부산대학교 도서관

Learning from Demonstration (LfD) is a widely utilized technology within the realm of robotics, and the task of writing holds particular significance in this context. Typically, algorithms for learning alphabet writing necessitate a demonstrated trajectory to acquire the requisite skills, thereby relying on sensors to record these trajectories. However, this approach introduces complexities when dealing with the composition of English words or sentences, as it requires the manual specification of starting and ending points for each individual letter. This research introduces an innovative methodology aimed at resolving this predicament, effectively obviating the necessity for physical demonstrations and the explicit designation of starting and ending points for replication. Instead, the proposed method entails the generation of binary images, followed by the extraction of skeleton curves and graph nodes. Subsequently, an iterative process is employed to ensure the absence of intermediate nodes within the extracted trajectories, thereby adeptly encapsulating the writing skill intrinsic to each alphabet letter. When reproducing the writing, the same technique is applied to process newly generated images, designating the existing nodes as the sequential starting and ending points from left to right. Through the adoption of this approach, the aggregation of distinct skills can be seamlessly realized. The efficacy of the proposed algorithm is substantiated through a comprehensive validation process encompassing both simulations and real-world experiments. This robust validation underscores the algorithm's proficiency in addressing the complexities associated with skill acquisition and reproduction, offering a promising avenue for advancements in robotic writing capabilities.