Multimodal perception-driven decision-making for human-robot interaction: a survey

Abstract

Multimodal perception is essential for enabling robots to understand and interact with complex environments and human users by integrating diverse sensory data, such as vision, language, and tactile information. This capability plays a crucial role in decision-making in dynamic, complex environments. This survey provides a comprehensive review of advancements in multimodal perception and its integration with decision-making in robotics from year 2004–2024. We systematically summarize existing multimodal perception-driven decision-making (MPDDM) frameworks, highlighting their advantages in dynamic environments and the methodologies employed in human-robot interaction (HRI). Beyond reviewing these frameworks, we analyze key challenges in multimodal perception and decision-making, focusing on technical integration and sensor noise, adaptation, domain generalization, and safety and robustness. Finally, we outline future research directions, emphasizing the need for adaptive multimodal fusion techniques, more efficient learning paradigms, and human-trusted decision-making frameworks to advance the HRI field.