Solving Robotic Trajectory Sequential Writing Problem via Learning Character's Structural and Sequential Information.

Area of Science:

• Robotics
• Artificial Intelligence
• Computer Vision

Background:

• Robotic calligraphy faces challenges in mimicking human writing aesthetics and intentions.
• Existing systems struggle with learning writing sequences from limited data.

Purpose of the Study:

• To develop a robotic calligraphy system capable of learning writing sequences with minimal data.
• To produce diverse and aesthetically pleasing calligraphy outputs comparable to human writing.

Main Methods:

• Utilized a gated recurrent unit (GRU) network for generating robotic writing actions.
• Incorporated a prelabeled trajectory sequence vector for enhanced learning.
• Employed a swarm optimization algorithm to optimize system parameters.
• Developed a novel evaluation method assessing shape, trajectory, and structural information.

Main Results:

• The system demonstrated competitive performance against state-of-the-art methods across multiple metrics (FID, MAE, PSNR, SSIM, PerLoss).
• Achieved high diversity in writing outcomes, measured by variance and entropy.
• Successfully learned from a small dataset, producing aesthetically pleasing and varied calligraphy.

Conclusions:

• The proposed GRU-based robotic motion planning system, enhanced by swarm optimization, effectively learns from limited data.
• The system generates diverse and aesthetically superior calligraphy, addressing key challenges in robotic writing.