Embodied Intelligence: Bridging Neuroscience, AI and Soft Robotics for Adaptive Physical AI

The study of natural and artificial intelligence in physical agents has a great interest both from a scientific and technological point of view. On the one hand a better understanding of how intelligent behaviors evolved in nature, in particular in plants, animals and humans have the highest scientific interest, on the other hand the reverse engineering of the emergence of those behaviors in nature could lead to the development of deeply bio inspired artificial intelligent agents (a.k.a ‘intelligent robots’) much more reliable, adaptive and capable of those built so far. Recent advances in robotics and artificial intelligence have highlighted the critical role of embodiment and of the compliance of the physical bodied of agents (hence the success of ‘soft robotics’ ) in enabling more adaptive, flexible, and robust intelligent behavior. In contrast to traditional disembodied computational models, natural biological intelligence emerges through continuous interaction between brain, body, and environment. In the meantime the fast progress in neuroscience and system biology raises some new intriguing questions.Despite rapid progress in machine learning and autonomous systems, many artificial agents still struggle with generalization, adaptability, and robustness outside controlled training environments. In particular, the utilization of Generative AI, Diffusion Models, LLMs (Large Language Models) and VLAs (Visual Language Action models) although rather successful in some applications (in particular Human Robot Interaction) finds serious challenges in the comparatively scarce availability of data and the real time needs of intelligent agents physically interacting in the real world. The possibility that the current approaches to AI could lead to human (or super human) intelligence (a.k.a. AGI, Artificial General Intelligence) by ‘scaling’ the current very successful approaches of Deep Learning architectures seems to be questioned by the difficulties of its application in robotics. A major challenge lies in bridging theoretical insights from neuroscience, in humans and animals, and cognitive science with practical implementations in robotics and AI, and a more deep and nuanced approach to AGI.This article collection (of inherent interdisciplinary characteristics) aims to explore the interface between neuroscience neurorobotics and deep learning from an embodied AI and soft robotics perspective. We believe that, by following the ‘synthetic methodology’ approach neuroscience brain studies and AI and robotics can help each other.Areas of interest include, but are not limited to:- Models of embodied intelligence- AGI and ASI in physical agents: challenges and opportunities- LLM, VLA, Generative AI and Diffusion Models in soft robotics- Models of Morphological computation- Neural mechanisms of sensorimotor integration and embodied cognition- Active inference, predictive processing, and dynamical systems approaches- Brain-inspired architectures for adaptive robotics- Emergence of intelligent behaviours in artificial agents- Systemic models in neuroscience- Animal and Human neuroscience- Evolution of intelligent behaviors in animal and humans- Approaches to intelligent robotics inspired by cybernetics (e.g. Braitenberg vehicles)- Simplexity (in A.Berthoz sense)in human animal neurophysiology and machines- Stigmergy in human animal neurophysiology and machines- Emergence of sensorimotor integration in natural and artificial agents- Embodiment aspects in the motor control in humans and animals Topic editors: Fabio Bonsignorio, Florian Röhrbein Source: https://www.frontiersin.org/research-topics/78193/embodied-intelligence-bridging-neuroscience-ai-and-soft-robotics-for-adaptive-physical-ai