蛇形机器人的损伤恢复

发布时间：2018-03-25 03:00

本文选题：蛇形机器人　切入点：损伤恢复　出处：《中山大学》2017年硕士论文

【摘要】：蛇形机器人由于其多自由度带来的运动灵活性开始被人们研制利用。几十年来,已经有很多蛇形机器人样机研制出来,甚至被应用到实际应用中。它们的机体结构多种多样,实现的运动形式也各有不同。并且有丰富的控制方法被提出及使用。但是对于蛇形机器人有一个问题是需要面对的,而目前人们对这个问题还没有太多关注,就是当蛇形机器人离开调试环境在复杂的自然环境中做自主运动时会不可避免的损坏。当蛇形机器人损伤时,现有的做法是要么停止当前作业由科研人员维修调试,要么根据自身预设的损伤模型进行检测并以预设的应对措施应对。而预设的损伤检测模型是有限的,人为的预想不可能覆盖所有不可预知的损伤情况。本文引入一种较新的智能试错学习算法应用在正交结构的蛇形机器人上,可以在不预设损伤模型的情况下,以不断的试错-学习过程指导蛇形机器人寻找更优的补偿行为。此算法目前被应用在六足机器人及机械臂上,还未被应用在蛇形机器人上。本文第一次将此算法应用在蛇形机器人上并在仿真工具上做多种方式的验证。首先,本文阐述了蛇形机器人自主损伤恢复的意义,并对当前蛇形机器人的研究现状及机器人的损伤恢复方面的研究做了介绍。其次,本文对智能试错学习这一框架算法做了展开介绍,包括两个重要的步骤—行为-性能映射表的建立、损伤恢复的适应过程。并讨论了蛇形机器人采用的一般控制方式蜿蜒曲线。同时介绍了本文要使用的集成有物理引擎的仿真工具。然后,我们对使用智能试错学习算法所需要的工作做了归纳总结。并对现有蛇形机器人平台做了必要的总结,包括正交结构、传感器系统等。然后,在归纳总结的框架下对我们将此框架算法应用在正交结构蛇形机器人上的工作及过程做了详细的分析与阐述。包括控制器的定义、行为空间的定义及选择、性能度量、算法内部方法选择、样本采集过程及具体实现等。最后,我们在仿真工具上对应用了智能试错学习这一框架算法的蛇形机器人做了多种方式的验证。包括单关节损伤的多种不同损伤角度的情况、双关节同时损伤的多种典型情况,及三关节、多关节等极端的损伤情况做了验证。
[Abstract]:Snake-like robots have been developed and utilized because of the flexibility of motion brought by their multi-degree of freedom. In recent decades, many snake robot prototypes have been developed and even applied to practical applications. And there are plenty of control methods that have been proposed and used. But there is a problem with snake robots that people don't pay much attention to. That is, when a snake robot leaves the debugging environment to do its own motion in a complex natural environment, it will inevitably be damaged. When the snake robot is damaged, the existing practice is to either stop the current operation and be repaired and debugged by the researchers. Either detect according to their own default damage model and respond with the default response. The default damage detection model is limited. This paper introduces a new intelligent trial and error learning algorithm which can be applied to a snake robot with orthogonal structure, without presupposing damage model. The algorithm is now applied to hexapod robots and robot manipulators to guide the snake robot to find better compensation behavior with continuous trial and error-learning process. This algorithm has not been applied to snake robot for the first time and has been verified in many ways by simulation tools. Firstly, the significance of self-damage recovery of snake robot is expounded. The present research situation of snake robot and the research of robot damage recovery are introduced. Secondly, this paper introduces the framework algorithm of intelligent trial and error learning. Consists of two important steps-the establishment of the Behavior-Performance Map, The adaptive process of damage recovery is discussed, and the general control method of serpentine robot is discussed. The simulation tools integrated with physical engine are also introduced in this paper. We make a summary of the work needed to use intelligent trial and error learning algorithm, and make a necessary summary of the existing snake robot platform, including orthogonal structure, sensor system and so on. In the framework of induction and summary, the work and process of applying this framework algorithm to the orthogonal snake robot are analyzed and explained in detail, including the definition of controller, the definition and choice of behavior space, the performance measurement, the definition of controller, the definition and selection of behavior space, and the measurement of performance, including the definition of controller, the definition and selection of behavior space. Algorithm internal method selection, sample collection process and specific implementation. Finally, In the simulation tool, we verify the snake robot with intelligent trial and error learning algorithm in many ways, including many different damage angles of single joint injury, and many typical cases of simultaneous joint damage. And three joints, multiple joints and other extreme injuries were verified.
【学位授予单位】：中山大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP242

【参考文献】