(-\Delta^\alpha/T)/Z$
563: is adopted where $Z$ is the normalization factor. Here $\alpha$ indicates forward, right, or left direction
564: relative to the heading direction, and
565: $\Delta^\alpha \equiv \rho^\alpha_\beta-\rho_\beta({\bf x},t)$
566: is the gradients of pheromone density to the candidate directions
567: where $\beta$ is the index to indicate the
568: recruit pheromone(for mode-II ants) or the foot pheromone(for mode-III ants).
569: Note the above weight is not applicable for ants in mode-I
570: which is making random walk independent of pheromone field.
571: %\scriptsize
572: }
573: \end{figure}
574:
575: As the mean for the communication among ants, two types of pheromones
576: are secreted and perceived by individual
577: ants according to their temporal modes which consist of: i)the semi-random walk mode (mode-I), ii)the exploring mode (mode-II) and
578: iii)the homing mode (mode-III) (Fig.11).
579: The kind of pheromones considered in the present model consists of:
580: i)recruit pheromone and ii)foot pheromone.
581: In addition these pheromones evaporate/decay in a constant rate, also
582: diffuse to the nearest sites.
583:
584: \subsection{Simulation}
585: All the simulations are performed on a $150 \times 150$ honeycomb lattice
586: with periodic boundary condition.
587: At the initial time step, N=500 ants are located at the nest,
588: then simultaneously released from the nest. Note the unit time
589: in our simulation corresponds to one Monte Calro step within which
590: N randomly chosen ants will sequentially take one individual step.
591: Under a proper combination of several fixed parameters, e.g., the evaporation/diffusion rate of pheromones,
592: and varying a pair of control parameters $\{M,T\}$,
593: \begin{figure}
594: \includegraphics[width=0.5\linewidth]{fig10.eps}
595: \caption{The method of feeding.
596: Feeding sites are located at two corners of an
597: equilateral triangle, the 3rd corner of which is the nest site.
598: The amount $M$ of food is supplied,
599: by turn, from alternative feeding sites at every feeding interval $T$.}
600: \end{figure}
601: \begin{figure}
602: \includegraphics[width=0.8\linewidth]{fig11.eps}
603: \caption{Three modes of ants. In mode-I ants make a semi-random walk secreting foot pheromone,
604: in mode-II ants explore new food following recruit pheromone in which mode they secrete foot pheromone,
605: in mode-III ant try to come back to nest following foot pheromone while secreting recruit pheromone.
606: The mode switch from mode-I to mode-II takes place if an ant reaches a site where the density of recruit pheromone
607: exceeds a threshold value, whereas a mode-II/mode-III ant turns into mode-III/mode-I on arriving at food/nest.
608: }
609: \end{figure}
610: the formation process of trail patterns and
611: the subsequent efficient foraging are investigated.
612: We are particularly interested in the relation between i)the geometries of trails
613: and ii)the accompanying foraging efficiency.
614:
615:
616:
617: \bigskip
618: \subsection{Typical trail shapes}
619:
620: Once a certain shape of trail is established it is recognized as one of following three types
621: i)V-shaped trail, ii)Y-shaped trail
622: or iii)/-shaped trail as shown in Fig.12.
623: \begin{figure}
624: \includegraphics[width=0.7\linewidth]{fig12.eps}
625: \caption{Typical shapes of trails obtained in the simulation.
626: i)V-shaped trail, ii)Y-shaped trail, iii)/-shaped trail.
627: Here dark shading in each figure
628: means the density of the ants averaged over a time significantly longer than $T$, the interval between
629: successive feeding events.}
630: \end{figure}
631: Notice that these shapes are recognizable and definable after averaging the density field of ants over a period much longer
632: than the feeding period $T$.
633: The condition, in the $M-T$ space, for the appearance of each of these trails are shown in Fig.13 which indicates
634: that the most relevant quantity for the trail geometry seems to be $M/T$.
635: As $M/T$ increases, the shape of trail
636: varies roughly from the V-shape to the Y-shape and finally the /-shaped
637: trail is obtained.
638:
639: \begin{figure}
640: \includegraphics[width=0.7\linewidth]{fig13.eps}
641: \bigskip
642: \caption{The relation between emergent trail patterns
643: and the combination of feeding schedule parameters,
644: \{$M$,$T$\}. The dark shading in each figure
645: means the density of ants averaged over a time significantly longer
646: than $T$ and over the ensemble of 5 simulations for the V-shaped and the Y-shaped trails.
647: For the /-shaped trail the ensemble average is not taken because of their intrinsic asymmetry.}
648: \end{figure}
649:
650: \bigskip
651: \noindent
652: \subsection{Foraging efficiency}
653:
654: To investigate the relationship between the efficiency of foraging
655: and trail geometry, we introduce averaged foraging efficiency $E_{av}$
656: which is the total amount of food
657: carried into the nest per unit time.
658: Hence, the relation between $M/T$ and $E_{av}$
659: is measured under various feeding rate $M/T$ (here $T$ is fixed at $T=600$). In figure 14,
660: a characteristic relation between $M/T$ and $E_{av}$ is seen
661: until $E_{av}$ reaches an saturation value.
662: \begin{figure}
663: \includegraphics[width=0.7\linewidth]{fig14.eps}
664: \caption{The relation between the
665: supplied amount of food per unit time $M/T$ and the long-term averaged foraging efficiency $E_{av}$.
666: There are five characteristic zones $A, A', B, C$ and $D$.
667: In zone $A$, the V-shaped trail appears, while in zones $A',B$
668: the Y-shaped trail appears, and in zones $C$ and $D$ the /-shaped trail is
669: seen. Up to the zone $A'$ the {\it perfect foraging} is
670: attained whereas in zone $D$ the amount of supplied food exceeds the
671: carrying capacity of ants thus $E_{av}$ reaches a saturation value $a$.}
672: \end{figure}
673: In this relation,
674: considering the emergent geometry of trails, roughly five characteristic zones
675: are recognized as indicated by symbols $A, A', B, C$ and $D$.
676: In zone $A$ with the emergence of the V-shaped trail, an almost linear $\frac{M}{T}-E_{av}$ relation is seen,
677: and its slope is 1 which corresponds to the {\it perfect foraging} , namely,
678: ants collect all the supplied food at both feeding sites
679: along the V-shaped trail.
680: On the other side, in zone $D$ in Fig.14, $\frac{M}{T} - E_{av}$ relation reaches a plateau, where
681: the averaged foraging efficiency $E_{av}$ amounts to the maximum possible foraging efficiency $E_{max}$
682: realized by the present number of ants.
683: The last situation occurs when the amount of supplied food at each feeding site exceeds (or is equal to) the carrying capacity of
684: ants, i.e., $M/2T \ge E_{max}$ is satisfied where the equivalence holds at the left edge of zone $D$.
685: These two zones $A$ and $D$ are characterized by simple foraging strategies with
686: a straightly extending trail(a pair of straight trails) for the /-shaped trail (for the V-shaped trail).
687:
688:
689: In the case of intermediate amount of
690: food between zone A and D, simple foraging tactics by use of straight trail(s)
691: fails to give an optimal foraging.
692: For example, as $M$ increases over a critical value $M_{C1}$ in zone $A'$,
693: if ants stick to the above mentioned simple strategy with the V-shaped trail (a pair of straight trails),
694: the sum of consuming periods for i) finding new food at alternative feeding site and that for
695: ii)carrying away all the food at the feeding site, exceeds the feeding interval $T$.
696: It means a part of food remains to be unexhausted even at the end of each feeding interval $T$.
697: However, in our simulation at zone $A'$, instead of sticking to such a simple strategy with the V-shaped trail,
698: ants {\it invent} Y-shaped trail which, by continuous use of the steadily maintained trunk trail
699: near the nest as a common path and
700: leveraging the junction of Y as the 'base camp' for the exploration of new food,
701: serves the system with a shorter finding time of new food at the alternative feeding site.
702: As a result, the {\it perfect foraging} is kept realized over $M=M_{C1}$ up to $M=M_{C2}$ in Fig.14\cite{note1}.
703: Certainly, along the Y-shaped trail, the distance ants need to walk from the nest to food is longer than
704: that along a V-shaped trail. Therefore, once a trail is established until all the food at the corresponding feeding site
705: is exhausted, the temporal foraging efficiency along the Y-shaped trail falls below that along the V-shaped trail.
706: In this sense, the transition from the V-shaped trail to the Y-shaped trail is a trade-off strategy within a restricted
707: food environment\cite{note1}.
708: This {\it perfect foraging} of both feeding sites of food is kept maintained before a certain value
709: of $M(=M_{c2}>M_{c1})$ over which no tactics can realize the perfect foraging.
710:
711: In figure 15 we show the theoretically estimated
712: foraging efficiencies realized by three types of trails \cite{Tao} as the functions of $M/T$.
713: The figure indicates that the most efficient trail pattern varies, as $M$ increases, from V( or Y) to Y
714: and finally to /. which estimation, at least qualitatively, corresponds to the simulation outcomes
715: shown in Fig.14.
716: It is remarkable that, even in the {\it imperfect foraging regime} over $M=M_{c2}$, through the proper change of trail geometry,
717: ants in the simulation keep making the almost optimal foraging close to the theoretically estimated
718: one in Fig.15 under respective feeding conditions.
719: As the extreme case in zone-D, the /-shaped trail is built which is optimized in the sense that
720: food is sufficient not to run out even when ants concentrate on one feeding site and in such a case
721: straight trail is most efficient because of the minimal path length between the nest and food.
722: Although the setup of the present simulation is very simple, similar arguments to the present are considered to hold in more complex and
723: more realistic situations like the trail formation under a larger number of feeding sites(Fig.16),
724: in which situations the presumably 'unintelligent' ants are guessed to make globally efficient behavior only via local information
725: carried by pheromones.
726:
727: Still, the basic question; why such an almost optimized foraging is realized only with the pheromone
728: trailing tactics, is left unsolved.
729: However it is strongly expected that the further exploration of the pheromone trailing will serve
730: us with some intrinsic insights on the reason why group behavior of presumably not intellectual elements
731: is kept realized in the nature. Also, the application of the pheromone trailing tactics to
732: a wider range of systems, e.g., interacting multi-robot system, is one of intriguing issue as
733: described in the next section.
734:
735: \bigskip
736: \bigskip
737: \begin{figure}
738: %\begin{figure}
739: \includegraphics[width=0.7\linewidth]{fig15.eps}
740: \caption{Theoretically estimated relations between $M/T$ and averaged foraging
741: efficiency $E_{av}$ for the V-shaped trail with
742: the real line, for the Y-shaped trail with the dotted line,
743: and for the /-shaped trail with the broken line, respectively.
744: According to these relations, the most efficient shape of trail varies,
745: as $M/T$ increases, from
746: V(or Y) to Y. Finally /-shape is seen most efficient. This way of shape shifting corresponds to the emergent trail
747: shapes in our simulation with $T$=600.
748: Note this is the output of a crude theoretical approximation, so it contains qualitative deviations
749: compared to those obtained in the simulation shown in Fig.14. }
750: %\end{figure}
751: \end{figure}
752: \begin{figure}
753: \includegraphics[width=0.5\linewidth]{fig16.eps}
754: \caption{Typical trail pattern in the case of four feeding sites}
755: \end{figure}
756:
757: \bigskip
758:
759: % robot experiments ---------------------------------------------
760: \section{Robot Experiments}
761:
762: In previous section, we discuss the mathematical model of ant-like system, but we are also trying to develop the robots system towards the realization of ant-inspired cooperative multi-robot system.
763:
764: In engineering field, research in the area of multi-robot systems has been very active, and many researchers and engineers are currently studying the behavior of robots inspired by ant colonies\cite{Cao}.
765: In most cases, physical media such as light, sound, radio wave are used for the communication between the robots, and some efforts have also been made for realizing chemical communication robot system\cite{Hayes}.
766: Unfortunately, it is not easy to get proper chemical sensors at this stage.
767: Moreover, chemical materials, especially gas, are invisible and it is quite difficult to observe how they spread and affect robots' behaviors.
768:
769: In our previous work, we propose "Virtual Dynamic Environment for Autonomous Robots (V-DEAR)" for real robot experiment\cite{DARS7}, in which the pheromones are virtually expressed by light information.
770:
771: \subsection{Experimental apparatus}
772:
773: In V-DEAR system, pheromones are replaced with computer graphics projected on the floor.
774: Robots decide their actions following the color information of the projected CG using color sensors.
775: As virtual pheromones are represented as CG, we can avoid the problems described above.
776: In addition, we can easily control the rate of diffusion, evaporation, diversity, etc. of the virtual chemical materials.
777:
778: Fig. \ref{VDEAR}(a) shows the schematic of this system.
779: This is composed of LC projector to project the CG and the CCD camera to trace the position of the robots in the field.
780: The robot moving on the field has sensors on the top to detect the color and brightness of the field(fig. \ref{VDEAR}(b)), and determines its behavior.
781:
782: \begin{figure}[h]
783: \includegraphics[width=\columnwidth]{fig17.eps}
784: \caption{Virtual Dynamic Environment for Autonomous Robots (V-DEAR). (a)Schematic of the system. (b)Color sensors on the top.}
785: \label{VDEAR}
786: \end{figure}
787:
788: Combining the position information of the robots acquired from CCD camera and the projected CG by projector, we can realize the dynamic interaction between the environment and robots.
789:
790:
791: \subsection{Task allocation}
792:
793: Using this system, we made experiments of task allocation.
794: In this experiment, each robot can take two states, staying and foraging, and decide one of the two states based on the total amount of food stock in the nest.
795: The stock is consumed constantly depending on the number of staying robots in the nest, and it is added when the foraging robot returns to the nest.
796:
797: Fig. \ref{fig2}(a) and (b) show the experimental field and the schematic, respectively.
798: A half of the field is "the nest" and the robots usually move around in this area.
799: The other half is "the work space" and the robots moving around this area carry back food.
800: V-DEAR system monitors the total amount of food stock and expresses it by the brightness of the nest.
801: Each robot detects it and decides their state autonomously.
802: The robots were adjusted to show nearly 1/3 working ratio in case of six robots system (Fig. \ref{fig2}(c)).
803: We reduced the number of robots by half under the same condition, but you can see the rate is kept 1/3 approximately (Fig. \ref{fig2}(d)).
804:
805:
806: \begin{figure}[h]
807: \begin{center}
808: \includegraphics[width=\columnwidth]{fig18.eps}
809: \caption{(a)Experimental field for task allocation. (b)Schematic of the
810: experiment. (c)The number of robots in each state(six robots). (d)The
811: number of robots in each state(three robots).}
812: \label{fig2}
813: \end{center}
814: \end{figure}
815:
816: \subsection{Pheromone Trailing}
817:
818: We also made foraging experiments by multi-robot system.
819: Fig. \ref{fig3}(a) shows the basic behavior of the foraging robots.
820: On discovering a food, the robot turns on a LED on the top and moves towards the nest.
821: The V-DEAR system detects the LED and projects a CG pheromone trail during the LED is turned on.
822: When the robot arrives at the nest, it turns off the LED and changes into the searching state.
823: If it finds the pheromone trail, it follows the trail.
824:
825: Fig. \ref{fig3}(b)(c) show the snapshot of the foraging by two robots in case of high evaporation rate and low evaporation rate, respectively, and the drawings at the right endpoint are the trajectories of the robots.
826: In case that the evaporation rate of the pheromone is high, the pheromone trail hardly remains.
827: However, the evaporation rate is low, a stable trail is formed between the food point and their nest.
828:
829: \begin{figure}[h]
830: \begin{center}
831: \includegraphics[width=\columnwidth]{fig19.eps}
832: \caption{(a)Basic behavior of the foraging robot. On discovering food, the robot lays chemical trail while returning to nest. The robot detecting the pheromone follows the trail.
833: (b)In case that the rate of the evaporation is high.
834: (c)In case that the rate of the evaporation is low.}
835: \label{fig3}
836: \end{center}
837: \end{figure}
838:
839:
840: % Summary --------------------------------------------------------
841: \section{summary}
842: Here in this article, we have investigated two topics
843: among intriguing behaviors exhibited by biological mass.
844: In \S 2, we proposed a model for multi(three)-tasks
845: allocation in which individuals feel
846: the situation of the whole of the colony
847: through the stock materials $w_j(t)$
848: and determines its behavior using
849: its internal variables $u_i(t)$ and potential $U(t)$.
850: The proportion regulation against several
851: types of disturbances was exhibited qualitatively.
852: As for general task rule like more complex ones or
853: especially asymmetric ones,
854: further improvement seems to be required.
855:
856: In \S 3, foraging behavior of ants was simulated, in which
857: ants rely only on the local information of two kinds of pheromones and
858: move according to a set of primitive rules.
859: However, they fulfill a highly efficient foraging flexibly adapting to various feeding conditions.
860: Although present simulations are restricted to rather simple
861: cases, same kind of argument is expected to be available for
862: more complex and more realistic situations.
863:
864: In \S 4, small scale experiments of
865: task allocation process and trail formation process
866: were examined using multi-robot and V-DEAR system
867: which simulates pheromones by light information.
868:
869:
870: The combination of the above investigations
871: would uncover the fundamental mechanism for
872: the formation, the maintenance and the evolution
873: of various types of collective dynamics exhibited by
874: a wide class of active elements.
875:
876:
877: % Acknowledgments --------------------------------------------------------
878: \begin{acknowledgements}
879: This work is supported by the
880: Japanese Grand-in-Aid for Encouragement of Young Scientists from the
881: Ministry of Education, Science and Culture (No.15760291 and No.16605008)
882: \end{acknowledgements}
883:
884: % references --------------------------------------------------------
885: \begin{thebibliography}{99}
886: \bibitem{Raper}K.B. Raper and J.Elisha, {\it Mitchell Scient. Soc.} {\bf 56} (1940) 241.
887: \bibitem{Bonner}J. T. Bonner, {\it Q. Rev. Biol.} {\bf 32} (1957) 232; {\it The Cellular Slime Molds}, Princeton Univ. Press, Princeton, New Jersey, 1967, 2nd ed.
888: \bibitem{Loomis}W. F. Loomis, {\it Dictyostelium discoideum: A developmental System}, Academic Press, New York, 1975.
889: \bibitem{Oyama} M. Oyama, K. Okamoto and I. Takeuchi, {\it J. Embryol. Exp. Morph. }{\bf 75} (1983) 293.
890: \bibitem{Wilson2}E.O. Wilson, {\it Sociobiology}, Harvard, Cambridge, MA, 1975.
891: \bibitem{Edelstein-Keshet}L. Edelstein-Keshet,
892: in: W. Alt (Ed.), {\it Lecture Notes in Biomathematics, Vol.89}, Springer, Berlin, 1990, p.528.
893: \bibitem{Partridge}B. L. Partridge, {\it Sci. Am.} {\bf 246} (1982) 90.
894: \bibitem{Inoue}M. Inoue, {\it Schooling of Fishes; behavior} (Kaiyo-shuppan, Tokyo, 1981) (in Japanese).
895: \bibitem{Shimoyama}N. Shimoyama et al. {\it Phys. Rev. Lett.}, {\bf 76} (1996) 3870.
896: \bibitem{Vicsek}T. Vicsek, A. Czirok, E. Ben-Jacob, I. Cohen, and O. Shochet, {\em Phys. Rev. Lett.}, {\bf 75} (1995) 1226--1229.
897: \bibitem{Wilson}E.O. Wilson, {\it The Insect Societies}, Oxford Univ. Press, London, 1971.
898: \bibitem{GlobalTuring}T. Mizuguchi and M. Sano, {\it Phys. Rev. Lett.}, {\bf 75} (1995) 966.
899: \bibitem{Cao} Y.U. Cao, A.S. Fukunaga and A.B. Kahng, "Cooperative Mobile Robotics:Antecedents and Directions," {\em Autonomous Robots}, 4, (1997) pp.7-27.
900: \bibitem{Hayes} A.T. Hayes, A. Martinoli, and R.M. Goodman, "Distributed Odor Source Localization", {\em IEEE Sensors}, Vol. 2, No. 3 (2002) pp.260-271.
901: \bibitem{DARS7} T. Kazama, K. Sugawara, and T. Watanabe, "Collecting Behavior of Interacting Robots with Virtual Pheromone," {\em Proc. 7th Int. Symp. on
902: Distributed Autonomous Robotic Systems}, (2004) pp.331-340.
903: \bibitem{Wil}
904: B.H\"{o}lldobler B.and Wilson E.O.
905: {\it The Ants}
906: {Belknap, Cambrige}
907: (1990)
908:
909:
910: \bibitem{De1}
911: {Deneubourg J.L.and Goss S.and Franks N. and Pasteels J.M.}
912: REVIEW{J. Insect Behavior} \textbf{2}(1989) 719.
913:
914: \bibitem{Sug1}
915: {N.Shimoyama, K.Sugawara, T.Mizuguchi, Y.Hayakawa, and M.Sano}
916: {Phys. Rev. Lett.} \textbf{76} (1996) 3870.
917:
918:
919: \bibitem{Hel}
920: {D.Helbing and F.Schweitzer and P.Moln\'{a}r}
921: {Phys. Rev. E} \textbf{56}{1997} 2527.
922:
923: \bibitem{Schw1}
924: {F.Schweitzer, K.Lao and F.Family}
925: {BioSystems} \textbf{41} (1997) 153.
926:
927: \bibitem{Bo}
928: {E.Bonabeau, M.Dorigo and G.Theraulaz}
929: {Swarm Intelligence}
930: {Oxford University press, Oxford}
931: {1999}.
932:
933:
934: \bibitem{Nishinari}
935: {K.Nishinari, D.chowdhury and A.Schadschneider}
936: {Phys. Rev. E} \textbf{67} (2003) 36120.
937:
938:
939: \bibitem{Tao}
940: {T.Tao, H.Nakagawa, M.Yamasaki and H.Nishimori}
941: {J. Phys. Soc. Jpn.} \textbf{73} (2004) to be pubslished.
942:
943: \bibitem{note1}
944: {Note that, in our simulation, the tranition value of $M(=M_c)$ from the V-shaped trail to the Y-shped trail is
945: not just the value $M_{C1}$ over which the perfect foraging using the V-shaped trail gets impossible
946: but lower than that.
947: It dose not mean, however, the abortion of optimized foraging at $M_c < M < M_{c1}$.
948: According to our theoretical analysis in ref\cite{Tao}., if $M<M_{c1}$
949: the foraging efficiency is equivalent between the foraging with the Y-shaped trail and that with V-shaped trail
950: as shown in Fig.14.}
951: \end{thebibliography}
952:
953: \end{document}
954:
955: