轮式足球机器人运动控制算法研究与实现

发布时间：2018-05-12 13:50

本文选题：轮式足球机器人 + 仿真　；参考：《成都理工大学》2009年硕士论文

【摘要】： 足球机器人是人工智能与机器人领域极富挑战性的高技术密集项目,同时又是人工智能技术的一个理想突破点。它涵盖了机器人学、人工智能和智能控制等多个领域,己成为研究多智能体系统和人工智能应用技术研究的重要实验平台。机器人踢足球,看似游戏,其实展示了一个国家信息和自动化技术的综合实力。作为开发足球机器人真实系统的辅助部分,仿真系统以其经济、灵活的特性一直受到人们的重视。实物的足球机器人因为控制难度大,实时性要求高和硬件的一系列问题,现在还很难做出很有技术性的动作。对于仿真比赛,由于平台近于理想,不像实物比赛易受周围环境的影响,并且平台为官方提供的统一平台,因而具有更好的客观性,更便于比赛的开展。先进的控制方法在仿真足球机器人比赛比在实物机器人比赛中更易得到应用和检验。本文正是以机器人足球比赛为背景,以The Robot Soccer Simulator为仿真平台,针对足球机器人运动控制系统进行深入研究,并进行了算法、性能和应用上的一系列改进,并可以作为其他仿真及决策系统开发的基础。本文首先对足球机器人比赛进行了回顾,分析了足球机器人关键技术、国内外研究现状、科研意义及应用。然后,介绍了仿真平台及仿真环境,并推导出足球机器人的运动学模型,为本文的后续研究提供了模型基础及平台环境。要让足球机器人实现战术策略,首先要有好的运动控制。常规PID算法在足球机器人控制中有广泛的应用,然而足球机器人控制过程机理复杂,难以确定精确的数学模型,并存在着不同程度的非线性、时变等不确定性,同时随着对机器人控制的要求进一步提高,利用常规的PID控制很难满足系统的要求。神经网络作为一门非常热门的交叉学科,以其强大的非线性映射能力、并行处理能力、自学习能力,在控制领域得到广泛的应用。文中研究分析了BP神经网络,BP神经网络是目前应用较多的一种神经网络结构,是一种性能优良的神经网络。主要研究了BP神经网络的数学理论,详细分析了几种流行的BP神经网络学算法的优缺点及其改进的BP算法。这些研究为后面机器人小车的运动控制研究做了铺垫。接着,将各种改进的BP算法与PID相结合,得出新的控制算法,并对各种算法的性能一一比较,仿真结果表明,这种改进方案与其他几种PID控制相比,超调量小、调节速度快、调整时间短,说明其具有更好的控制特性;另外,稳态误差也较小。所以,改进的BP神经网络PID控制的控制精度更高,从而会获得令人满意的效果。因此,将BP神经网络运用于PID控制中,能够有效克服经典PID控制器在被控对象具有非线性、时变不确定性和难以建立精确的数学模型时出现的参数整定不良和性能欠佳等缺陷。本文还研究了基于BP神经网络的PID控制器结构和算法,利用改进的BP神经网络对PID控制参数进行在线自整定,构造一个具有参数自整定能力、稳定的PID控制器。将这种改进的BP PID控制器应用到机器人小车到定点运动及圆周圆周运动中,通过仿真平台及MATLAB实验,从实验结果里我们看到这种新型的PID控制器在一定程度上提高了系统的鲁棒性,使小车的运动更加稳定和轨迹更加平滑,明显提高了小车的控制性能。路劲规划问题一直是足球机器人研究的热点和难点,机器人在有障碍物的情况下,寻找一条恰当的路径,能从给定起点到终点,使机器人在运动过程中能安全、无碰撞地绕过所有的障碍物。因此,在很大程度上,路径规划问题就是避障问题。路径规划成功,机器人能快速完成给定任务,但是如果失败,机器人的行动受阻,动作难以完成,甚至影响整个策略的实现,直接影响比赛结果。所以说路径规划任务在足球机器人系统中占有很重要的地位。本文设计了一种应用SVM模式识别分类技术进行机器人路径规划的方法。支持向量机是一种基于小样本统计理论的学习机,具有完备的理论基础和严格的理论体系,支持向量机是能够提高学习机的泛化能力,此外,存在全局唯一最优解。我们把障碍物分为两类,SVM在满足最慈 ?嗉涓舻奶跫鋿?产生一个非线性分类面,从而产生一个安全的平滑的路径,本文利用SVM这个性质进行路径规划研究。首先,得到一组小车阵型,将小车离散化为样本点,然后设置一些样本引导点和向导点,下一步就是寻找一条可行的路径,通过MATLAB仿真,我们将得出一条避开障碍物的路径。对于不同的障碍物模式,起始点和目标点可能处于不同的中间区域,因此当搜索步数大于某一个阈值的时候我们终止搜索。在下一步的搜索中当找不到符合条件的下一个使V 1的安全点时,也终止搜索,因为这时候在中间区域两边的障碍物的距离太小,而不能安全越过障碍物。经过多重搜索,我们得出几条路径曲线,通过比较,我们将选择路径最短、最平滑的曲线作为小车的实际路径。因此,实验仿真证明了利用SVM取得很好的效果,机器人能够寻找到一条最优路径,为实际机器人足球比赛提供了很好的理论基础。最后,对本文的工作做了总结,指出了工作的成果意义及不足,并对今后的进一步工作进行了展望。
[Abstract]:Soccer robot is a highly challenging and high technology intensive project in the field of artificial intelligence and robot. It is also an ideal breakthrough point of artificial intelligence technology. It covers many fields such as robotics, artificial intelligence and intelligent control. It has become an important experimental platform for the research of the application technology of multi-agent system and artificial intelligence. Robot playing football, seemingly game, actually shows the comprehensive strength of a country's information and automation technology.
As an auxiliary part of developing the real robot soccer system, the simulation system has attracted the attention of its people for its economic and flexible characteristics.
Real soccer robot is difficult to control because of its difficulty in control, high real-time performance and a series of hardware problems.
For the simulation competition, because the platform is close to the ideal, unlike the physical competition easily affected by the surrounding environment, and the platform is a unified platform provided by the official platform, it has better objectivity and more convenient for the competition to carry out. The advanced control method is more easily applied and tested in the simulation of soccer robot competition than in the physical robot competition.
This paper takes the robot soccer game as the background and takes the The Robot Soccer Simulator as the simulation platform to study the motion control system of the soccer robot, and carries out a series of improvements in algorithm, performance and application, and can be used as the basis for other simulation and decision making system development.
This paper first reviewed the soccer robot competition, analyzed the key technology of the soccer robot, the research status at home and abroad, the research significance and the application. Then, the simulation platform and the simulation environment were introduced, and the kinematics model of the soccer robot was deduced, which provided the model foundation and platform environment for the follow-up study of this paper.
In order to make the soccer robot realize the tactical strategy, it must have good motion control first. The conventional PID algorithm is widely used in the control of soccer robot. However, the mechanism of the control process of the soccer robot is complex and it is difficult to determine the precise mathematical model. There are different degrees of non linear, time-varying and other uncertainties, at the same time, with the control of the robot. The requirement of the system is further improved. It is difficult to satisfy the system requirements by using conventional PID control.
As a very popular interdisciplinary subject, neural network has been widely used in control field with its powerful nonlinear mapping ability, parallel processing capability and self-learning ability.
In this paper, the BP neural network is studied and analyzed, and the BP neural network is a kind of neural network structure which is widely used at present. It is a kind of neural network with excellent performance. It mainly studies the mathematical theory of BP neural network, analyzes the advantages and disadvantages of several popular BP neural network algorithms and its improved BP algorithm. These studies are the rear machines. The research on the motion control of the human car has been paved.
Then, a variety of improved BP algorithms are combined with PID to get a new control algorithm and compare the performance of all kinds of algorithms. The simulation results show that, compared with several other PID controls, the overshoot is small, the adjustment speed is fast, the adjustment time is short, and the control characteristics are better. In addition, the steady-state error is also small. The improved BP neural network PID control has higher control accuracy, and it will achieve satisfactory results.
Therefore, the application of BP neural network to PID control can effectively overcome the defects of the classical PID controller, such as the parameter setting and the poor performance, when the controlled object is nonlinear, time-varying and difficult to establish an accurate mathematical model.
In this paper, the structure and algorithm of PID controller based on BP neural network are also studied. The improved BP neural network is used to self-tuning PID control parameters online, and a PID controller with parameter self-tuning ability and stability is constructed. This improved BP PID controller is applied to the robot car to fixed point motion and circumference circle motion. Through the simulation platform and the MATLAB experiment, we see from the experimental results that this new PID controller improves the robustness of the system to a certain extent, makes the motion of the car more stable and smooth, and obviously improves the control performance of the car.
The problem of road strength planning has always been a hot and difficult point in the research of soccer robot. In the case of obstacles, the robot can find a proper path, which can make the robot safe and avoid all obstacles in the process of motion from a given starting point to the end. Therefore, to a great extent, the path planning problem is the obstacle avoidance.
If the path planning is successful, the robot can complete the given task quickly, but if the robot fails, the action of the robot is blocked, the action is difficult to complete, and even the realization of the whole strategy is affected, and the result of the game is directly affected. Therefore, the path planning task plays an important role in the soccer robot system.
In this paper, a method of robot path planning using SVM pattern recognition classification technology is designed. Support vector machine is a learning machine based on small sample statistics theory. It has complete theoretical basis and strict theoretical system. Support vector machine can improve the generalization ability of the learning machine. In addition, there is a global unique optimal solution.
We divide the obstacles into two categories, and the SVM is satisfied with the most kind. We produce a nonlinear classification surface to produce a safe and smooth path. In this paper, we use the property of SVM to study the path planning. First, we get a group of small car formations, scatter the car into a sample point, and then set some sample guide points and directions. The next step is to find a feasible path. Through MATLAB simulation, we will get a path to avoid obstacles.
For different barrier patterns, the starting point and the target point may be in different middle regions, so we terminate the search when the search step is greater than a certain threshold. In the next search, when the next secure point for V 1 is not found, the search is terminated, because the barrier on both sides of the middle area is at this time. The distance between obstructions is too small to cross the barrier safely.
After multiple searches, we get several path curves. By comparison, we will choose the shortest path and the smoothest curve as the actual path of the car.
Therefore, the experimental simulation proves that the use of SVM has achieved good results. The robot can find an optimal path and provide a good theoretical basis for the actual robot soccer game.
Finally, the work of this paper is summarized, the significance and shortcomings of the work are pointed out, and the further work in the future is prospected.

【学位授予单位】：成都理工大学
【学位级别】：硕士
【学位授予年份】：2009
【分类号】：TP242

【引证文献】