【作者】 王梅；

【导师】 吴铁军；

【作者基本信息】 浙江大学 ， 控制科学与工程， 2007， 博士

【摘要】 多关节自治机器人协作系统属于多智能体机器人系统的范畴,在承载能力、操作精确性、灵活性、系统的可靠性、对变化不确定环境的适应性等方面都比传统的机器人有更大的优越性。协作系统具有空间分布、功能分布、时间分布等特点,通过共享资源(信息、知识等)弥补了单机器人能力的不足,扩大完成任务的能力范围。系统内机器人资源的冗余性提高完成任务的可能性、增加了系统的可靠性与稳定性。运动规划是机器人决策的重要环节之一,多个关节式机器人的协同工作对协作运动规划提出了更高的要求。现有的集中式规划方法和分散规划—集中协调的统一协调式规划方法难以满足应用系统对空间—时间综合优化、速度约束、协作任务多样化和系统可靠性等多方面的综合要求,无法发挥分布式自治多机器人协作系统的优势。本论文提出了一种多关节自治机器人协作系统的分布式协作运动规划方法。系统中各机器人个体的规划子系统相对独立、平等地根据各自的局部信息进行自身的运动规划,仅通过少量异步通信,获取其它机器人的决策信息,完成相互避碰和特定的协同作业等任务,系统总体能够实现空间—时间的综合优化目标,规划系统对于一定程度上的通讯不可靠具有抗干扰性。因此在多机器人外太空作业和外星球探险、野外自主协同作业、深海协同作业等新的机器人应用领域具有重要的应用价值。本文的主要研究工作包括:1、给出了待求解的多关节自治机器人协作系统运动规划问题的数学描述。从优化角度,明确了多关节自治机器人协作运动规划的优化目标和运动路径综合评价标准,详细分析了影响机器人运动的各类约束条件。论证了空间优化与时间优化的一致性和可分离性问题,说明空间—时间综合优化指标的重要意义。提出了对协作运动规划方法的性能要求。2、针对多关节自治机器人协作系统,提出一种新的分布式协作运动规划(Distributed Cooperative Motion Planning,DCMP)方法,在同一规划层次上实现空间—时间运动路径综合优化。以分布式求解为目标,通过严格的数学推导,将多关节自治机器人协作系统的整体运动规划任务分解为各机器人个体的独立子规划。在满足综合优化目标的前提下,推导出能够分布式迭代计算的子系统优化模型。DCMP方法解耦了各机器人协作运动子规划的同步性,使其可以异步并行进行。对DCMP方法的性能,尤其是抗通讯干扰的能力,进行了分析研究。DCMP方法在协作运动规划的体系结构和协作机制方面进行了有益的尝试和创新。3、利用协同进化计算作为DCMP方法的算法实现工具。在满足数学严谨性的基础上,采用计算智能方法求解各机器人协作运动子规划。DCMP方法充分利用了协同进化算法的并行计算能力、在优化搜索方面的启发功能和对约束条件的处理方法,实现了运动路径的空间—时间综合优化,并解决了其它方法难以实现的多关节式机器人动态约束(如速度约束)问题。4、从多关节自治机器人协作运动规划的特点出发,讨论了基于协同进化计算的DCMP方法在个体编码方法、遗传操作算子、约束条件处理方法、评价函数计算、以及预估值的更新计算等方面的具体实现。该算法具有普适的计算框架,适用于各种机器人协作工作任务。针对同时启动不同时终止、同时启动同时终止和末端距离约束等不同的协作任务,给出了不同的处理办法。最终通过仿真实验证明算法有效性。5、本文以浙江大学信息科学与工程学院智能系统与决策研究所多机器人协作控制实验室的多关节自治机器人协作项目为应用背景。在DCMP方法基本原理和基于协同进化计算的算法实现基础上,针对实际系统的特点,发展基于DCMP方法的相关实用技术。针对高维C空间搜索困难的问题,借鉴路标法的降维概念及路标点选择技术,提高DCMP方法在高维空间的优化搜索能力;为了适应多协作任务的运动规划要求,改进DCMP方法,为高层次的协作任务规划提供更多的选择余地。藉此提高算法的现实可行性和实用价值。最终通过实验验证DCMP方法的有效性、DCMP方法与协作任务规划器配合的能力、以及通讯不可靠情况下的工作能力。更多还原

【Abstract】 Autonomous multi-joint robot system, which belongs to the Multi-Agent Robot System (MARS), has advantages over traditional robots in load bearing, accurate operating, flexibility, and adaptability in uncertain environment. The robots in cooperative system are spatial, timing and functional distributed. As a result of resources sharing, say, information and knowledge, the system overcomes the lack in capability of single robot, broadens the application fields of robots. Redundancy in robot resource improves reliability, stability, and the possibility to accomplish tasks.Motion planning is an important issue in robot decision making. In multi-robot cooperation, the demanding for cooperative motion planning is much higher. Centralized planning and decoupled planning have difficulties in space-time integrated optimization, velocity constraints handling, cooperative task diversity, and system reliability. They can not take full advantage of the distributed cooperative robot system.A distributed cooperative motion planning (DCMP) for autonomous multi-joint robot system is proposed in this paper. The motion planning subsystems of individuals in the system are independent and equal. They conduct collision-free motion planning according to limited decision information of other robots, which is acquired by asynchronous communication, therefore fulfill given cooperative tasks. The planning method has the ability to realize space-time integrated optimization, and to work under certain unreliable communication. It is valuable in many new application fields, such as outer-space exploration, interstellar exploration, autonomous cooperative fieldwork and deep-water cooperationIn this dissertation, the main research is focused on the following works:1) The problem of cooperative motion planning for autonomous multi-joint robot system is described mathematically. Optimization goal, evaluation criteria of path, constraints that influence the motion of robots are singled out in detail. By analyzing the consistency and separability of time optimization and space optimization in cooperative motion planning, the importance of the space-time integrated optimization criteria is emphasized. The performance requirements of cooperative motion planning method are specified.2) A new distributed cooperative motion planning (DCMP) method is proposed for cooperation of multiple autonomous multi-joint robots. The space-time integrated optimization is achieved on the same planning hierarchy. Objective to distribute problem-solving of the cooperative motion planning, the system is decomposed, by mathematical derivation strictly, into subsystems of each robot that can be calculated iteratively and independently in an asynchronous way. The performance of DCMP method, especially the resistance capability to the communication interference, is analyzed. cooperative motion planning method are specified. The DCMP method try to improve the collaboration architecture and coordination mechanism of cooperative motion planning.3) Based on mathematical strictness, cooperative co-evolutionary method is exploited in the DCMP method for optimal solution finding in sub-planning of each robot. Benefit from its parallel computation, heuristic search and constraints handling ability, the DCMP method realizes space-time integrated optimization naturally, and solves the problem of dynamic constraints, such as velocity constraint, which is very difficult for most cooperative planning methods.4) According to the characteristics of cooperative motion planning of autonomous multi-joint robot system, key points in the DCMP implementation, such as real-valued problem-specific chromosome representation, problem-specific evolutionary operators design, constraints handling, fitness evaluation method and estimation updating, are discussed in detail. The planner is a general framework and can be used for different kinds of tasks, such as simultaneous start-asynchronous stop, simultaneous start-simultaneous stop, and terminal distance keeping. The effectiveness of the DCMP method is proved by simulation results.5) A motion planning based on the DCMP method is applied to a pilot multiple autonomous multi-joint robot system in the Multi-robot Cooperative Control Lab, of which is the Institute of Intelligent Systems & Decision Making of Zhejiang University. Based on the principles of the DCMP method and its cooperative co-evolutionary implementation, some practical techniques are developed to enhance its practical feasibility and value. Motivated by the reducing dimension concept and point selection technology of the roadmap method, the capability of DCMP method in exploring in very high dimensional composite configuration space is enhanced. The DCMP is improved to satisfy the requirements of high-level cooperative mission planning and to provide more choice. The effectiveness, as well as its ability to work with the cooperative mission planning and under unreliable communication, are validated on a number of experiments.更多还原