矿井救援多机器人系统任务分配优化算法研究.pdfVIP

摘 要 近年来，机器人技术得到了快速的发展。与单机器人系统相比，多机器人系 统有着无法比拟的优势，然而多机器人系统能否优于单机器人系统的关键是多机 器人系统中的机器人成员能否通力合作，高效地完成全局任务。本文以多机器人 系统任务分配算法为研究对象，在对国内外研究现状进行深入分析的基础上，针 对矿井事故救援难度大，矿井环境动态性复杂性的特点，提出了一种基于效用评 价的矿井救援多机器人系统任务分配算法。 本文对多机器人系统进行简单介绍，并详细介绍了模糊神经网络和 Q 学习 算法，在此基础上设计了基于模糊-神经 Q 学习的效用评价网络结构，通过实时 学习与参数调整，提高了评价精度。设计全局奖励函数从而使每个机器人根据各 自的任务评价进行任务分配，实现全局任务的分配。并将其应用于矿井救援中， 实现了救援伤者任务效用评价和任务选择，通过算法的比较，验证了方法的可行 性和精确性。最后，设计了矿井救援机器人的硬件结构和软件流程，重点部分是 矿井机器人救援的导航算法实现，提出多机器人系统仿真的解决方案。极大地降 低矿井环境的不确定性和危险性，但是由于条件和时间的限制，在很多方面还有 待改进和深入研究。 关键词： 矿井救援 多机器人系统任务分配 多机器人仿真系统 I Abstract Recently, Robot technology is developing rapidly. Compared with single-robot systems, multi-robot systems have more advantages. But it has become a key problem that whether they can cooperate with each other effectively or not. The research objectives of this paper are a task allocation algorithm based on multi-robot systems. On the basis of deep analysis to the research in and out of China, according to the characteristics of rescue difficulties of mine accident and dynamics, we proposed a task allocation algorithm based on utility valuation . The paper introduces multi-robot systems and thesis of Q-learning and adaptive neural-fuzzy inference system .particularly make a study of utility evaluation algorithm based on Q-learning and adaptive neural-fuzzy inference system which has a good generalization. In this algorithm utility evaluation parameters are modified gradually by on-line learning. We designed different global rewarding functions, which make robots can realize global task allocation and applied in mine rescue. The realization of task election and utility evaluation of are stated in detail in this dissertation. Through the comparison with algorithms ,the feasi