Generative Neural Models in Robotics: Pioneering the Embodied AI Revolution

The concept of machines with the capability for self-reproduction and the creation of useful artifacts has intrigued thinkers like John von Neumann, whose seminal contributions continue to influence contemporary technology. Yet, the modern discourse has evolved, emphasizing machines' autonomous performance of beneficial tasks over self-replication. This shift is markedly noticeable in domains such as transportation, elderly care, household management, and manufacturing, where the past decade has witnessed a gradual but consistent advancement. This progress is characterized by innovations in specialized tasks and tools, which, however, lack integration into a cohesive universal framework. Conversely, the growth in AI fields not inherently tied to physical interaction has seen an exponential rise, fueled by significant increases in the resources allocated for developing neural models. This surge has led to groundbreaking commercial innovations in textual, visual, and auditory content generation becoming widely accessible in remarkably short times. Yet, the question remains: How do these advances in generative AI translate to the physical realm, particularly in enhancing robotic capabilities? My presentation delves into this inquiry, analyzing it through the prism of cutting-edge research and emerging trends, aiming to shed light on how generative neural models are spearheading the revolution in Embodied AI.