An Interactive Spatial Structure Study Using Machine Learning

Beyond generalized assertions and rigid definitions of location-independent knowledge work, nomadic workers are not bound to traditional office space. A pattern of flexible spatial needs emerges, which they seek to facilitate their lifestyle. How can architectural spaces be designed to adapt and transform dynamically like machines in response to their surroundings, particularly to benefit digital nomads? We propose a novel design concept where architectural spaces employ reinforcement learning techniques to respond dynamically to their environment. We developed a Unity-based simulator to explore this concept, using reinforcement learning to acquire adaptive spatial movement and configuration policies. Additionally, a prototype robot was integrated to validate our approach. The results demonstrate that machine learning-driven training of spatial patterns can foster specific interactions between individuals and their environment. This engagement has the potential to alleviate the challenges and hardships faced by digital nomads, making their living and working spaces more adaptable and responsive.