题名 | 对环境适应性的实验研究 |
作者 | 赵凯 |
学位类别 | 工学博士 |
答辩日期 | 1999-06-01 |
授予单位 | 中国科学院自动化研究所 |
授予地点 | 中国科学院自动化研究所 |
导师 | 王珏 |
关键词 | 适应性 环境 直接表示 Agent adaptation environment direct representation agent |
学位专业 | 模式识别与智能系统 |
中文摘要 | 传统AI假设世界是封闭的,以推理和搜索为核心,对适应性的研究有所忽 略,直到Holland提出了适应性框架和遗传算法。适应性的研究假设环境是黑 箱,Agent通过与环境的交互获得信息,以解决问题或进行学习。 在本文中,将环境分为三类:自然环境、社会环境和人环境。并在这种分 类的基础上对不同环境以及它们的组合做了一个实验系列,观察了实验现象, 讨论了对复杂性不同的环境适应的特点。强调了fitness function就是环境的 观点,并将实验建立在这个观点之上。 对环境的表示强调直接表示,而不是建立数学模型。Agent所见到的就是它 的环境,而不是将环境抽象为一个状态。 提出了染色体-蛋白质表示模型,它是一个基于Geneic Programmjng的 通用的规划方法。我们将其应用于排课表、机器人路径规划等问题。在单树的 基础上发展了多树表示,建立了多树操作,探讨了树表示的顺序性,并对实验 中出现的短基因现象从Schema原理的角度作了解释。 建立了染色体-蛋白质模型之后,以其为方法,以机器人路径规划问题为 例,对自然环境、社会环境和人环境下的适应分别作了探讨。 在对自然环境的适应中,对环境已知、未知,机器人质点、刚体等四种情 况做了实验,验证了Agent对已知和未知两种自然环境的适应能力。 在对社会环境的适应中,对纯社会环境、社会环境和自然坏境,Agent合作、 竞争等四种情况做了实验,观察到了机器人自由合作、编队合作,竞争时自动 退让、等待等解决资源冲突的现象,验证了在未知和变化的社会环境和自然环 境中,适应性方法可以完成合作和竞争问题。 在对人环境的适应中,首先讨论了对人适应的特点和方法,其次在自然、 社会、人三种环境同时存在的情况下,对Agent同时适应三者做了实验。各实 验中,自然环境和社会环境一致,但根据Agent一方、双方适应人,人的情感 一致、不一致分为四种情况,验证了Agent对人的适应是复制人的情感,形成 自我个性。 |
英文摘要 | In traditional AI, the world is assumed to be closed. Therefore, reasoning and search are taken as Leading technique And research of adaptation to the environment is greatly ignored, until John Holland proposed his adaptive plan and Genetic Algorithms in 1975. In an adaptive problem, the total environment ii considered as a black box. An agent tries to solve a problem or finishes learning through interaction with environment. In this dissertation, the environment is divided into three classes: natural environment, social environment and human. Under this classification, we take a series of experiments in different types of environment, observe the phenomena, and discuss the characters of adaptation to specific environment. We emphasize the point of view that fitness function is just environment, and establish our experiments on it. The direct representation method is used to represent environment, but not the mathematical model. The environment of an agent is just what it "sees". It is unnecessary to transform a specific environment to a state, which is usually used. The Chromosome-Protein Scheme(CP) is proposed as a general model of planning. It is based on Genetic Programming. The CP scheme is used to solve time table and robot path planning problem. Based on single-tree representation, multi-tree form of CP is proposed. And then the new genetic operators of multi-tree are established. In addition, the order character of hierarchical representation is studied. The short-gene priority phenomenon is analyzed according to the Schema theory, After the establishment of Chromosome-Protein Scheme, we study the cases of adaptation to natural, social environment and human, and take robot path planning as example In the adaptation to social environment, four experiments are designed. They are considered in different conditions, including known or unknown enyironment, point or rigid body of robot. The results of experiments show the ability of agents for adaptation to known and unknown environment. In the adaptation to social environment, four experiments are designed. They are considered in different conditions, including pure social environment or social and natural environment, cooperation or competition. Some interesting result:; are observed, such as random cooperation, team cooperation, conceding or waiting in resource competition. The results of experiments show the ability of agents for cooperation and competition under unknown and changing environment. About the adaptation to human, the special characters and methods are firstly discussed. Then several experiments are taken when three kinds of environment is considered together. Four experiments are designed in different conditions, including one or two agent adapting to human, different or same emotion of human. The results show that agent's adaptation to human is the process of copying human emotion and c |
语种 | 中文 |
其他标识符 | 540 |
内容类型 | 学位论文 |
源URL | [http://ir.ia.ac.cn/handle/173211/5695] ![]() |
专题 | 毕业生_博士学位论文 |
推荐引用方式 GB/T 7714 | 赵凯. 对环境适应性的实验研究[D]. 中国科学院自动化研究所. 中国科学院自动化研究所. 1999. |
个性服务 |
查看访问统计 |
相关权益政策 |
暂无数据 |
收藏/分享 |
除非特别说明,本系统中所有内容都受版权保护,并保留所有权利。
修改评论