Simulating Early Childhood Drawing Behaviors under Physical Constraints Using Reinforcement Learning

The ability to draw is a skill that children naturally develop over time. The interaction between a child's body and the environment plays a crucial role in this developmental process, yet there has been limited research exploring how physical properties of these interactions influence drawing development. We address this issue through a simulation study using a novel system: a reinforcement learning drawing agent within a simulated physical environment of a pen and a canvas. We demonstrate that acquired drawing behaviors can diverge based solely on the difference in physical configurations, particularly viscous resistance and that drawing patterns resembling those of young children can emerge. Also, we show that appropriate physical setups are important for an agent to efficiently learn drawing. These findings imply that when discussing how children's drawing behaviors develop, it is preferable to take the physical characteristics of the drawing environment into account. Moreover, our proposed system can be a new tool to investigate the development of children's drawing behaviors.