Enhancing Robotic Sketching with a Deep Learning Framework

A deep-learning framework to improve the capabilities of a robot sketching agent
Deep learning algorithms have shown remarkable results across a wide range of disciplines, including the arts. Many computer scientists have developed models which can produce artistic works such as poems, sketches and paintings.

Researchers from Seoul National University recently developed a new framework for deep artistic learning that is intended to improve the capabilities of a robot sketcher. The framework was introduced in a pre-published paper on arXiv and presented at ICRA2022. It allows a robot to simultaneously learn stroke-based rendering as well as motor control.

The primary motivation behind our research was the desire to create something cool using non-rule based mechanisms, such as deep-learning. We thought that drawing would be cool to demonstrate if it were a robot performing the drawing instead of a human,\” Ganghun Lee told TechXplore. Recent deep learning techniques in the artistic field have produced some amazing results, but they are mostly generative models that produce whole pixel outputs at once.