基于冲突搜索的多智能体路径规划研究进展

多智能体路径规划是人工智能领域一个经典的搜索问题，基于冲突的搜索算法是当前解决该问题的最优算法之一。文中讨论了多智能体路径规划的基础研究，对国内外近年来基于冲突搜索算法及其变体的研究成果进行了分类，根据改进方式将其变体分为4类，包括分割策略的改进、启发式算法、对典型冲突的处理和次优算法。同时介绍了基于冲突的搜索算法在多智能体路径规划的扩展问题中的应用。最后根据当前算法的优缺点，指出了目前面临的挑战，并针对这些挑战给出了未来可能的研究方向。     

一种基于改进冲突搜索的多机器人路径规划算法

针对智能仓储环境下多载位自主移动机器人集群拣选-配送路径规划问题,提出一种改进型基于冲突搜索的多智能体路径规划算法.在模型方面,采用多载位机器人替代KIVA机器人,建立以最小化拣选-配送时间以及无效路径比为目标的数学规划模型.在算法方面,首先,提出一种基于优先级规则的多智能体冲突消解加速策略;然后,设计基于动态规划的单机器人拣选序列优化算法;最后,设计考虑转向惩罚的增强A*算法搜索机器人最优路径.实验结果表明:所提出模型与KIVA系统相比有较大优越性;所提出算法能够有效缩短拣选-配送时间、减少无效路径时间.     

多智能体路径规划综述

多智能体路径规划（multi-agent path finding,MAPF）是为多个智能体规划路径的问题,关键约束是多个智能体同时沿着规划路径行进而不会发生冲突。MAPF在物流、军事、安防等领域有着大量应用。对国内外关于MAPF的主要研究成果进行系统整理和分类,按照规划方式不同,MAPF算法分为集中式规划算法和分布式执行算法。集中式规划算法是最经典和最常用的MAPF算法,主要分为基于[A*]搜索、基于冲突搜索、基于代价增长树和基于规约四种算法。分布式执行算法是人工智能领域兴起的基于强化学习的MAPF算法,按照改进技术不同,分布式执行算法分为专家演示型、改进通信型和任务分解型三种算法。基于上述分类,比较MAPF各种算法的特点和适用性,分析现有算法的优点和不足,指出现有算法面临的挑战并对未来工作进行了展望。     

大规模AGV的改进时间窗路径规划

在机器人路径规划中,A*算法搜索路径时存在大量冗余节点,随着任务量增加,其搜索效率也会急剧下降,因此无法适应大规模任务下的路径规划。为此提出一种改进时间窗的有界次优A*算法用于求解大规模自动导引车（automatic guided vehicle,AGV）路径规划问题。算法使用时间启发式,并在搜索过程中采用时空搜索,规划无冲突的最优或次优路径。算法主要进行了三处改进：采用时间启发式,缩短了路径时间;采用动态时间窗算法,避免多次路径规划;优化了聚焦搜索算子,降低负反馈。通过MATLAB实验结果证明改进后的算法在进行多机器人路径规划时,能快速有效地规划出无冲突的平滑次优路径,搜索效率高,稳定性强。     

基于几何方法的多智能体群体刚性运动的路径规划

利用微分几何的方法在特殊欧氏群(Special Euclidean Group)上研究刚性运动.对于已知边界约束条件的多智能体群体刚性运动路径规划问题，采用了虚拟结构的模式；通过建立基于能量的最优准则，在特殊欧氏群上推导了多智能体协调运动的最优轨迹.所得到的最优轨迹是光滑的，而且与空间惯性坐标系的选择无关.该方法对于描述和规划多智能体的群体刚性运动具有广泛的适用性.     

基于改进CBS算法的自动化码头多AGV无冲突路径规划

针对自动化集装箱码头上自动引导车（automated guided vehicle,AGV）数量增加导致冲突更频繁。提出一种改进的基于冲突的搜索（conflict based search,CBS）算法。底层采用基于曼哈顿距离的A*算法,上层结合二叉树原理建立冲突树对AGV之间的冲突进行规避。以最小化AGV在岸桥和堆场之间的总路径长度为目标,使用栅格法建立AGV路网模型。考虑AGV之间的点冲突与边冲突,将自动化码头多AGV无冲突路径规划问题规约为多智能体寻径问题。实验结果表明,所提出的算法在保证堵塞率为0%的前提下,缩短总路径长度并提高运算速度,验证算法的有效性。     

基于深度强化学习的多配送中心车辆路径规划

多配送中心车辆路径规划(multi-depot vehicle routing problem, MDVRP)是现阶段供应链应用较为广泛的问题模型,现有算法多采用启发式方法,其求解速度慢且无法保证解的质量,因此研究快速且有效的求解算法具有重要的学术意义和应用价值.以最小化总车辆路径距离为目标,提出一种基于多智能体深度强化学习的求解模型.首先,定义多配送中心车辆路径问题的多智能体强化学习形式,包括状态、动作、回报以及状态转移函数,使模型能够利用多智能体强化学习训练;然后通过对MDVRP的节点邻居及遮掩机制的定义,基于注意力机制设计由多个智能体网络构成的策略网络模型,并利用策略梯度算法进行训练以获得能够快速求解的模型;接着,利用2-opt局部搜索策略和采样搜索策略改进解的质量;最后,通过对不同规模问题仿真实验以及与其他算法进行对比,验证所提出的多智能体深度强化学习模型及其与搜索策略的结合能够快速获得高质量的解.     

基于协同进化的多智能体机器人路径规划

协同进化是一种新兴的、简单有效的智能优化方法,具有较好的收敛性、鲁棒性和高效性,在多目标优化问题中得到很广泛应用。将其应用到复杂环境下多智能体机器人的路径规划中,并设计适应度评价函数。同时,引入一系列新的变异操作算子,有效地对多智能体机器人规划的路径进行优化,加速了整体的规划速度,避免规划陷入局部最优,从而获得多智能体系统的全局最优或次优解。最后给出了的仿真结果证明方法可行、有效。     

海上落水目标协同搜寻路径规划算法研究

海上落水目标协同搜寻路径规划与多旅行商问题相似。论文所提出的算法由待救目标分类和搜寻路径规划两个子算法组成。首先经过基于遗传算法的K-means目标聚类,确定染色体数量及长度,解决海上搜救目标分类问题。然后经过多染色体遗传算法,获得多种搜救设备协同的海上搜救最优路径。计算结果表明,论文提出的海上落水目标协同搜寻路径规划算法,能够有效降低算法的搜索范围,提高算法的运行速度和全局搜索能力,提高海上落水目标搜救效率。     

基于构型空间法的机器人路径规划研究

提出了一种静态环境下的机器人路径规划仿生算法,该算法用构型空间法对场景进行建模,模拟蚂蚁群体觅食的智能行为,由多只蚂蚁协作完成最优路径的搜索。搜索过程在基于蚁群优化算法的基础上增加了最近邻居策略和目标导引函数,使搜索过程快速高效。并在实验环境中对机器人路径规划进行仿真,结果显示在多障碍物下也能迅速规划出最优路径,表明研究的可行性和有效性。     

A genetic algorithm for multiprocessor scheduling

The problem of multiprocessor scheduling can be stated as finding a schedule for a general task graph to be executed on a multiprocessor system so that the schedule length can be minimized. This scheduling problem is known to be NP-hard, and methods based on heuristic search have been proposed to obtain optimal and suboptimal solutions. Genetic algorithms have recently received much attention as a class of robust stochastic search algorithms for various optimization problems. In this paper, an efficient method based on genetic algorithms is developed to solve the multiprocessor scheduling problem. The representation of the search node is based on the order of the tasks being executed in each individual processor. The genetic operator proposed is based on the precedence relations between the tasks in the task graph. Simulation results comparing the proposed genetic algorithm, the list scheduling algorithm, and the optimal schedule using random task graphs, and a robot inverse dynamics computational task graph are presented     

A stochastic algorithm for makespan minimized multi-agent path planning in discrete space

Makespan minimized multi-agent path planning (MAPP) requires the minimization of the time taken by the slowest agents to reach its destination. The resulting minimax objective function is non-smooth and the search for an optimal solution in MAPP can be intractable. In this work, a maximum entropy function is adopted to approximate the minimax objective function. An iterative algorithm named probabilistic iterative makespan minimization (PIMM) is then proposed to approximate a makespan minimized MAPP solution by solving a sequence of computationally hard MAPP minimization problems with a linear objective function. At each iteration, a novel local search algorithm called probabilistic iterative path coordination (PIPC) is used to find a sufficiently good solution for each MAPP minimization problem. Experimental results from comparative studies with existing MAPP algorithms show that the proposed algorithm strikes a good tradeoff between the quality of the makespan minimized solution and the computational cost incurred.     

随机梯度下降算法研究进展

在机器学习领域中, 梯度下降算法是求解最优化问题最重要、最基础的方法. 随着数据规模的不断扩大, 传统的梯度下降算法已不能有效地解决大规模机器学习问题. 随机梯度下降算法在迭代过程中随机选择一个或几个样本的梯度来替代总体梯度, 以达到降低计算复杂度的目的. 近年来, 随机梯度下降算法已成为机器学习特别是深度学习研究的焦点. 随着对搜索方向和步长的不断探索, 涌现出随机梯度下降算法的众多改进版本, 本文对这些算法的主要研究进展进行了综述. 将随机梯度下降算法的改进策略大致分为动量、方差缩减、增量梯度和自适应学习率等四种. 其中, 前三种主要是校正梯度或搜索方向, 第四种对参数变量的不同分量自适应地设计步长. 着重介绍了各种策略下随机梯度下降算法的核心思想、原理, 探讨了不同算法之间的区别与联系. 将主要的随机梯度下降算法应用到逻辑回归和深度卷积神经网络等机器学习任务中, 并定量地比较了这些算法的实际性能. 文末总结了本文的主要研究工作, 并展望了随机梯度下降算法的未来发展方向.     

多因素制约的迷宫问题最优解的求解算法

介绍迷宫问题及其最优解,引入多因素制约的迷宫问题。重点讨论多因素制约迷宫问题最优解的含义及基于广度优先搜索的求解算法,并通过两个实例分析如何基于广度优先搜索算法求解这类迷宫问题的最优解,并给出算法的伪代码。最后,进一步讨论和总结这类迷宫问题最优解的求解算法。     

On efficient sensor scheduling for linear dynamical systems

Consider a set of sensors estimating the state of a process in which only one of these sensors can operate at each time-step due to constraints on the overall system. The problem addressed here is to choose which sensor should operate at each time-step to minimize a weighted function of the error covariances of the state estimates. This work investigates the development of tractable algorithms to solve for the optimal and suboptimal sensor schedules. A condition on the non-optimality of an initialization of the schedule is developed. Using this condition, both an optimal and a suboptimal algorithm are devised to prune the search tree of all possible sensor schedules. The suboptimal algorithm trades off the quality of the solution and the complexity of the problem through a tuning parameter. The performance of the suboptimal algorithm is also investigated and an analytical error bound is provided. Numerical simulations are conducted to demonstrate the performance of the proposed algorithms, and the application of the algorithms in active robotic mapping is explored.     

Two-machine group scheduling problems in discrete parts manufacturing with sequence-dependent setups

This paper focuses on minimizing the total completion time in two-machine group scheduling problems with sequence-dependent setups that are typically found in discrete parts manufacturing. As the problem is characterized as strongly NP-hard, three search algorithms based on tabu search are developed for solving industry-size scheduling problems. Four different lower bounding mechanisms are developed to identify a lower bound for all problems attempted, and the largest of the four is aptly used in the evaluation of the percentage deviation of the search algorithms to assess their efficacy. The problem sizes are classified as small, medium and large, and to accommodate the variability that might exist in the sequence-dependent setup times on both machines, three different scenarios are considered. Such finer levels of classification have resulted in the generation of nine different categories of problem instances, thus facilitating the performance of a very detailed statistical experimental design to assess the efficacy and efficiency of the three search algorithms. The search algorithm based on long-term memory with maximal frequencies either recorded a statistically better makespan or one that is indifferent when compared with the other two with all three scenarios and problem sizes. Hence, it is recommended for solving the research problem. Under the three scenarios, the average percentage deviation for all sizes of problem instances solved has been remarkably low. In particular, a mathematical programming based lower bounding mechanism, which focuses on converting (reducing) the original sequence-dependent group scheduling problem with several jobs in each group to a sequence-dependent job scheduling problem, has served well in identifying a high quality lower bound for the original problem, making it possible to evaluate a lower average percentage deviation for the search algorithm. Also, a 16–17-fold reduction in average computation time for solving a large problem instance with the recommended search algorithm compared with identifying just the lower bound of (not solving) the same instance by the mathematical programming based mechanism speaks strongly in favor of the search algorithm for solving industry-size group scheduling problems.     

基于小样本学习的图像分类技术综述

图像分类的应用场景非常广泛,很多场景下难以收集到足够多的数据来训练模型,利用小样本学习进行图像分类可解决训练数据量小的问题.本文对近年来的小样本图像分类算法进行了详细综述,根据不同的建模方式,将现有算法分为卷积神经网络模型和图神经网络模型两大类,其中基于卷积神经网络模型的算法包括四种学习范式:迁移学习、元学习、对偶学习...     

融合多策略的增强麻雀搜索算法及其应用

针对麻雀搜索算法（SSA）易陷入局部最优和寻优精度低等问题,提出一种融合多策略的增强麻雀搜索算法（ESSA）。首先,在发现者飞行位置引入莱维飞行和云自适应权重,以扩大算法搜索范围并丰富其种群多样性;其次,通过基于模糊控制的自适应透镜成像策略对当前最优位置进行反向学习,以增强算法跳出局部最优的能力;最后选用CEC2017中的12个函数作为测试集,将ESSA和标准SSA,以及其他四种改进麻雀算法（ISSA、MSSSA、HSSA、SHSSA）进行性能测试。实验结果表明ESSA能够获得更好的搜索性能。将ESSA算法应用于三维无人机路径规划问题,仿真结果表明ESSA在无人机三维路径寻优上也能获取最优的结果。     

基于改进多目标骨干粒子群算法的电力系统环境经济调度

针对粒子群优化算法种群多样性不足、易陷入局部寻优的问题,提出一种基于改进多目标骨干粒子群优化算法(improved bare-bones multi-objective particle swarm optimization,IBBMOPSO)的电力系统环境经济调度的求解方法.IBBMOPSO采用一种搜索权重非线性递减...