【作者】 李艳萍；

【导师】 冯林；

【作者基本信息】 大连理工大学 ， 控制理论与控制工程， 2010， 硕士

【摘要】 移动机器人是一种集环境感知、动态决策与规划、行为控制与执行等多种功能于一体的综合系统,由于其较高的自主性、智能性和对外界环境的自适应性,使得移动机器人在各行业有一个广阔的应用前景,并成为机器人领域的一个重要分支。近几年来,移动机器人已广泛应用于星球探测、军事侦察、医疗服务、危险及恶劣环境作业、农业生产等方面,本文中重点研究移动机器人在农业中的应用。自主学习的农业机器人要求所有机器人都是全自主的,即传感器信息获取、任务规划、运动控制都是由机器人自身完成。对于自主学习的机器人,需要一种机器学习策略的支持,进行信息获取、任务规划和运动控制。机器学习研究的一个主要问题就是针对一个学习任务,通过选取一组具有代表性的特征属性,构建一个模型。而机器学习最初的研究动机就是为了使计算机系统具有人的学习能力从而实现人工智能。之后我们又提出了自主式机器学习方法(即自主学习),摆脱了在知识获取过程中对领域专家知识依赖的约束。20世纪90年代初期,随着统计学习理论的逐步完善,基于此理论的算法——支持向量机也随之出现。支持向量机现在已成为一种倍受关注的机器学习。由于SVM训练数据样本时,需要消耗大量的时间,而为了加快训练的速度,本文提出了一种新的自主学习策略——基于聚类分析的增量式支持向量机的学习策略。结合时间权重和数据特征分布对数据进行分析,使算法适合于处理随时间变化的大量数据。特别的,我们通过对比实验验证了这种算法的有效性和实用性。由于加工偏差及机器人调试中后期结构的调整,势必造成质心相对几何中心的偏移。常规做法是在机器人制作完成后施加配重微调质心,但配重的加入会带来更多的能耗。为了解决该问题,本文提出了基于质心偏移时对机器人的运动控制方法,通过建立四轮机器人的控制模型,分析了质心偏移对机器人运动状态的产生的影响,并采用线性规划的方法对机器人各轮电机的出力分布进行了优化,得到质心偏移时在保证合力最大情况下电机的最省能耗。实验结果验证了该算法的有效性和正确性。更多还原

【Abstract】 Mobile robot is a integrated system that includes several functions. Because of its high autonomy, intelligence and adaptability to the external environment, the mobile robot has a broad application prospects in various industries, and becomes an important branch in the field of robotics. In recent years, mobile robots have been widely used in planetary exploration, military reconnaissance, medical services, hazardous and harsh environment operations, agriculture, etc. This paper focuses on the mobile robot applications in agriculture.Autonomous learning requires that all agricultural robots are fully autonomous, that is to say, sensor information acquisition, mission planning, and motion control are done by the robot itself. In recent years, a method named Learning from Demonstration used in agriculture robot has been gradually developed and become the research point in artificial intelligence and machine learning areas。In the early 90s, Support Vector Machine presents with the gradual improvement of statistical learning theory, based on this theory algorithm. Support Vector Machine now has become a closely watched classification technology.When SVM trains data samples, it needs to consume a large amount of time. So to accelerate the training speed, we propose a new learning strategies—the learning strategy based on clustering of increamental svm. We analyze datas by combinating of time weight and data features to make the algorithm suitable for processing large amounts of datas. In particular, we prove the effectiveness and practical of this algorithm with the experiments.Mobile Robot often has an eccentric centroid because of the limited precision in producing causes. That results in difference of robot’s final position between expected one and actual one since most control approaches are based on an ideal model whose centroid is not eccentric, thus extra energy is needed for robots to adjust that difference. To solve this problem, we build a model with eccentric centroid to analyze the distribution of traction on each wheel in this condition and get an optimization solution set of this traction with linear programming algorithm. Specifically, we conduct simulation experiments to demonstrate the algorithm is of positive effect in reducing cost of extra energy.更多还原