论文部分内容阅读
对撞球机器人的母球控制问题展开研究,设计了一种基于神经网络(NN)的控制器,使机器人能够控制母球在击打目标球后按照预定的模式运动至目标点——即完成走位。针对该问题非线性且非光滑的特点,对坐标系进行阐述并给出机器人击球的模型;在光滑的假设下使用理论分析的方法建立母球的运动学模型与边库反弹的理想镜像模型;进而使用神经网络方法对理想模型进行修正,并对不同的轨迹模式进行分析与分类。测试结果表明:经过训练的机器人能够掌握各种模式的走位,统计结果与模型分析结果相吻合;相比于单一使用神经网络方法,本文使用理论分析与神经网络相结合的方法能够有效地提升网络的品质,降低训练的误差。
The research on the cue ball control problem of the cue ball robot is based on a neural network (NN) controller that enables the robot to control the cue ball to hit the target ball and then move to the target point according to a predetermined pattern - that is to say, Bit. Aiming at the non-linear and non-smooth feature of the problem, the coordinate system is expounded and the model of robotic shot is given. Using the theoretical analysis, the ideal mirror model of the kinematic model of the ball and the bouncing ; Then using neural network method to correct the ideal model, and different trajectory patterns for analysis and classification. The test results show that the trained robots can grasp the walk of various modes and the statistical results are in good agreement with the model analysis results. Compared with the single-use neural network method, the method combining theoretical analysis with neural network can effectively improve Network quality, reduce training errors.