This comprehensive survey explores how multimodal fusion techniques and vision-language models are revolutionizing robot vision capabilities across critical applications.

• Systematically analyzes fusion methods for semantic scene understanding, SLAM, and 3D object detection
• Evaluates the effectiveness of LLM-based vision-language models compared to traditional fusion approaches
• Highlights advancements in navigation and robot manipulation through multimodal integration
• Provides a roadmap for engineering teams developing next-generation robotic perception systems

For engineering teams, this research offers valuable insights into optimal fusion architectures for specific robotics tasks, helping bridge the gap between academic research and industrial implementation.

Multimodal Fusion and Vision-Language Models: A Survey for Robot Vision