【作者】 杨茂；

【导师】 田彦涛；

【作者基本信息】 吉林大学 ， 控制理论与控制工程， 2010， 博士

【摘要】 本文主要研究了群体机器人系统的协同适应性问题。目的是通过基于局部信息交互下分布式控制、优化与学习,实现群体机器人系统对于动态复杂环境的适应,进而揭示群体智能系统中涌现行为的规律及其可控性。分别从群体机器人运动学模型、动力学模型以及分布式强化学习三个方面四个角度进行了较为深入的研究。1、对于运动学模型下协作,通过结合Vicsek模型及人工势场法,实现了群体机器人在复杂环境中的同步运动,借助人工协调场方法实现了群体机器人的避障运动。2、对于动力学模型下的协同,通过结合虚拟力、外部环境因素以及连接矩阵的方法,针对静态环境和动态环境分别设计了基于局部信息交互的分布式控制器,并进行了相应的稳定性分析,实现了群体机器人系统的同步运动,利用改进的粒子群优化算法对控制器中的参数进行优化,降低了群体机器人系统的能量损耗。3、提出了基于内部平均动能的分布式控制器设计方法,并进行了稳定性分析,有效地实现了群体机器人在复杂环境下的觅食任务。4、对于分布式强化学习,由于状态空间过大,机器人数量较多,导致强化学习的收敛速度过慢,适量的通信能够加快学习速度,这里提出几种基于黑板结构通信的协作强化学习算法,充分利用了多机器人系统的分布式感知能力去探索学习空间与收集经验,进而提高强化学习收敛速度。本文研究得到了国家自然科学基金项目“复杂环境下群体机器人系统协同适应性理论与方法研究”(60675057)以及吉林大学2009研究生创新研究计划“动态环境下群体机器人协同方法研究”(20091020)的资助。更多还原

【Abstract】 With the development of computing technology, sensor technology, communication technology, control theory, artificial intelligence, and some research of robot formed by a number of interdisciplinary have also entered a new stage. Thus swarm robotics is produced by some social insects in nature-inspired. Swarm robots system is a special class of multi-robot system, with the features of robustness, adaptability, and scalability.Research on swarm robotic system has important significance in theory and practice. In theory, with the further research of swarm robotics, it will help reveal the emergence of a fundamental mechanism for intelligent behavior. In practice, a mature swarm robots system can be in the ship manufacturing, product assembly, transportation systems, military equipment, aerospace and other areas of the completion of certain dangerous work independently and therefore have high potential applications.Coadaptivity for swarm robots system means the ability of the autonomous robot optimize their own control strategies constantly, and adjust behavior to meet the dynamic changes in the environment and the characteristics of the task, and ultimately the overall optimality through the local information with other robot and external environment in a complex dynamic environment.This dissertation research on some problems of the coadaptivity for swarm robots system, which are based on the tasks of foraging and flocking control. The work is supported by the National Natural Science Fund of China under Grant 60675057.Name of the project is“coadaptivity theory and methods research for swarm robots system in a complex dynamic environment”.The major work of this dissertation studies on the following four aspects:1. The distributed control strategy is studied for flocking under swarm robots system kinematics model. The achievement and maintenance of flocking formation is implemented for swarm robots systems though the method combined the Vicsek model and improved artificial potential field in an environment without obstacles. In order to achieve flocking and avoiding the static obstacles in the environmentrun, the combination of update rule for Vicsek model and artificial coordinating field is adopted to design coorespoding control strategy, then the improved particle swarm optimization algorithm is presented to optimize the coorespoding parameters in order to achieve a stable flocking behavior. The data of simulation experiments show that the distributed control strategy can implement the flocking behavior for swarm robots system effectively in the absence of or with static obstacles environment.2. The distributed control strategy is designed for flocking under swarm robots system dynamics model based on local information exchange. Analysis of the stability of the distributed controller, and estimates the corresponding finish time of the flocking behavior. The improved particle swarm optimization algorithm is presented to optimize the coorespoding parameters in order to minimize the energy consumption during the process of the motion. In order to solve the problem how to design the controller cause the swarm robots system flocking in an dynamic environment. The distributed control strategy is designed based on local information exchange. In order to analyze the stability of the non-autonomous system, Barbalat lemma is introduced under the coorespoding assumption,such that the speed of all the individuals converge to the same curves. Simulation results show that for the above two cases, the designed distributed control strategy can achieve a stable flocking behavior for the swarm robots system effectively and rapidly.3. The distributed c ontrol strategy is designed for social swarm foraging in an consistent environment under swarm robots system dynamics model based on interal average kinetic energy, so that the swarm robots system can finish the foraging task efficiently under the non-flocking condition based on local information exchange, and prove that the value of intermal average kinetic energy will eventually converge to the prior expectation value in a damping environment. Simulation results show that the convergence lower value of internal average kinetic energy, make the swarm robots system as a whole cover a smaller area in the search space, by contrast the convergence higher value of internal average kinetic energy, make the swarm robots system as a whole cover a larger area in the search space, then the swarm robots system can find the extreme value of the environment function more efficiently, to finish the swarm social foraging task.4. The method of cooperative Q-learning is presented based on blackboard architecture, targeted at some shortcomings as follows: poor scalability of the point to point communication, too much communication traffic and too slow convergence speed of reinforcement learning. The learning process is executed at the blackboard architecture making use of the advantage of robots number and distributed sensing capability in the training scenario to explore the learning space and collect experiences. Communication is essential for swarm robots system which can be used to share experiences, parameters and control policies. Resent research proofed that proper communication can largely improve the performance of swarm robots system.It can achieve to independence of each robot reinforcement learning in experience sharing by learning automata and improved particle swarm optimization algorithm. Simulation results show that the model can improve the learning speed and reduce communication traffic.In summary, some problems of the coadaptivity for swarm robots system are studied in this dissertation. The main purpose of this work is to establish a complete theory of coadaptivity and its implementation for swarm robots system. And then, Simulation experiments are performed for the purpose of related verification and analysis.更多还原