Learning-enabled collaborative intelligence in search and rescue robotics: a survey of paradigms, models, and challenges

Search and Rescue (SAR) robotics increasingly depends on collaborative intelligence to operate effectively in dynamic, uncertain, and time-sensitive environments. Modern SAR missions require coordinated interaction among heterogeneous robotic agents as well as structured collaboration between humans and intelligent systems for perception, decision-making, and adaptive mission execution. This paper presents a structured narrative review supported by a systematic literature search of learning-enabled collaborative intelligence frameworks in SAR robotics, organized around three paradigms: human-to-machine (H2M), machine-to-machine (M2M), and machine-to-human (M2H) learning. The review examines recent developments in reinforcement learning, federated and decentralized learning, knowledge distillation, swarm intelligence, semantic communication, and the growing integration of large language models (LLMs) and vision-language models (VLMs) for knowledge grounding, transfer, and explainability. Beyond algorithmic advances, the analysis emphasizes deployment-oriented challenges including partial observability, communication constraints, uncertainty management, model reliability, and real-time operational requirements. Rather than conducting a fully PRISMA-compliant systematic review, this work synthesizes representative and influential studies to identify emerging trends, architectural trade-offs, deployment limitations, and future research directions for scalable and trustworthy collaborative SAR systems.