Reinforcement Learning Based 3-D Surface Calligraphy Robot

We present a robotic calligraphy system designed for writing on complex curved surfaces, addressing the growing interest in applying robotic techniques to irregular substrates. Our system integrates a six-axis industrial robotic arm and a depth camera, enabling it to reproduce 2D target calligraphy images onto various 3D surfaces. By framing the writing process as an interactive task between the robot and its environment, we incorporate a reinforcement learning agent to optimize the writing strategy, ensuring generalized, distortion-free calligraphy across diverse surfaces. We specifically designed a reward function suited to the curved surface writing task. For the essential 2D-to-3D mapping, we employ the Least Squares Conformal Maps (LSCM) algorithm to minimize distortion. Experimental results on multiple curved surfaces demonstrate that the system achieves aesthetically pleasing calligraphy, highlighting its potential for advanced robotic applications in art and design.