Disaster Robotics in the Deep Learning Era]{Disaster Robotics in the Deep Learning Era: A Systematic Review of Platforms, Autonomy, Methods, Datasets, Challenges, and Future Research Directions
Disaster robotics enables responders to sense and act at a disaster site from a safe distance, converting responder risk into an equipment one. The field is increasingly shaped by deep learning and, most recently, by foundation models. However, existing surveys adopt a single perspective or platform class, predate the deep learning era or remain largely narrative. Moreover, none grades every reviewed work on a common validation maturity scale. This paper presents a systematic review of disaster robotics in the deep learning era, restricted to disaster-connected works evaluated on disaster data or in real, replicated or simulated emergency settings. The evolution of the field is traced through successive research phases, from teleoperated platforms to foundation-model interfaces and responder-grounded field autonomy. A multi-criteria taxonomy is formed, covering response tasks, robotic platforms, perception, autonomy, learning strategies, foundation model types, human-robot teaming, coordination, communication and deployment, disaster domains, and validation maturity. Each criterion is analyzed comparatively, with critical insights into the strengths and limitations of its categories under field conditions. Additionally, public datasets and benchmarks are reviewed. Current challenges are also paired with future research directions, from data acquisition and field hardening to verifiable autonomy, responder trust and evaluation. The analysis shows that systems deployed at real incidents still rely predominantly on model-based methods. It also shows that recent research leans increasingly on simulation and benchmark evaluation. Finally, foundation models have entered the field mainly as perception front ends and mission-level planners rather than as embodied policies, without yet reaching a real incident.
Authors
- Vasileios Argyriou (ORCID: https://orcid.org/0000-0003-4679-8049)
- Panagiotis G. Sarigiannidis (ORCID: https://orcid.org/0000-0001-6042-0355)
- Iraklis Varlamis (ORCID: https://orcid.org/0000-0002-0876-8167)
- Georgios Th. Papadopoulos (ORCID: https://orcid.org/0000-0003-1686-421X)
- Θωμάς Λάγκας (ORCID: https://orcid.org/0000-0002-0749-9794)
- Aggelos Psiris
- Georgios Angelidis (ORCID: https://orcid.org/0009-0005-1060-8647)
Institutions
- Democritus University of Thrace (GR)
- Kingston University London (GB)
- Harokopio University of Athens (GR)
- University of Western Macedonia (GR)
Publication Details
- Journal
- Research Square
- Published
- 2026-11-19
- DOI
- https://doi.org/10.21203/rs.3.rs-11587551/v1
- Primary Topic
- Robotics and Automated Systems
- Type
- preprint