AI Breakthrough: Robots Emulate Human DrawingRobotic Artistry: Humanoid Robots Learn to Sketch

Advancements in the field of deep learning and the development of generative models have propelled the generation of art through artificial intelligence, producing works that are increasingly compelling. Nonetheless, it's predominantly software algorithms, rather than physical entities, that are behind the creation of such AI-driven art.

Scientists from Universidad Complutense de Madrid (UCM) and Universidad Carlos III de Madrid (UC3M) have recently advanced a model based on deep learning that empowers humanoid robots to perform sketching tasks akin to those of human artists. Their research, documented in Cognitive Systems Research, demonstrates the potential for robots to partake in the artistic creation process actively.

Raúl Fernandez-Fernandez, one of the study's co-authors, explained to Tech Xplore the team's ambition to develop a robotic application that would intrigue both the academic community and the wider public. They conceptualized the idea of a robot undertaking artistic activities as a novel and striking endeavor, leading to the creation of an art-producing humanoid robot.

While most robotic systems designed for drawing or painting mimic the functionality of printers by replicating pre-designed images, the team aimed to differentiate by developing a robot capable of using deep reinforcement learning to sketch images stroke by stroke in a manner reminiscent of human drawing techniques.

  • "Our objective wasn't merely to engineer a robot capable of generating complex artworks, but to build a sophisticated physical robot painter," said Fernandez-Fernandez. "Our focus was on enhancing the robot's control mechanisms used in painting applications."

  • Over recent years, Fernandez-Fernandez and his team have been exploring sophisticated and efficient algorithms to orchestrate the actions of robots in creative tasks. Their latest publication integrates promising approaches from their previous research, marking a significant step in robotic creativity.

  • "This endeavor draws inspiration from two significant previous works," Fernandez-Fernandez stated. "One of our earlier research efforts investigated utilizing the Quick Draw! Dataset for training robotic painters. Another introduced Deep-Q-Learning for executing complex movement trajectories, including the expression of emotions through art."

The innovative sketching system developed by the researchers relies on a Deep-Q-Learning framework, initially presented in another study. This framework has been refined to enable robots to execute intricate manual tasks across various settings.

"The system's neural network is segmented into three interconnected networks," elaborated Fernandez-Fernandez. "A global network identifies high-level canvas features, a local network pinpoints detailed features near the painting area, and an output network generates subsequent painting positions based on these identified features."

The model is further informed by additional channels that convey information about distance and painting tools, enhancing its ability to sketch with precision. To refine the robot's painting skills to mimic human-like artistry more closely, a pre-training phase utilizing a random stroke generator was incorporated.

"We adopted double Q-learning to mitigate overestimation issues and devised a custom reward function for training," Fernandez-Fernandez added. "Moreover, we integrated a sketch classification network to identify high-level sketch features, using its output as a reward during the final stages of a painting epoch, thus allowing for greater painting flexibility."

The challenge of translating AI-generated image positions to a real-world canvas was addressed by creating a discretized virtual space within the physical canvas, enabling the robot to accurately translate model-provided painting positions.

"A key achievement of our work is the integration of advanced control algorithms within a real robot painting application," concluded Fernandez-Fernandez. "This demonstrates significant improvements in robot control for painting applications, paving the way for original and sophisticated uses beyond traditional problems."

Ann Castro
Ann Castro Author
Ann Castro carries a total of 7 years experience in the healthcare domain. She owns a Master’s of Medicine Degree. She bagged numerous awards by contributing in the medical field with her ground-breaking notions. Ann has developed her own style of working and known for accuracy in her work. She loves trekking. She visits new places whenever she gets free time.