基于视觉的打乒乓球机器人仿人击球策略研究

项目名称： 基于视觉的打乒乓球机器人仿人击球策略研究

项目编号： No.61273337

项目类型： 面上项目

立项/批准年度： 2013

项目学科： 自动化技术、计算机技术

项目作者： 谭民

作者单位： 中国科学院自动化研究所

项目金额： 83万元

中文摘要： 本项目以机器人打乒乓球为背景，研究机器人学习和实现人的运动技能,提出基于视觉的机器人仿人击球策略和方法，为提高机器人的智能性提供必要的理论与技术基础。 以乒乓球、球拍与手臂为观测对象，以乒乓球的运动测量结果触发手臂和球拍的测量，以乒乓球的运动转折点区分击球运动与辅助运动。对直接测量出的球拍位姿和基于小臂和大臂推算出的位姿进行融合，再经拟合获得带时标和姿态的平滑轨迹，估计球拍运动参数。在对人的击球效果评价的基础上，以姿态的变化和轨迹的转折等作为度量，识别推挡、削球等运动技能类型。对于分类后的运动技能，采用基于运动基元或基本轨迹的方法描述球拍运动，从而提取出运动模式。根据机器人应用运动模式回球后的效果，进行运动模式修正。结合来球的击打点参数和竞技策略给出的期望出球参数，以模糊算法确定机器人球拍采用的运动模式，并转换为机器人各个关节的运动参数，进行仿人运动技能的回球，实现机器人与人的竞技运动。

中文关键词： 乒乓球机器人；机器视觉；自主学习；定点回球；仿人策略

英文摘要： Aiming at playing table tennis using a robot, this project will investigate how a robot to learn and realize human's motion skill. The vision-based humanoid striking strategy will be proposed for the playing table tennis robot. This project will provide necessary theory and technology for improving the intelligence of robots. The table tennis ball, racket and human's arm are selected as the observing objects. The measuring results of the ball's motion are used to trigger the measurement of the arm and the racket. The returning point of the ball is employed to distinguish the striking and auxiliary motions. The measured pose of the racket is fused with the pose of the racket calculated from the lower and upper arms. Then the smooth trajectory with sampling time and racket orientation is obtained via fitting the poses of the racket in different sampling moments. The motion parameters of the racket are estimated with the trajectory. Based on the assessment of the striking results, the basic motion types such as push-and-block and chop stroke are classified according to the pose variation and the trajectory turning. Each type is described with motion primitives or basic trajectories. Then the motion patterns in them are extracted and recognized. The motion patterns are modified according to the performances in the

英文关键词： table tennis robot；machine vision；self-learning；striking with fixed point；humanoid strategy