演化群机器人：智能计算与机器人技术的融合

群智能机器人

需积分: 12 177 浏览量更新于2024-07-19 1 收藏 8.16MB PDF 举报

身份认证购VIP最低享 7 折!

30元优惠券

"Evolutionary Swarm Robotics (演化群机器人)" 是一本深入探讨智能计算与机器人技术的专著，尤其关注群智能机器人领域。作者Vito Trianni通过本书介绍了如何利用进化算法来发展和优化自主机器人的群体行为。正文: 演化群机器人是近年来机器人学和人工智能领域的一个热点研究方向，它结合了生物学中的群集行为理论、进化计算以及控制理论等多学科知识。这本书"Evolutionary Swarm Robotics" 提供了一个全面的视角，详细阐述了如何利用演化计算方法来设计和优化自主机器人的群体行为，以实现更高效、更复杂的任务执行能力。首先，演化计算是模拟生物进化过程的一种计算模型，包括遗传算法、遗传编程、粒子群优化等。这些算法在解决复杂优化问题时表现出强大的能力，尤其适用于群机器人系统中个体间相互作用和集体行为的学习和优化。群智能机器人系统由多个自主机器人组成，它们通过简单的规则和局部信息交换来协调行动，模仿自然界中如蚂蚁、鸟群等生物群体的行为。在本书中，作者可能详细讨论了如何定义这些简单规则，以及如何通过演化过程来逐步改进和适应环境变化。书中可能会涵盖以下关键知识点： 1. **群体行为模型**：介绍并分析各种基于生物启发的群体行为模型，如Boids模型（Birds, Ostriches, and Doves）和自组织临界性理论。 2. **演化策略**：讲解如何运用遗传算法或粒子群优化等演化策略来优化机器人群体的行为参数，以提高整体系统的性能。 3. **通信机制**：探讨机器人之间的通信方式，包括距离感知、信息传递和局部规则，以及如何设计有效的通信协议以促进群体协作。 4. **适应性和鲁棒性**：研究机器人如何适应环境变化和应对不确定性，以及如何通过演化提高系统的整体鲁棒性。 5. **实验与应用**：展示实际的实验结果和案例研究，可能包括模拟实验和真实世界部署，例如搜索与救援、环境监测、构建和维护结构等任务。 6. **未来挑战与前景**：讨论群机器人领域的当前挑战，如大规模群体的管理、能源效率、安全性问题，以及可能的技术发展方向。 "Evolutionary Swarm Robotics" 是对智能计算与机器人技术研究者的重要参考，它不仅提供了理论基础，还可能包含丰富的实践经验和实验数据，有助于读者深入理解和掌握群机器人系统的演化设计方法。

资源详情

资源推荐

4 1 Introduction

However, designing a self-organising behaviour for a group of simulated

and/or real robots is not a trivial task. The classic approach to this design

problem consists in two decomposition phases: (i) the behaviour of the system

should be described as the result of interactions among individual behaviours,

and (ii) the individual behaviours must be encoded into controllers. Both

phases are complex because they attempt to decompose a process (the global

behaviour or the individual one) that is a result of dynamical interactions

among its sub-components (interactions among individuals or between indi-

vidual actions and the environment). These dynamic aspects are in general

diﬃcult to be predicted by the observer. In such a context, we believe that Evo-

lutionary Robotics (ER) should be the methodology to be exploited (Harvey

et al., 1993; Nolﬁ and Floreano, 2000; Harvey et al., 2005). ER bypasses the

problem of decomposition at both the levels of ﬁnding the mechanisms that

lead to the emergent global behaviour, and of implementing those mechanisms

into a suitable controller. In fact, ER relies on the evaluation of the system

as a whole, that is, on the emergence of the desired global behaviour starting

from the deﬁnition of the individual controllers. Moreover, ER can exploit

the richness of possible solutions oﬀered by the dynamic robot-environment

interactions, which may not be a priori evident to the experimenter (Nolﬁ

and Floreano, 2000; Dorigo et al., 2004).

In this book, we propose the use of ER techniques for the design of

self-organising group behaviours, for both simulated and real robots. In this

respect, the contribution brought forth by our research is twofold. From an en-

gineering perspective, we propose an automatic methodology for synthesising

complex behaviours in a robotic system. We believe that evolutionary tech-

niques should be used in order to obtain robust and eﬃcient group behaviours

based on self-organisation. From a more theoretical point of view, we show

that simple sensory-motor mechanisms are at the base of complex cognitive

phenomena, both at the individual and at the collective level. We therefore

propose the study of adaptive, intelligent behaviours as the result of evolu-

tionary selective pressures. In summary, our work tries to mediate between

two apparently opposed perspectives: engineering and cognitive science. The

former seeks for the optimal design, that can display the best performance at

the lowest cost. The latter seeks for tangible models that can explain intelli-

gence. Both these perspectives inform the research described in this following

chapters.

As mentioned above, the object of our research is a swarm robotic system

composed of a number of autonomous mobile robots—referred to as s-bots

(see Figure 1.1a and b)—which have the ability to connect to each other

forming a physical structure—referred to as a swarm-bot (see Figure 1.1cand

Mondada et al., 2004, for further details). Exploiting the cooperation among

its components, the swarm-bot can solve problems the single s-bots are not

able to cope with. In the swarm-bot form, the s-bots are attached to each

other, therefore forming a single robotic system that can move and reconﬁgure.

For example, the swarm-bot might have to take diﬀerent shapes in order to

1 Introduction 5

(a) (b) (c)

Fig. 1.1. (a, b) The real s-bot. The traction system is composed by both tracks

and wheels. On top of it, a rotating turret is mounted, which holds many sensory

systems and the rigid gripper for physical connections. (c) A swarm-bot moving in

a outdoor environment.

go through a narrow passage or overcome an obstacle. Physical connections

between s-bots are essential for solving many collective tasks. Additionally, the

physical interactions can be exploited for self-organisation of the swarm-bot.

However, for tasks such as searching for a goal location, or tracing an optimal

path to a goal, a swarm of unconnected s-bots may be more eﬃcient.

The swarm-bot is a very innovative robotic artifact, developed by a consor-

tium of European universities and AI labs, which participated to the SWARM-

BOTS project.

From the mechatronics point of view, the s-bots are considered

to be among the most complex robots in their size to date. The research pre-

sented in this book summarises some of the results obrained in the attempt

to design suitable control systems capable of exploiting the features of these

robots and of displaying coordination and cooperation abilities.

Besides the aspects related to the particular technology used for our re-

search, we performed much experimental work in the attempt to to assess arti-

ﬁcial evolution as a viable methodology for the development of swarm robotics

controllers. The use of evolutionary techniques for the synthesis of group be-

haviours has been relatively modest up to now, as shown in Section 4.3,and

even less work has been done about the synthesis of self-organising group

behaviours through artiﬁcial evolution. In this book, we collected various ex-

periments aimed at the synthesis of self-organising behaviours, and we show

how evolutionary robotics can be exploited in this respect. We also show how

through the anlysis of the self-organisng behaviour evolved in an artiﬁcial sys-

tem, it is possible to shed light on the basic principles underlying collective

intelligence.

The SWARM-BOTS project was funded by the Future and Emerging Tech-

nologies programme (IST-FET) of the European Commission, under grant IST-

2000-31010. It started in October 2001 and ended in March 2005. See also

http://www.swarm-bots.org.

Embodied Cognitive Science

The study of self-organising behaviours for a multi-robot system can be pur-

sued following two diﬀerent approaches. On the one hand, one can be moti-

vated by an engineering perspective, aiming at the design of robust, eﬃcient

and adaptive controllers for a complex robotic system. On the other hand, one

can try to synthesise an aritiﬁcial system in order to uncover the rules govern-

ing collective intelligence. In the former case, the development of a working

artifact is the ultimate goal. In the latter case, a working artifact is used

as a descriptive model to study cognitive processes. These two approaches

only seemingly contrast: they are the two sides of the same coin—a synthetic

approach to the study of Embodied Cognitive Science (see also Pfeifer and

Scheier, 1999). This chapter is devoted to a brief history of this research ﬁeld,

which underpins and motivates the studies presented in this book.

2.1 Back to the Origins: Artiﬁcial Intelligence

and Cybernetics

The question about “what is intelligence” has been tackled in many diﬀerent

ways, and various deﬁnitions have been proposed trying to account for com-

mon sense notions and scientiﬁc observations (see Pfeifer and Scheier, 1999,

p. 6). The debate confronted the ideas of many philosophers and psycholo-

gists, and consensus was hard to ﬁnd. At the beginning of the XX Century,

the technological advancements and the advent of computers enriched this

debate with the following question: can a machine be intelligent?

Mainly two disciplines tried to give an answer: Artiﬁcial Intelligence (AI)

and Cybernetics. The former originates from the work of Turing (1936, 1937),

which represented a milestone for the theory of computation and computabil-

ity and gave a formal deﬁnition of the universal computing machine. This

work was followed by the development of the ﬁrst general-purpose computers

during the early 1940’s, under the pressure of the Second World War and with

Vito Trianni: Evolutionary Swarm Robotics, Studies in Computational Intelligence

(SCI) 108,7–21 (2008)

www.springerlink.com

 Springer-Verlag Berlin Heidelberg 2008

8 2 Embodied Cognitive Science

the contribution of many scientist, among which Von Neumann (1945).

The

scientiﬁc community was impressed by the problem-solving ability of these

computing machines and by the envisioned potentials. Soon, rather than be-

ing a simple tool, the computer started to be considered as an artiﬁcial brain

that could mimic human reasoning.

In the same period, analog machines were built, that could demonstrate

some form of “intelligent behaviour”. In 1912, the military research in the

USA celebrated the electric dog, built by John Hammond Jr. and Benjamin

Miessner (see Figure 2.1a). This machine—sadly the precursor of what nowa-

days is called, using an oxymoron, intelligent weapon—was able to direct itself

and move toward a light source. Some years later, Grey Walter (1950, 1951,

1953) demonstrated how Elmer and Elsie, two electric tortoises,

could display

complex behaviours, as if they were “alive” (for more details, see Figure 2.1b

and Holland, 2003):

Not in looks, but in action, the model must resemble an animal.

Therefore, it must have these or some measure of these attributes:

exploration, curiosity, free-will in the sense of unpredictability, goal-

seeking, self-regulation, avoidance of dilemmas, foresight, memory,

learning, forgetting, association of ideas, form recognition, and the

elements of social accommodation. Such is life.

Grey Walter, 1953, pp. 120-121

[Elsie] lingers before a mirror, ﬂickering, twittering and jigging like

a clumsy Narcissus. The behavior of a creature thus engaged with its

own reﬂection is quite speciﬁc, and on a purely empirical basis, if it

were observed in an animal, might be accepted as evidence of some

degree of self-awareness.

Grey Walter, 1953, pp. 128-129

Grey Walter’s tortoises were built following the principles of Cybernetics.

This discipline aimed at the understanding of the basic principles that under-

pin the behaviour of “the animal and the machine”, as it was indicated in the

subtitle of Wiener’s book (Wiener, 1948, 1961). Cybernetics was mainly driven

by control theory and statistical information theory. Animals were modelled

and studied as if they were machines, in order to uncover the basic principles

of their behaviour. In doing so, the environment was taken into account as

a fundamental part of the system under observation. The environment was

considered as a source of energy and disturbances ﬂowing through the ob-

served system, which reacts trying to maintain its internal equilibrium. This

led to the introduction of the concept of homeostasis, the ability to maintain

Von Neumann gave his name to the architecture of most of the non-parallel-

processing computers. However, it has to be recognised that many years before the

work of Von Neumann, implementations of general-purpose computing machines

already existed.

The only surviving tortoise was repaired by Dr. Owen Holland and is conserved

at the Burden Neurological Institute in Bristol.

2.2 Artiﬁcial Intelligence: From Dawn to Dusk 9

(a) (b)

Fig. 2.1. The origin of cybernetics. (a) The electric dog, built in 1912 by John

Hammond Jr. and Benjamin Miessner. This machine could orient itself and move

toward a light source. (b) The electric tortoise. This machine was provided with light

sensors to perceive the ambient light, and a bumper to detect collisions. It reacted

to the perceived light approaching it or moving away, depending on the perceived

intensity. In this way, it was able to show complex behaviours and learning abilities,

thus mimicking a living organism (

 UWE Bristol).

“internal order”—i.e., certain parameters within a given range of values—

notwithstanding the external inﬂuence from the environment (Ashby, 1956,

1960). Adaptive behaviour and learning were studied under this perspective,

and physical instantiations of cybernetics theories took the form of various

electro-mechanical homeostatic machines, including Grey Walter’s tortoises.

Unfortunately, cyberneticists missed to participate to the digital revolution of

the time, so that soon their lucky star faded away, dimmed by the raising sun

of Artiﬁcial Intelligence.

2.2 Artiﬁcial Intelligence: From Dawn to Dusk

While Grey Walter was busy cabling his tortoises, the AI community was

rapidly growing. The year 1956 was a key point for this discipline. Two im-

portant meetings were organised. The Symposium of Information Theory took

place at the MIT in Boston. During this meeting, Newell and Simon (1956)

presented a program that could demonstrate theorems in logic. The second

meeting was the so called Dartmouth Conference, a six-week meeting that

brought together the most prominent researchers in the ﬁeld. The meeting

led to the establishment of the “brain-computer metaphor”: the processes of

the mind were considered completely logical, so that they could have been

simulated by a well-deﬁned program. From this point of view, the mental

processes are assimilated to a program, to be executed by a computer, which

in turn takes the place of the brain. On the extreme of this line, the Func-

tionalism stated that what matters are only the programs—i.e., the mental

剩余191页未读，继续阅读

brbrm

粉丝: 2
资源: 5

演化群机器人：智能计算与机器人技术的融合

Multi-Robot Systems Swarm Robotic

liziqun.rar_ea_evolutionary_kennedy_swarm_粒子群算法

多重自适应协同进化粒子群算法（ACEPSO

neutral evolutionary计算

3篇有关粒子群算法的文献

除了粒子群算法还有什么算法可以优化BP神经网络提高预测精度的文献有哪些

粒子群算法相对其他算法的优点和对应文献

与粒子群算法有关的文献

composite differential evolution for constrained evolutionary optimization.

三方演化博弈的matlab代码举例

Evolutionary Programming算法流程

matlab实现多目标粒子群

复杂网络演化博弈代码

粒子群算法优化支持向量机英文

R语言 实现两方演化博弈

改进粒子群算法MOPSO

粒子群多目标轨迹规划python

Multi-objective Evolutionary Federated Learning

渐进交付的流程 Evolutionary Delivery与最强最美产品 Maximal Beautiful Product(MBP)的特点？

evolutionary algorithms

最新资源

R语言实现两方演化博弈