Mutual human-robot understanding for a robot-enhanced society: the crucial development of shared embodied cognition

The conception of autonomous, intelligent, collaborative robots has been the subject of science fiction rather than science in the second half of the previous century, with practical applications limited to industrial machines without any level of autonomous, intelligent, and collaborative capacity. The new century is facing the challenge of pressing industrial and social revolutions (4, 5, 6, …) with the prospect of infiltrating robots in every sector of human society; however, this dissemination will be possible if and only if acceptable degrees of autonomy, intelligence, and collaborative capacity can be achieved. Scientific and technological innovations are needed within a highly multidisciplinary framework, with a critical integration strategy and functional characterization that must ask a fundamental question: the design of autonomous, intelligent, collaborative robots should aim at a unified single template to be mass-produced including a standard setup procedure for the functional adaptation of any single prototype, or should the design aim at "baby" robots with a minimal set of sensory-motor-cognitive capabilities as the starting point of a training and educational process in close connection with human companions (masters, partners, final users)? The former alternative is supported by EAI, i.e., the Embodied variant of the Artificial Intelligence family of computational tools based on large foundation models. The latter alternative is bio-inspired; namely, it attempts to replicate the computational structure of Embodied Cognitive Science. Both formulations imply embodiment as a core issue. Still, we think this concept has a markedly different meaning and practical implications in the two cases, although we are still far away from the practical implementations of either roadmap. In this opinion paper, we explain why we think the bio-inspired approach is better than the EAI approach in providing a feasible roadmap for developing autonomous, intelligent, collaborative robots. In particular, we focus on the importance of collaborative human-robot interactions conceived in a general sense, ranging from haptic interactions in joint physical efforts (e.g., loading/unloading) to cognitive interactions for joint strategic planning of complex tasks. We envision this type of collaboration only made possible by a deep human-robot mutual understanding based on a structural equivalence of their embodied cognitive architecture, based on an active, first-person acquisition of experience rather than a passive download of third-person knowledge.