【作者】 姜健；

【导师】 赵杰；

【作者基本信息】 哈尔滨工业大学 ， 机械电子工程， 2008， 博士

【摘要】 人工智能科学研究的进展和通讯技术的进步为多移动机器人系统的深入研究和广泛应用奠定了牢固的技术基础。随着人类要求机器人完成任务难度的提高,多移动机器人系统的应用范围越来越广泛。许多任务由多个机器人协同完成要优于采用单个机器人。多移动机器人系统不仅能够有效地利用机器人之间的相对定位及信息共享机制,提高系统效率和任务完成的质量,而且可以完成单一机器人无法完成的时间和空间并行的多任务。然而,多个机器人之间如果没有一个良好的协作机制,不仅会由于机器人之间的重复工作和彼此干扰而导致系统整体性能下降,而且会由于冲突加剧和发生死锁现象而产生机器人安全隐患甚至使系统发生瘫痪而无法完成预定任务。因此多移动机器人协作方法研究是多移动机器人系统研究的核心问题。本文重点研究了基于心理学概念的多移动机器人协作方法和基于生物启发策略的多移动机器人协作方法。并对多移动机器人紧耦合任务和松耦合任务分别进行了强化学习方法的研究。分析和比较了各种单体机器人控制结构,在此基础上提出了一种面向多移动机器人协作的基于目标型行为有限状态机的单体机器人控制结构。分析和比较了各种多机器人群体控制结构的特点,在此基础上提出了一种基于群智能方法的分布式群体控制结构。为了更好地说明任务概念和为基于任务的多移动机器人协作研究奠定基础,提出了一种MTB多移动机器人任务层次结构。受蚂蚁等社会性昆虫营养交哺现象的启发,提出了一种基于交哺行为的Mission级任务一致性保持方法,使多移动机器人系统可以自主地实现任务的一致性保持,为通信范围有限或者通信信息需要保密的敌对环境下的系统任务命令的有效传播提供了一种鲁棒和可靠的群智能方法。在模拟蚂蚁等社会性生物采用信息素进行信息交流的隐式通信方式的基础上,提出了一种基于排斥信息素型蚁群算法(PR-ACA)的多移动机器人自主任务分配方法。进行了未知非结构化环境下的多移动机器人协作搜集仿真实验。仿真结果表明,采用该方法可以实现多移动机器人搜集任务的自主分配,有效减少机器人的空间冲突,尤其在机器人数量较多的情况下,更能显示出该方法的优势。提出了一种基于心理状态参数的多移动机器人协作行为决策方法。机器人根据对自身内外环境及队友状态的估计产生心理状态参数。利用神经网络建立心理状态参数与机器人协作倾向阈值之间的映射关系。在遇到需要产生协作行为的场合根据这些阈值来快速产生反应。实验结果表明所提方法既可以保证机器人根据环境做出合理的判断,又可以保证机器人反应的快速性。针对目前应用于多移动机器人协作的拍卖方法很少考虑机器人参与拍卖的时机是否适当这一问题,在拍卖方法中引入了心理学的焦虑概念,提出了基于焦虑/拍卖的多机器人协作方法(AACM)。实验结果表明,与单纯的拍卖方法相比,该方法能够提高多移动机器人协作搜集任务的执行效率。针对紧耦合任务的特点,应用强化学习方法进行了多移动机器人编队导航任务的研究。高层采用“条件——行为对”强化学习机制解决了多移动机器人长障碍物的绕行方向选择问题。中层采用一种角色交叉包含式控制结构作为编队保持机制。底层采用“动作——状态对”强化学习机制进行避碰规则的学习。通过多移动机器人编队导航仿真实验证明了所提三层控制方法对于解决多机器人编队导航任务的有效性和高效性。针对松耦合任务的特点,作为所提面向多移动机器人协作的单体机器人控制结构的进一步研究,提出了一种带“共享区”的基于目标型行为的强化学习算法,研究了通过强化学习机制获得基于目标型行为有限状态机的方法,通过多机器人搜集仿真实验证明了该方法的可行性。最后,建立了面向多移动机器人协作方法性能验证的多移动机器人实验系统。通过实验对所提的多移动机器人协作方法进行了验证。实验结果证明了所提多移动机器人协作方法的有效性和高效性。更多还原

【Abstract】 The development of the artificial intelligtnce science and the communication technique lays the firm foundation for further research and extensive application of multiple mobie robots system. As robots are charged with increasingly difficult tasks, the application domain of multiple mobile robots system expands. Many tasks can be better achieved by a team of robots than by a single robot. Multiple mobie robots system not only can effectively utilize the relative localization technique and information sharing mechanism to advance the system efficiency and the quality of accomplishing missions, but also can accomplish the spatial parallelism tasks which can not be realized by a single robot.However, if there is not a good cooperation mechanism among multiple robots, the whole system performance will decrease because of repeat work and mutual interference. The whole system will have safety hidden danger and even break down and can not accomplish the scheduled mission. So the research of multiple mobile robots cooperation method is the key problem of multiple mobile robots system research. The multiple mobile robots cooperation method based on psychology concept and the multiple mobile robots cooperation method based on biological inspired strategies are emphatically analyzed. At the same time, the RL method for loosely-coupled task and tightly-coupled task are respectively studied. Various single robot control structures are analyzed and compared. On the basis of this, a kind of single robot control structure named Target Type Behavior FSM was presented. Various group architectures are analyzed and compared. On the basis of this, a kind of distributed structure based on swarm intelligent method was presented. To fully describe the task conception and to lay a foundation for task based multi-robot cooperation methods research, a kind of task hierarchical structure called MTB was presented.Simulating the trophallaxis behavior of the social insects such as bees and ants, a kind of multi-robot systems mission consistency maintenance method was presented. It provided a robust and reliable swarm intelligence method for the effective communication of multiple mobile robots mission especially in limited communication range environment and security environment. The paper presented a multi-robot systems mission consistency maintenance method based on trophallaxis. A multiple mobile robots task allocation method based on a kind of Ant Colony Algorithm (ACA) named Repulsion Pheromone Ant Colony Algorithm (PR-ACA) was presented. Under the unknown unstructured environment, multi-robot cooperative foraging experiment was performed. Experiment results showed that the method presented in the paper could make multiple robots perform task allocation autonomously and effectively, and could decrease spatial confliction among robots, especially when the group was large.A multiple mobile robots cooperation behavior decision method based on psychological state parameters was presented. Robots gave birth to psychological state parameters based on the estimations of environment, teammates and themselves. The mapping relationship between psychological state parameters and cooperation tendency threshold values were set up with neural network. Robots could make decision on the basis of these threshold values on cooperation occasions. Experiment results showed that the multi-robot cooperation method could ensure both the rationality and the speediness of robots’decision-making. Few had been thought about the application occasion problem of the auction method for the cooperative multi-robot foraging. To solve the problem, anxiety conception was introduced. A new algorithm named Anxiety and Auction Cooperation Method (AACM), combined anxiety concepton with auction method, was presented. Experiment results showed that, compared with the only auction method, the method pesented in the paper could improve the efficiency of performing multiple mobile robots cooperation foraging.The paper studied multiple mobile robots formation and navigations according to the respective characteristics of tightly-coupled tasks. The high level used station-behavior pair reinforcement learning to learn the circumambulating direction according to the obstacles. A Role-Cross-Subsumption control structure was presented to be the middle layer of the three layers control structure. The low level used action-station pair reinforcement learning to learn the regulations for preventing collisions. Multiple mobile robots formation and navigations stimulation experiments validated the effectiveness and high efficiency of the method. As the further research of the single robot control structure for multiple mobile robots cooperation in the paper, a kind of Target Type Behavior RL with share section was presented. The paper studied the method to obtain Target Type Behavior FSM with RL. Stimulation experiments validated the feasibility of the method.Finally, a hardware experiment system for validating multiple mobile robots cooperation methods was built. The experimental results validated the system efficiency of multiple mobile robots cooperation methods presented in the paper.更多还原