【作者】 余伶俐；

【导师】 蔡自兴；

【作者基本信息】 中南大学 ， 控制科学与工程， 2010， 博士

【摘要】 多机器人任务规划是根据分配准则将任务分派至各机器人,并按最佳规划路径执行任务,能使多移动机器人系统高效完成任务。由于任务的复杂性与多样性,若无统一协调与统筹规划,将导致机器人系统资源消耗过大,甚至执行时发生故障。因此,多机器人任务规划是复杂任务高效完成的基石,同时移动机器人传感器系统故障诊断是任务成功规划的保障。在建立机器人团队控制平台的基础上,本文深入研究多移动机器人的任务探测方法,建立了任务失败概率最小的分配模型,设计了求解该分配模型的当代学习自适应离散粒子群算法。提出了求解多机器人任务规划方法及动态增量任务规划的策略,成功用于MORCS-2机器人团队中,有效解决了负载均衡的规划问题。同时,为了对机器人系统完成任务提供有效保障,研究了移动机器人传感器系统的故障诊断,并通过MORCS-1传感器系统进行了有效性的验证。论文主要工作及创新性成果如下：针对多机器人协作均分任务探测问题,研究一种均分点蚁群算法。利用多组蚂蚁群协作搜索策略,设计了一种蚁群算法的求解结构。根据任务均衡探测的原则定义了评价函数,避免了机器人最大负载过重问题。最后利用2-opt技术解决各子周游路径的交叉,获得了总规划路径较优解。实验结果表明,该算法可获得任务均衡探测的较优解,能解决多机器人系统中大规模任务均衡探测问题。针对多机器人执行任务失败概率最小的分配问题,综合考虑机器人任务完成效率、机器人能力以及任务性质等因素,建立了多机器人任务分配的数学模型。并提出一种当代学习自适应混合离散粒子群算法求解该模型。该算法依据粒子多样性变化规律,引入自适应扰动算子,以保持种群进化能力。设计了当代学习因子以体现粒子当代学习能力,改进其运动方程,有效地提高算法的鲁棒性。通过融入近邻搜索变异策略,极大地提升算法的局部求精能力。经实验表明：当代学习自适应混合离散粒子群算法具有强寻优能力和鲁棒性,同时也验证了任务分配模型的合理性。将多机器人任务规划分解为任务分配与路由规划两部分,分别提出一种空间正交分配技术求解任务分配问题,设计异质交互式文化混合算法体系框架,解决最佳路由规划问题。其中任务分配根据三维空间建模原理,利用空间正交试验方法,以负载均衡为目标更新并确定吸引算子,降低计算复杂度。提出一种异质交互式仿生群协进化体系框架,包括基于佳点集遗传算法的上层知识空间、基于离散粒子群优化的底层主群空间、自上而下的影响机制和自下而上的接受机制。并利用佳点集初始化主群空间,使初始粒子群均匀分布于可行域中；定义了粒子进化模型和进化力指标,提高种群的多样性和算法稳定性。最后,将空间正交分配异质文化混合算法在MORCS-2机器人团队平台上得到了充分的验证。在此基础上,设计一种基于规则的贪婪策略求解随机增量任务重规划问题,使得在重分配后机器人负载仍保证均衡,通过TSPLIB中不同任务地图进行测试,验证了算法的合理性。在任务规划过程中若机器人航迹推算系统发生故障却未得到及时诊断,很大可能导致机器人任务执行失败。针对此类移动机器人航迹推算系统的故障诊断问题,提出一种多模态进化Rao-Blackwellized粒子滤波器(multi-modality evolutionary Rao-Blackwellized particle filter, MERBPF)算法。该算法利用粒子滤波器估计机器人故障状态,采用卡尔曼滤波精确计算运动状态,有效地降低高维状态空间复杂度。为解决由粒子贫乏引起的不一致性问题,根据粒子多样性加入扰动因子,融入交叉种群与变异种群优化策略。以专家规则判定运动状态所对应的ERBPF,构造了复杂逻辑表述方法。通过实验表明,在强过程噪声情况下,MERBPF表现出较高的鲁棒性,降低了机器人航迹推算系统故障诊断的误诊率。更多还原

【Abstract】 Multi-robot mission planning assign tasks to each robot under the allocation criteria, and plan task execution order in accordance with the optimal path requirement, which could complete tasks efficiently by mobile-robot system. Because of tasks’complexity and diversity, if robots don’t coordinate together, it may lead to excessive consumption of robotic system costs, or even result in robots malfunction. Therefore, multi-robot mission planning is the cornerstone of the completion of complex tasks. Meanwhile robot-sensor system fault diagnosis is the basic guarantee for successful mission planning.Based on the establishment of heterogeneous robot team control platform, this paper carries on a in-depth study of multi-robot system tasks detection method, establishes a task allocation model according to the minimum probability of mission failure, and designs a optimization algorithm to solve the allocation model. On this basis, this paper puts forward a strategy for multi-robot task planning methods solving and dynamic incremental mission planning. This strategy is used in MORCS-2 robot team, which has achieved apparent accomplishment. At the same time, in order to ensure the proper completion of the planned tasks, this paper carries on the fault diagnosis study of robot sensor system. Main research work and innovative achievements are as follows:An equal division point ant colony algorithm (EDPACA) is proposed to solve the multi-robot collaboration mission exploration. the algorithm is designed by a novel solution construction through multi-group ants search cooperatively strategy and a more reasonable evaluation function is define which consider sufficiently equal allocation exploration mission, and avoid a max-consuming robot overload. At last, the crossover problems of sub-circular paths are solved by 2-opt method. The experiment results show that the proposed algorithm is available to gain better solution, and solved multi-robot system a large-scale of tasks balance exploration problem.Aiming to multi-robot collaborative tasks allocation problem for the smallest failure probability, tasks allocation mathematical model is established firstly, which considers three factors comprehensively:the efficiency of executing mission, the ability of robot and the nature of the mission. Current learning Discrete Particle Swarm Optimization Algorithm (CLDPSO) is proposed to solve this model. Adaptive perturbation factor is introduced according to the population heterogeneity to keep particle swarm evolutional capability. Based on the excellent performance of particle swarm society learning ability and individual learning ability in DPSO, we propose a new conception current learning factor to improve the DPSO kinetic equation, and the robost of CLDPSO is better. Finally nearby neighbor mutant strategy is added to increase local search capabilities. The experiment results show that CLDPSO has strong optimization ability and robustness, meanwhile the rationality of the task allocation model is verifiedMulti-robot mission planning divides into task allocation and route planning subdivision, we design spatial orthogonal cluster algorithm for multi-robot task assignment problem, and propose novel system architecture of heterogeneous interactive cultural hybrid algorithm to solve the best route planning problem. The tasks assignment problem adopts 3-D space model, utilizes spatial orthogonal test technology. The attractor position are updated according to load balance objective function, this approach is high validity but lower complexity. Heterogeneous interactive cultural hybrid algorithm firstly initializes population space using good-point-set in order to make particles swarm uniform distribution in feasible region. Secondly, novel evolution model and particle evolution ability indexes are redefined, which increase particles swarm diversity and improve algorithm stability. At last the results are shown that SOCHCHA is superior and significant. Meanwhile, we practice SOCHCHA on MORCS-2 robot-team platform which exhibits the algorithm practicability. On this basis, a rule-based greedy algorithm for stochastic incremental task re-planning is designed, which makes mobile-robot load balance after re-planning, the algorithm is reasonable which is verified by using different TSPLIB mission maps.During mission planning process, if mobile-robots dead reckoning system breaks down, and don’t diagnose it on time, which may lead to fail for robot execution tasks. A multi-modality Rao-Blackwellized evolutionary particle filter (MERBPF) algorithm is devised for those fault diagnosis problems. Particle filter is utilized to estimate robot fault state and Kalman Filter is used to calculate accurately kinetic state, so as to drop the complexity of high-dimensional state space. The inconsistency from particle degeneration problem is solved by integrating swarms’intercross and mutation strategy, and adding disturbance factors accoding to diversity. Robot moving states are determined by expert rules reasoning mechanism and monitored by each different ERBPF. Finally the multi-modality ERBPF are formed which express complex logic clearly. MERBPF maintains a strong robustness even under the strong process noise. Meanwhile MERBPF reduces diagnostic errors rate for fault diagnosis of robot’s dead reckoning system.更多还原