Virtual table tennis ball player hitting training method based on reinforcement learning

Abstract

The invention discloses a virtual table tennis ball player hitting training method based on reinforcement learning. The method comprises the following steps of 1) designing a task scene and a task flow; 2) training a ball hitting strategy of a ball bat by using the reinforcement learning method; 3) estimating a motion condition of each joint during ball hitting of a human body by using an inversekinematics algorithm; and 4) training a movement strategy of a root node through the reinforcement learning. According to the method, a virtual ball player who can carry out ball hitting with a reasonable posture and high accuracy can be obtained by designing a simple reward function without training data; complicated ball hitting rules do not need to be designed; meanwhile, due to the low-consumption characteristic of the forward operation of the reinforcement learning, a ball hitting action of the virtual ball player can keep the high frame rate stably, and a user has good interactive experience.

Description

technical field [0001] The invention relates to the fields of virtual reality and reinforcement learning, in particular to a reinforcement learning-based virtual table tennis player batting training method. Background technique [0002] Virtual reality has always been one of the key research topics in the computer field. In recent years, with the advent and development of virtual reality devices such as HTCVive and Oculus, the development of virtual reality technology has also reached a new height, and the applications of virtual reality emerge in endlessly. At present , virtual reality has been widely used in military, education, entertainment and other fields. With the low cost and civilianization of virtual reality equipment, people have more and more contacts with virtual reality applications, and the degree of contact is getting deeper and deeper. The requirements for the quality of virtual reality applications are also getting higher and higher. People not only hope W...

AI Technical Summary

Problems solved by technology

[0004] In the task of designing a virtual table tennis player, the rule-based method needs to design complex hitting rules, which is difficult to design and has high computing costs, while the method based on imitation learning or supervised learning needs to collect training data for training, and the training cost is relatively high. high

Images