基于神经网络的移动机器人路径规划方法研究

发布时间：2018-05-19 22:09

本文选题：机器人 + 路径规划　；参考：《哈尔滨工业大学》2017年硕士论文

【摘要】：随着科学技术的进步,机器人学得到了长足的发展,机器人可以将人类从繁重的重复劳动中解脱出来,从工业领域到大众生活,机器人发挥着越来越重要的作用。路径规划是机器人学的核心内容之一,得到了众多学者的深入研究,具有非常重要的意义,本文主要研究了一种基于神经网络和混合粒子群算法相结合的移动机器人路径规划方法。在环境信息表示方面,研究了多层前向网络、hopfield神经网络、ART神经网络等在机器人路径规划方面的应用,由于多层前向网络在表示障碍物信息时计算简单,易于并行,并且无需训练权值,结合未知环境下路径规划的特点,最后确定用多层前向网络表示环境信息。课题使用了一种混合粒子群算法DHPSO进行子路径规划。针对惯性权重随迭代次数递减的标准粒子群算法(SPSO)全局收敛性强但是收敛速度慢和压缩因子粒子群算法(PSOCF)全局收敛性弱但是收敛速度很快的特点,提出了一种双种群混合交叉粒子群算法DHPSO,种群一和种群二分别使用SPSO和PSOCF的方法进行迭代,每隔一定的迭代次数,种群一将自身的较优粒子交换给种群二。DHPSO结合了SPSO和PSOCF的优点,不仅全局收敛性较强,同时具有很快的收敛速度。最后,在MATLABR2016a实验平台上对粒子群算法和路径规划进行了仿真实验,在单模态函数和多模态函数下的仿真实验证明了DHPSO算法的优越性。在路径规划的实验仿真方面,进行了多种环境下的路径规划,包括简单和复杂静态环境下的路径规划、环境中存在动态障碍物情况下的路径规划以及二维编码情况下的路径规划。同时进行了路径规划方面的几点思考,包括评价函数的选择、坐标转换以及粒子群算法在路径规划方面的一些需要留意的地方等。路径规划的仿真实验结果说明了总的路径规划的有效性,具有一定的实用价值。
[Abstract]:With the progress of science and technology, robotics has made great progress. Robot can extricate mankind from the heavy repeated work, from the industrial field to public life, robot plays an increasingly important role. Path planning is one of the core contents of robotics and has been deeply studied by many scholars. In this paper, a path planning method for mobile robot based on neural network and hybrid particle swarm optimization is studied. In the field of environmental information representation, the application of multi-layer forward network (ART) neural network in robot path planning is studied. Because the multi-layer forward network is simple to compute when representing obstacle information, it is easy to parallel. And without training weights, combined with the characteristics of path planning in unknown environment, the multi-layer forward network is used to represent the environment information. In this paper, a hybrid particle swarm optimization (DHPSO) algorithm is used for subpath planning. For the standard particle swarm optimization (SPSO) algorithm with decreasing inertial weight with the iterative times, the global convergence is strong but the convergence rate is slow, and the compression factor particle swarm optimization algorithm has the characteristics of weak global convergence and fast convergence rate. In this paper, a two-population hybrid crossover particle swarm optimization (DHPSO) algorithm is proposed. Population one and population two are iterated by SPSO and PSOCF, respectively, every certain number of iterations. Population one exchanges its own better particles with population II. DHPSO combines the advantages of SPSO and PSOCF. It not only has strong global convergence, but also has a fast convergence rate. Finally, the particle swarm optimization (PSO) algorithm and path planning are simulated on the MATLABR2016a platform, and the superiority of the DHPSO algorithm is proved by the simulation experiments under the single-mode function and the multi-mode function. In the aspect of experimental simulation of path planning, path planning is carried out in many environments, including simple and complex static environment. Path planning in the presence of dynamic obstacles and in two-dimensional coding. At the same time, some thoughts on path planning are given, including the selection of evaluation function, coordinate transformation and some points needing attention in the aspect of path planning based on particle swarm optimization. The simulation results of path planning show that the overall path planning is effective and has some practical value.
【学位授予单位】：哈尔滨工业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP242;TP183

【参考文献】