Graduation Semester and Year
Fall 2025
Language
English
Document Type
Thesis
Degree Name
Master of Science in Computer Science
Department
Computer Science and Engineering
First Advisor
Dr. Diego Patiño
Second Advisor
Dr. Cesar Torres
Third Advisor
Dr. Alex Dillhoff
Abstract
Most computational art systems rely on generative models that produce a complete artwork in a single pass, without capturing the gradual, decision-driven process through which human artists construct visual pieces. Prior research in sequential, stroke-based image generation, including differentiable neural painters and model-based reinforcement learning agents, has explored step-by-step creation, but these systems typically aim to reconstruct the input image within the same visual representation space, closely matching brushstrokes, textures, or colors to the target. In contrast, this thesis investigates sequential art creation in a different artistic representation, where the final artwork does not share the same visual form as the input image. A Deep Deterministic Policy Gradient (DDPG) framework with CNN-based actor and critic networks is developed to study how an artificial agent learns to “paint” through thousands of incremental actions within a custom canvas environment. A central contribution of this work is a systematic analysis of reward function design, identified as the primary challenge in enabling coherent artistic progress. Through extensive evaluation of pixel-based, structural, perceptual, and embedding-based reward families, the results show that misaligned rewards lead to instability, policy collapse, or non-artistic behavior, while carefully shaped intermediate rewards are essential for guiding step-by-step creation. Overall, the study demonstrates that while DDPG provides a viable architecture for progressive art generation, effective reward engineering is the decisive factor in enabling agents to construct artwork deliberately over time.
Keywords
Reinforcement Learning, Artificial Intelligence, Large Language Models, AI in Art, Computational Creativity, Deep Deterministic Policy Gradient, Reward Function Design, Stroke Based Rendering, Policy Stability, String Art
Disciplines
Artificial Intelligence and Robotics | Graphics and Human Computer Interfaces | Painting
License
This work is licensed under a Creative Commons Attribution 4.0 International License.
Recommended Citation
Pothula, Asmin, "A Deep Reinforcement Learning Framework for Sequential Art Creation" (2025). Computer Science and Engineering Theses. 539.
https://mavmatrix.uta.edu/cse_theses/539
Included in
Artificial Intelligence and Robotics Commons, Graphics and Human Computer Interfaces Commons, Painting Commons