Exploring Deep Learning for Autonomous Kinetic Art

Autonomous kinetic art is a new interdisciplinary field at the edge of artificial intelligence, robotics, and practice of art, where the movement has become a primary form of artistic expression. Conventional kinetic art objects depend mainly on deterministic processes or a rule-based approach, restricting them to adapting, learning and reacting in any meaningful way to dynamic conditions. The paper discusses how deep learning can be used to empower autonomous kinetic art systems with adaptive, expressive and context-aware mechanical behavior. It suggests a detailed conceptualization of kinetic art as a cyber-physical system that involves multimodal perception and learning-based artistic intelligence, generation of movement and actuation of machinery in a closed-loop structure. Different types of neural network models, convolutional, recurrent and transformer-based are studied in terms of their functions in spatial perception, temporal coherence, and long-range expressive consistency. Motion planning and adaptive control systems are based on learning to convert the abstract neural representations into realizable and expressive motion in the real world taking mechanical and safety limits into account. Experimental assessment uses both quantitative performance metrics and qualitative artistic evaluation using quantitative data (table and plot based) to confirm the smoothness of motion, responsiveness, stability, and energy efficiency of various interaction cases. The case studies of autonomous kinetic installations also illustrate that the system can support its continuous functioning, the development of behaviors and the increase in the level of audience interest. The conclusions state that deep learning has allowed transforming the paradigm of kinetic art of pre-programmed motion into adaptive, learning-driven mechanic expression, which has altered the definition of artistic authorship and pushed the limits of computational creativity and embodied artificial intelligence

Artificial neural network · Conceptualization · Deep learning · Perception · Embodied and Extended Cognition · Human Motion and Animation · Music Technology and Sound Studies