基于生成对抗网络与长短时记忆网络的机器人书法系统

打开文本图片集

中图分类号：TP393 文献标识码：A文章编号：1673-9868（2025）07-0231-14

Abstract： Machine calligraphy，as one of crucial robotic manipulator applications in industrial manufacturing，faces significant challenges. Its active writing mechanism requires extensive training datasets containing information of writing trajectory sequences，and manual annotation of these data is a laborious task.

To address this issue，this paper proposes a machine callgraphy writing system based on Generative Adversarial Networks （GAN） and Long Short-Term Memory Networks （LSTM）. The writing system converts Chinese character stroke images into trajectory sequences without using stroke trajectory coding information，overcoming the problem of missng traditional writing trajectory information. Specifically，a GAN architecture was initially constructed，in which，LSTM networks was combined with a discriminative network to reduce the scale of the training dataset. Subsequently，the LSTM network generated new trajectory points gradually through multiple cycles，allwing the robot to progressively complete the entire Chinese calligraphy writing process. Finally，a discriminative network was employed to evaluate the output of the LSTM network to assst the robot finding the optimal strategy. Reinforcement learning algorithm was introduced to further enhance system performance. Experimental results demonstrate that the proposed system can efficiently produce high-quality and aesthetically pleasing Chinese caligraphy.

Key words： generative adversarial networks; long short-term memory networks; reinforcement learning algorithm;Chinese calligraphy；robot writing system

汉字书法，亦称书法，是一种具有魅力且悠久的汉字艺术表现形式。（剩余12004字）