Font Size: a A A

Research On Robotic Generation Method Of Chinese Calligraphy Strokes Based On Generative Adversarial Networks And Inverse Reinforcement Learning

Posted on:2019-11-13Degree:MasterType:Thesis
Country:ChinaCandidate:J T LvFull Text:PDF
GTID:2428330545997909Subject:Computer technology
Abstract/Summary:PDF Full Text Request
As an important research field in which artificial intelligence and automated control are combined,intelligent robots are making rapid progress.Robotic Chinese calligraphy is an important application of intelligent robot technology.It is a highly demanding task for artistic creativity,control complexity and operation coordination.It can not only promote Chinese traditional culture,but also play an important role in robots achieve other complex actions which need creative creativity abilities.This paper aims to improve the level of writing of Chinese characters in robots,increase writing diversity,reduce the dependence of writing process on human engineers,and use generative adversarial networks and reinforcement learning techniques to enable robots to realize the diversity of strokes of Chinese characters according to sample fonts and human aesthetic preferences.The paper firstly constructs a robot Chinese stroke writing frame based on the generative adversarial networks.It uses a policy gradient method based on reinforcement learning to train the writing module.Through self-adversarial training,the robot learns writing strokes on the one hand and learning evaluation on the other hand.Robot learns stroke styles from the database of stroke samples,realizes self-learning of strokes writing action in Chinese characters according to stroke sample results.Based on this,in a simple human-computer interaction method,a method based on inverse reinforcement learning was used to construct a robot Chinese stroke writing method based on what human preferences,so that the ultimate writing style of the robot was freed from the limitations of the data sample,by learning human aesthetic preferences.The aesthetic preferences achieve a high level of writing of complex Chinese strokes and the effect of human participants on the control of robot writing results.Experiments show that the method constructed in this paper completes the task of the robot autonomously learning Chinese characters strokes.Through the study of stroke samples and human aesthetic preferences,the robot can write high-level strokes of Chinese characters,and the writing effects are diverse and stylized.In the future work,it is possible to construct a method of learning in the shape of a Chinese character,to further improve the robot's ability to learn,and to achieve the writing of a complete Chinese character.
Keywords/Search Tags:Robotic Calligraphy, Generative adversarial networks(GAN), Inverse Reinforcement Learning(IRL)
PDF Full Text Request
Related items