分布式人工智能与机器学习在多智能体系统中的应用探索

人工智能

需积分: 5 6 浏览量更新于2024-06-28 收藏 5.19MB PDF 举报

身份认证购VIP最低享 7 折!

领优惠券(最高得80元）

"这篇论文是BRAC大学计算机科学与工程系的本科生毕业论文，主题为‘计算机生成的人工生命模型：猎人捕猎猎物的算法’，由Md. Mahbub-Ul Haque、Surid Imam Khan Sur、Rehnuma Binta Shahriar和Supti Rani Saha共同完成。论文于2021年提交，并在9月26日被接受，作为获得计算机科学学士学位的部分要求。指导教师包括Dr. M. Kaykobad（著名教授）和Tanvir Kaykobad（博士生），分别来自BRAC大学和皇后大学计算学院。" 在本文中，分布式人工智能与机器学习在多智能体环境中的应用是核心讨论点。分布式人工智能（Distributed Artificial Intelligence，DAI）是指在多个智能体之间共享和协调知识以实现共同目标的领域。这些智能体可以是独立的程序或者硬件设备，它们通过通信和合作来解决复杂问题。在多智能体系统中，每个智能体都有其独特的感知和行动能力，能够根据环境变化做出决策。机器学习（Machine Learning，ML）是人工智能的一个分支，它允许系统通过经验学习和改进，而无需显式编程。在多智能体环境中，机器学习可以用于训练智能体以适应不断变化的环境，优化决策过程，并提高整体性能。例如，强化学习（Reinforcement Learning, RL）是一种常用的机器学习方法，它通过奖励和惩罚机制让智能体学习最佳策略。论文中的“猎人捕猎猎物的算法”可能涉及模拟现实世界中的捕食者-猎物动态，利用分布式AI和机器学习技术。这种模拟可以帮助我们理解智能体如何在复杂环境中协作、竞争和生存。智能体可能通过学习和适应来优化其狩猎策略，同时考虑其他猎人和猎物的行为。此外，论文可能还探讨了如何设计有效的通信协议，使智能体能有效地交换信息，以及如何处理不确定性和局部信息的问题。这些挑战在多智能体系统中尤为显著，因为每个智能体只能获取有限的环境信息，并需要在不完整的信息下做出决策。这篇论文深入研究了在分布式人工智能框架下，如何利用机器学习算法让多智能体系统在捕食者-猎物的动态中自我学习和优化，为理解和应用这些技术在实际问题中提供了理论基础和实验平台。

资源详情

资源推荐

ten by Andrew Vande Moere* [8], which is based on self-organization and behavior

simulation principles that can represent dynamic data evolution by extending the

concept of information ﬂocking. Time-varying data-sets contain data objects that

are altered in time due to continuously executed, time-dependent data updates.

Static State Replacement, Time-Series Plots, Static State Morphing, Control Appli-

cations, and Equilibrium Attainment are used to visualize time-varying data-sets.

This paper has described the algorithms and principles that drive the information

ﬂocking method and how they can be applied to a real-world and almost chaotic

data-set.Then Kapi, Sunar, and Zamri’s paper ‘A Review on Informed Search Algo-

rithms for Video Games Pathﬁnding’ [23] is based on Pathﬁnding algorithms that

have received increasing attention in many applications such as video games and

robotics, metabolic pathways in the past few years. The ideal algorithm A* can

speed by enhancing it from various perspectives as it reduces memory consumption

and no post-processing is needed. By using swarm intelligence in the pathﬁnding

technique, so many other ﬁelds are growing. Hence, the recent research progress in

the game’s navigation emphasizes optimization using four diﬀerent perspectives.The

paper ‘Particle Swarm Optimization: A Survey of Historical and Recent Develop-

ments with Hybridization Perspectives’ written by Sengupta, Basak, and Alan Peters

II [18] is based on Particle Swarm Optimization (PSO), which is the foundations and

frontiers of advances in PSO. Here, Fuzzy Adaptive (FA) is used to make fuzzy sets

and membership rules. However, The topology of the swarm of particles sets up a

degree of a network of its members.According to Artaxo et al., paper ‘Autonomous

Cooperative Flight Control for Airship Swarms’ [19] is a paper investigating the de-

sign and implementation of controllers for autonomous cooperative airships ﬂights

using two diﬀerent approaches. Two diﬀerent swarm strategies have been tested

based on the Reynolds Boids algorithm and the Robotic Particle Swarm Optimi-

sation algorithm paper.Then Hahn et al. wrote the paper ‘Foraging Swarms using

Multi-Agent Reinforcement Learning’ [22], where they evaluated the Boids forag-

ing with the help of multi-agent reinforcement learning (MARL). They also showed

that each predator moves freely without provoking their ﬂock-mates. Lastly, they

evaluated that each predator learns from the ﬂock-mates and can easily select its

targeted object. Using the idea of the Boids Algorithm, Mavhemwa and Nyangani

wrote the paper ‘Uniform spatial subdivision to improve Boids Algorithm in a gam-

ing environment’ [16] about creating a video game that simulates the real-life crowd

behavior. He also showed several approaches that can improve the ﬂocking simula-

tion. A robotic environment full of ﬁsh school and this ﬁsh school’s ﬁsh maintaining

the Boids algorithm is written by Connor et al. in their paper ‘ Analysis of Robotic

Fish Using Swarming Rules with Limited Sensory Input’ [17]. They also showed

the beneﬁts of using Virtual Reality that can make the ﬁsh modify itself and can

interact with the ﬂock-mates easily.‘Simulation of the Flocking Behavior of Birds

With the Boids algorithm’ written by Erneholm [12], which undergoes the ﬂocking

behavior. Moreover, along with the algorithm, he compares two diﬀerent neighbor-

hoods of Boids.Joselli et al.’s paper ‘A Flocking Boids Simulation and Optimization

Structure for Mobile Multicore Architectures’ [14] is the paper where they use the

approach of ﬂocking Boids Algorithm to high-end mobile multicore architectures.

They used the ﬂocking Boids algorithm simulation on the Android render script

API. For the development of our day-to-day life, people are converting their lives to

robotic systems. For this reason, Kasmarik et al. made the evolution of the Boids

Flocking algorithm and Swarm behavior in their paper ‘Autonomous Recognition

of Collective Behaviour in Robot Swarms’ [24]. In their paper, they used several

functions to recognize the ﬂocking behavior to simulate a robot. For our thesis

implementation we also used some ideas about convex hull algorithm. Jarvis ﬁrst

wrote the paper ‘On the Identiﬁcation of the Convex Hull of a Finite Set of Points

in the Plane’ [2] that gives the proper deﬁnition of the convex hull algorithm. The

algorithm shows that we can create a convex hull with some ﬁnite number of points

by three simple steps.But while calculating the convex hull of a simple polygon,

people were failing to use the algorithm. In 1981 Toussaint and Avis wrote the pa-

per, ‘On a convex hull algorithm for polygons and its application to triangulation

problems’ [3], where they showed how using the externally weak visible polygon can

easily make a convex hull. We can make internal and external triangulate of the

polygon at a particular time. Lastly, we used the simulation plateform ’Godot’ to

implement our ideas in a visual representation. For this reason, we took the help of

the book ’Introducing Godot: Why Migrate?’ written by Thorn and Alan [25]. All

this previous works inspired us and helped to determine our objectives of the thesis,

which we talked about in the next section.

1.3 Objectives of the Thesis

Our research is motivated to simulate a natural phenomenon between a herd of

deer and a pack of wolves. We also want to simulate the movements derived by

the ﬂocking algorithm and the eﬀects on ﬂocks by their surrendering ﬂockmates.

Moreover, another main objective of this thesis is to show the way of hunting and

communication between the wolves, which ultimately represents a situation of real-

life hunting and the outcome. We want to observe the reaction caused by the

artiﬁcial intelligence of both deer herds and wolves in some artiﬁcial situations. This

whole analysis leads us to the understanding of the decision-making capabilities of

artiﬁcially generated life models.

The objectives of this research can be divided into two main sections, imple-

mentation of prey behavior using Boid’s algorithm introduced by Reynolds [4] and

simulation of the predators’ behaviors.

Implementing Flocking Behavior of Prey

Our ﬁrst objective is to simulate the behavior of prey. Using Boids Algorithm

by Reynolds [4], we set the moving rules for each prey individually, which helps

to simulate the ﬂocking behavior. We also using fuzzy logic by Bajec, Mraz, and

Zimic [7] to get a more accurate and realistic simulation, implying the individual

prey’s observation and analytical capability within a range. Also, by adding an

escape behavior to prey, the whole movement of the prey group can be controlled.

Simulating Predator’s Characteristics

Our second objective is to simulate predators. The group of predators will be led by a

leader, the movement and diction of other predators in the group will be manipulated

by the leader’s decision. To make the formation of hunting and communicate with

each other, swarm intelligence can be implied. We will introduce simple rules like

detecting the target, chasing the target, isolating the target, and terminating the

target, making the simulation of the wolves hunting.

Our other objective is to analyze the outcome from the simulation and compare

it with real-life scenarios. These will lead us to make predictions in situations similar

to the simulations. Moreover, our research may interest other researchers in working

on more real-life scenarios and unleash new scope of this thesis like determining the

ﬁre execution path of a building compound, estimating the path of movement of a

mob, swarm drones, or robots, war fair simulations and, other natural phenomenons.

After we had discussed the objectives, we had to organize our thesis in a order.

Which helps the readers to follow the way, also helps for better understanding.

1.4 Organization of the Thesis

In Chapter 1, we introduce the ﬂock and ﬂocking behavior and a brief literature

survey. We discussed the objectives, scope, and organization of this thesis. In

Chapter 2, we have made a review of literature in the ﬁeld. We also elaborately cov-

ered more important papers discussing Boids ﬂocking algorithms, convex hulls, and

video simulations. In Chapter 3, we have discussed the idea of animal behavior and

some basic rules of simulating prey. We have also discussed the problems we have

faced in simulating behaviors of prey and predators. Chapter 4 contains details of

the algorithms and their implementation. We have simulated behaviors of multiple

intelligent predators led by a leader and multiple prey. We have used appropriate

ﬁgures and diagrams to increase comprehension of the process. Chapter 5 contains

the results that were expected before the experiment, which was guessed by analyz-

ing the theories. The result we got after the experiments is also in this chapter and

compared with the expected results. We concluded our thesis in Chapter 6. This

chapter contains the concepts of this thesis’s future work, which can motivate other

researchers to work further with this topic. In the end, a section of the bibliography

contains all the references given in the whole thesis.

剩余67页未读，继续阅读

承让@

粉丝: 6
资源: 380

会员权益专享

分布式人工智能与机器学习在多智能体系统中的应用探索

多智能体系统中的分布式强化学习研究现状

分布式机器学习平台与算法综述.pdf

分布式机器学习系统 工程与实战.pptx

多智能体分布式强化学习

介绍一下分布式人工智能及应用

多智能体强化学习与分布式计算的联系

智能体分布式优化问题

人工智能在多智能体系统中的应用

怎么样实现分布式人工智能

集中式事件触发多智能体与分布式事件触发多智能体的对比

基于分布式DQN多智能体强化学习的算法流程

disc课程:多智能体分布式控制讲义

分布式智能计算系统的国内研究现状

多智能体如何做到完全分布式

一个docker集群部署了机器学习服务算是分布式机器学习平台吗

基于事件触发机制多智能体分布式算法

ubuntu虚拟机搭建分布式机器学习平台

基于docker分布式机器学习

多智能体系统分布式协调控制

多智能体分布式编队协同控制

会员权益专享

最新资源

分布式机器学习系统工程与实战.pptx