飞行物体轨迹预测与拦截回击对军事、体育、工业等均具有重要研究意义和应用价值。但飞行中的旋转特性为问题研究带来了挑战。本项目以旋转飞行乒乓球为研究对象，开展飞行物体状态估计、轨迹预测和拦截回击等算法研究。当前研究通常采用黑箱或者简化的离散模型，本项目从构建旋转观测视觉系统入手，通过对目标表面自然标识的识别定位实现基于观测的实时旋转状态估计；结合受力分析和旋转观测，推导旋转飞行模型连续形式，以解决传统离散模型不能用于模型参数反馈学习及周期迭代计算效率低、预测不准确的问题；在现有静态球桌碰撞模型研究的基础上，推导旋转球和运动球拍的碰撞模型；采用概率神经网络辨识模型参数，并提出基于模型的旋转状态估计和自适应运动状态滤波方法，实现对旋转乒乓球的精确轨迹预测；通过问题建模和优化求解方法研究，实现旋转乒乓球回球规划；最后在乒乓球仿人机器人上验证技术的可行性与有效性，实现旋转乒乓球对打演示，并推广应用。

Trajectory prediction, interception and hitting back of flying objects have significant research and promotion value in many fields such as military, sport, industry etc. The study is even more challenging when the flying object is spinning on its own. In this project we use Ping-Pong robots as the platform to study algorithms for flying state estimation, trajectory prediction and interception planning of the spinning ball. Previous studies on flying trajectory prediction of objects are generally based on black-box model or simplified discrete model. In this proposal, first a new vision system will be proposed to observe and estimate the spinning accurately and directly by recognizing the natural marks on the target surface. Then the continuous form of the dynamics model can be deduced based on force analysis and spin observation, while traditional discrete model cannot provide feedback information for model parameters learning and lack of effectiveness and accuracy for trajectory prediction. Basing on existing modeling method for ball-table collision, collision model between the spinning ball and the moving racket will also be derived. All model parameters will be learned using probabilistic neural network. And the learned results can be applied in new algorithms for spin state estimation based on the dynamics model and move state estimation based on the adaptive filter, which help to achieve precise prediction of the spinning ball's flight trajectory. Then the motion planning algorithm of the robot to intercept and hit back the Ping-Pong ball will be studied using optimization approach. All techniques and algorithms will be applied on a real humanoid robot to play spinning Ping-Pong ball to verify their practicability and effectiveness. More than play spinning ball flexibly, we will also try to promote our techniques to other tasks and domains.