Emerging Multimodal Artificial Sensory Fusion Paradigms Toward Advanced Embodied Intelligence.

Abstract

In the endeavor of embodied intelligence to connect the digital world and physical reality, artificial multimodal sensory fusion has emerged as a cornerstone for endowing agents with robust perception and natural interaction capabilities. Specifically, multisensory fusion constructs a biomimetic sensing framework capable of parallel acquisition and processing of multi-source information. Such architectures augment information perception and ensure higher decision-making reliability within complex environments. However, a systematic review of this rapidly evolving field is noticeably absent. This review bridges this gap by providing a comprehensive analysis of various multimodal sensory fusion approaches. We introduce the configurations of multimodal sensory fusion, as well as the operating mechanisms of sensors and artificial synapses. In particular, various biomimetic multimodal fusion paradigms are critically analyzed, such as visual-tactile and visual-olfactory fusion, with detailed discussions of their device structure and fusion mechanisms. These multisensory fusion systems show great potential in various application fields. Finally, the emerging opportunities are discussed and outlined. By highlighting the critical potential of multimodal sensing fusion, this review aims to inspire further research on fusion strategies and practical applications, facilitating the evolution of embodied intelligence toward more human level.