中国科学院自动化研究所机构知识库

In this paper, we present an active learning enabled robotic painting system, called ShadowPainter, which acquires artist-specific painting information from the artwork creating process and achieves robotic reproduction of the artwork. The artist's painting process information, including interactive trajectories of paintbrushes with the environment and states of the canvas, is collected by a novel Visual Measurement System (VMS). A Robotic Painting System (RPS), accompanied by the VSM, is developed to reproduce human paintings by actively imitating the measured painting process. The critical factors that influence the final painting performance of the robot are revealed. At the end of this paper, the reproduced artworks and the painting ability of the RPS are evaluated by local and global criteria and metrics. The experimental results show that our ShadowPainter can reproduce human-level brush strokes, painting techniques, and overall paintings. Compared with the existing work, our system produces natural strokes and painting details that are closer to human artworks.