【作者】 钟秋波；

【导师】 潘启树； 洪炳镕；

【作者基本信息】 哈尔滨工业大学 ， 计算机应用技术， 2011， 博士

【摘要】 随着科技的不断进步,各种类型的机器人相继的研发成功,类人机器人由于其外形和动作与人类似,与其它类型的机器人相比,更容易被人类所接受。而类人机器人的运动规划是类人机器人研究的基本问题,也是非常重要的问题之一。类人机器人要在人类的复杂环境中从事各种任务,需要获取不同的环境信息,进行各种不同的复杂运动,从而能适应人类的生活环境,更好的为人类服务。本文在国家863计划重点项目的子项目“竞技与娱乐多机器人系统”课题的资助下,以提高类人机器人在竞技娱乐活动中的娱乐观赏性和竞技能力为目的,对类人机器人运动规划的若干关键技术进行系统深入的研究。主要包括稳定性控制、倒地动作最优控制、复杂运动规划和类人型足球竞技实验平台等方面的内容:第一,建立了类人机器人7连杆运动学和动力学一般模型,并根据机器人不同类型的运动分别建立了前后倒地动作和上下楼梯、上下斜坡运动模型。对于类人机器人来说,其稳定性是首先需要解决的问题。一些研究人员已经对类人机器人的稳定性控制研究提出了各种不同的稳定判定标准和控制方法,本文在比较各种稳定判断准则下,通过引入一种二阶锥控制方法,对类人机器人运动稳定性进行控制,获得了较好的效果。第二,在类人机器人的运动过程中,各种不确定因数可能导致其发生摔倒。当摔倒不可避免的情况下,本文研究了类人机器人倒地动作的最优控制问题,通过对倒地动作深入分析,建立了倒地运动学方程,并引入参数优化和强化技术对倒地动作进行最优控制。针对参数优化方法在求解二次规划问题(SQP)中的不足,使用一种基于改进的二次规划滤子算法对该方法的寻优过程进行了改进,并和极小值原理控制方法进行了比较。第三,由于类人机器人外形类人,除了能在平地上行走外,还需要进行楼梯和斜坡等各种环境地面上的复杂运动,以便能够真正适应人类的生活环境。研究人员已经对类人机器人的上下楼梯和斜坡进行了各种研究,但是由于类人机器人是一个多自由度的非线性复杂系统,单独使用传统控制方法和一些智能控制算法并不能很好的对其进行复杂运动的控制。本文分别使用神经网络和模糊控制对类人机器人的上下楼梯运动进行离线学习训练,通过嵌入式视觉系统采集环境信息进行在线控制,同时为了解决训练过程的耗时长、结果难收敛的缺点,在基本微粒群优化算法的基础上,使用一种改进的微粒群优化算法,分别对神经网络的权值和模糊控制器的规则进行优化,实验表明,该方法可以有效的减少训练时间并获得类人机器人上楼梯运动中较高的稳定控制效果。第四,针对类人型足球机器人的比赛特点,本文对3个对抗3个类人型机器人(3vs3)比赛系统进行研究,采用主从控制系统框架对机器人进行任务级的控制。根据比赛过程中所需的基本动作,设计了基于嵌入式视觉的在线快速曲线行走运动,并对点球子系统的射门与守门进行深入研究,类人机器人通过嵌入式视觉实时采集比赛环境信息,通过有限状态机的方式进行决策。比赛显示该系统能够获得较好的控制效果。更多还原

【Abstract】 With the development of science, kinds of robots have been developed sucessfully. Compared with other type of robot, humanoid robot is easier to accepted by people because its shape and actions are similar with human. Motion planning for humanoid robot is the basic research and also one of the best important problems in the area of humanoid robot. If humanoid robot wants to do kinds of works in the complex environment of human, it need to acquire the different environmental information to do kinds of complex motions, so as to adapt to the human environment and better service to humanity.This thesis is funded by the national 863 key project "sports and entertainment multi-robot systems", and with the purpose of improving the entertainment appreciation and competitive ability of humanoid robot in the competitive entertainment activities, some key technologies of motion planning for humanoid robot are on further research, which includes the stability control, optimal control of falling action, complex motion planning, motion planning in the complex environment and experimental platform of 3 vs 3 humanoid robot soccer game.Firstly, a seven link model of kinematics and dynamics for humanoid robot is built and according to the different motions of robot, models of falling forward and backward actions, climbing up and down the stairs and slopes are present. For humanoid robot, the first problem required to solve is the stability. Some researchers have present kinds of different stability criterions and control methods for the stable control of humanoid robot. This theses introduced a control method based on second-order cone to control the stability of robot and good results are obtained.Secondly, during the process of motion for humanoid robot, it may fall due to some uncertainty factors. Since the falling is inevitable, this theses has done research on the optimal control of falling for humanoid robot. By deeply anlysis of falling action, the kinematics equation of falling is built and optimal control for falling action based on parameter optimum and enhance technology is proposed. For the lack of solving SQP(sequence of quadratic programming) in the parameter optimum method, a new method based on improved SQP filter algorithm is present to improve the optimization process. Good optimization results are obtained by comparing with control of the minimum principleThirdly, since the humanoid robot’s shape is similar to people, it is not only walking on the flat but also do some complex motions in environment such as stairs and slopes to adapt the people’s living environment really. Researchers has done some researches on the motions for humanoid robot on stairs and slopes. However, humanoid robot is regarded as a multi-degrees and nonlinear complex system, traditional control methods and some intelligent algorithms are not good enough to control its complex motion. In this theses, two methods of nerual network and fuzzy logic control are present to learn offline and control the climbing stairs motion of robot online respectively. To solve the shortcoming of difficult to converge and long time consuming during the trainning process, an improved PSO(Particle Swarm Optimization) optimal algorithm is proposed to train the weights of nerual network and rules of fuzzy logic controller. Experiments show this methods can effectively reduce training time of stairs motion for humanoid robot and get the higher stability.Forthly, for the features of humanoid robot soccer game, this theses present a game control system for 3 vs 3. Master-slave control system is used to control robot based on the task-level. For the basic motions required in the game, fast curve walking based on embedded vision system is designed. Deep research on penalty kick system include shooting and goalkeeping are achieved. The information from environment are gathered by the embedded vision system online, decisions are made through the finite state machine. Experiments show the system can get achieve good control effect.更多还原