Multimodal Perception-Driven Decision-Making for HRI: A Survey

Authors: Wenzheng Zhao, Kruthika Gangaraju, Fengpei Yuan (Worcester Polytechnic Institute) Published: Frontiers in Robotics and AI, August 2025 Coverage: 2004–2024

Overview

Reviews how robots integrate diverse sensory data — vision, language, and tactile — to understand environments and interact with humans. Focuses on Multimodal Perception-Driven Decision-Making (MPDDM) frameworks.

Application Domains

  1. Social and Assistive Robotics — emotion recognition, companion robots, healthcare assistance, rehabilitation
  2. Navigation and Mobile Robotics — obstacle avoidance, socially-aware autonomous movement
  3. Industrial Collaborative Robotics — worker safety, object manipulation
  4. General-Purpose Robotics — complex task understanding, adaptive planning

Major Challenges

  • Sensor fusion complexity — noise and integration across modalities
  • Generalization — learned behaviors don’t transfer to unseen environments
  • Safety and robustness — reliable performance in dynamic human spaces
  • Computational demands — real-time multimodal processing
  • Human variability — unpredictable behavior, cultural differences in social interaction norms

Future Directions

  • Adaptive multimodal fusion techniques
  • More efficient learning paradigms
  • Human-trusted decision-making frameworks