【作者】 吴宪祥；

【导师】 郭宝龙；

【作者基本信息】 西安电子科技大学 ， 电路与系统， 2009， 博士

【摘要】 近年来飞速发展的机器人足球比赛系统为人工智能理论的研究提供了一个标准的实验平台。MiroSot是目前开展最为广泛的集控式机器人足球比赛系统之一,一般包括视觉、决策、通信和足球机器人四个子系统,其研究目标是快速准确获取赛场态势并给出合理的决策。相关技术涵盖了机器人学、计算机视觉、传感器融合、实时推理、运动规划与运动控制、无线通信、机器学习、自治智能体和多智能体协作等多个研究领域,引起了各国学者的高度重视。本文重点研究了对MiroSot系统高层决策至关重要的微型足球机器人快速鲁棒位姿辨识技术,并结合近年来新兴的粒子群优化算法研究了群智能路径规划技术。本文的主要研究内容如下:(1)研究了MiroSot系统的体系结构、工作原理及其视觉、决策、通信和足球机器人四个子系统的功能与硬件设计,给出了移动机器人路径规划的框架。分析了移动机器人路径规划的问题描述和特点,对传统的移动机器人路径规划算法和新兴的智能路径规划算法进行了总结,比较了各种路径规划算法的优点和不足,探讨了路径规划技术进一步研究的方向。(2)提出了一种基于较短轴补偿逼近的微型足球机器人位姿辨识算法(SASA)。根据微型足球机器人设计队标色块的对称性特点,提出了一种基于较短轴分割的微型足球机器人色标分块方案,在此基础上给出了基于较短轴补偿逼近的微型足球机器人位姿辨识算法SASA。实验结果表明,SASA算法有效减少了目标机器人位姿辨识的计算量,改进了位姿辨识的计算速度和准确度,提高了系统的实时性。(3)提出了一种利用相位相关技术进行MiroSot系统微型足球机器人位姿辨识算法(PCGR),并构造了八边形对数极坐标傅里叶变换算法(OLPFFT),提高了运算速度和精度。将分割得到的机器人目标图像和参考图像进行快速离散傅里叶变换后,转换到对数极坐标系下,将笛卡儿坐标空间中图像的旋转和缩放转化为对数极坐标空间中图像的二维平移,进而采用相位相关法得到小车的朝向角。为了提高对数极坐标傅里叶变换的运算速度和精度,构造了一种八边形对数极坐标网格来逼近对数极坐标网格,并给出了八边形对数极坐标网格上的快速傅里叶变换算法(Octa-Log-Polar Fourier Transform, OLPFFT)。实验结果表明,PCGR算法精度高,鲁棒性好。(4)提出了基于Lotka-Volterra模型的双群协同竞争粒子群优化算法(LVPSO)。最优路径规划问题的本质是优化计算,在LVPSO算法中,针对粒子群优化算法易于出现早熟收敛的问题,借鉴种群生态学中著名的Lotka-Volterra双群协同竞争模型,给出了两种种群协同竞争方案,通过群内和群间竞争增加粒子多样性,提高种群摆脱局部极值的能力。对5个典型基准测试函数进行优化实验表明,LVPSO算法在收敛速度和优化精度方面均有非常好的表现。(5)提出了一种基于LVPSO和Ferguson样条的MiroSot机器人路径规划算法(LVPSOFS)。利用三次Ferguson样条描述移动机器人路径,将路径规划问题转化为三次样条曲线的参数优化问题,借助LVPSO进行路径优化。实验结果表明,LVPSOFS算法可以有效地实现障碍环境下机器人的无碰撞路径规划,并且实现路径优化,规划路径平滑、合理,利于机器人的运动控制,符合人工规划的意图。更多还原

【Abstract】 Robot soccer system, one of the hotspots in the field of artificial intelligence (AI)in recent years, is a typical multi-agent system and it provides a standard testingplatform for evaluation of Artificial Intelligence theories and algorithms. FIRA(Federation of International Robot-soccer Association) MiroSot (Micro Robot WorldCup Soccer Tournament) is one of the most popular robot soccer tournaments. MiroSotis a kind of centralized control robot soccer systems which includes four subsystems toget the field situation rapidly and accurately and make reasonable decision, such as:vision subsystem, decision-making subsystem, communication subsystem and robotsubsystem. Many researchers have devoted to this project because it is involved inmany research fields, such as robotics, computer vision, sensor-fusion, real-timereasoning, path planning and motion control, wireless communication, machine learning,autonomous agents, multi-agent collaboration, and so on. Rapid robust gesturerecognition algorithm and swarm-intelligent path planning algorithm are well studied inthis dissertation for their importance to high decision-making.The main contributions of this dissertation are summarized as follows:(1) The system architecture and working principle of MiroSot, and the structure,fucction and hardware design of all four MiroSot subsystems, including visionsubsystem, decision-making subsystem, communication subsystem and robotsubsystem, are introduced respectively. A common framework of mobile robotpath planning is proposed. The description and characteristics of mobile robot pathplanning problem, the traditional methods and the up-to-date methods areintroduced, and the advantages and disadvantages of these algorithms are analyzed.Furthermore, the trend of mobile robot path planning is described.(2) A color tag recognition algorithm based on the shorter axis segmentation andsuccessive approximation technique (SASA) is proposed. By analyzing thesymmetry of color tag, a shorter-axis based successive approximation algorithm isproposed. The experiment results show that the proposed gesture recognitionalgorithm is of lower computing complexity and higher recognition accuracy.(3) A novel phase-correlation based gesture recognition algorithm (PCGR) in MiroSot is proposed,and the Octa-log-polar Fourier transform (OLPFT) is presented to improve the computingspeed and precision. In the proposed algorithm, the object image and the referenced image are transformed into log-polar coordinate space after FFT, by which rotation and scaling inCartesian coordinate space can be reduced to 2-D translation in log-polar coordinate space.The orientation of the robots can be estimated by phase correlation technique. Octa-log-polarFourier transform (OLPFT) is proposed to estimate the log-polar DFT. The OLPFT estimatesthe DFT on octa-log-polar grid, which is geometrically more similar to the log-polar grid thanpseudo-log-polar grid which can lead to higher accuracy. Experimental results show that theproposed algorithm has high precision and robustness.(4) A novel Particle Swarm Optimization Algorithm inspired by Lotka-Volterra Model(LVPSO) is proposed in this paper to avoid the premature convergence problem.Path planning is an optimistic computation problem essentially. The famousLotka-Volterra Model in Ecology is introduced into basic particle swarmoptimization algorithm. Two different cooperative-competitive schemes have beendiscussed. The diversity of particles is increased by intraspecific and interspecificcompetition, and the ability for particles to break away from the local extremum isimproved remarkably. The experimental results show that the proposed LVPSOalgorithm can converge in higher speed and higher precision by optimizing fivetypical benchmark functions.(5) A novel path planning approach using LVPSO and Ferguson splines (LVPSOFS)was proposed to get an optimal smooth path for a micro soccer robot. In theproposed algorithm, a path is described as a string of Ferguson splines. The pathplanning is then equivalent to optimization of parameters of particular cubicFerguson splines. The proposed Particle Swarm Optimization Algorithm inspiredby Lotka-Volterra Model (LVPSO) was introduced to optimize the path for its fastconvergence and global search character. Experimental results prove therationality and practicability of the proposed algorithm, which can getconvergence rapidly, with a collision-avoiding smooth optimal path being plannedfleetly.更多还原