In this dissertation, based on the high performance units and system of humanoid robot, the new method of visual data processing and arm motion planning is proposed aiming at the task of robot playing pingpang. In order to develop and deploy the key unit of high-speed visual perception system, an optimized model of flighting pingpang is constructed. And also, an effective robot arm hitting strategy is investigated based on predicted data.The main contributions of this dissertation are summarized as follows:Firstly, a pingpong trajectory prediction method is proposed on a three-dimensional flight model and a collision model based on the high-speed visual perception system.Then, a motion model that adopts intelligent training method is built and deployed. In the model, the model parameter is constructed by visual data; and a new two-level learning hierarchy for designing RBF networks is proposed, which combines an integrated algorithm (ROLS+D-opt) and electromagnetism-like mechanism optimization algorithm, to learn the key parameters of the pingpong motion model.Finally, a hitting strategy of pingpong playing is proposed. Based on optimization model, the planning algorithm completed the interphase between the visual system and the arm unit, and verified its effectiveness by simulation and humanoid robot. The simulation and the practical pingpong playing both demonstrate effectiveness of the proposed algorithm. |