We are pleased to share our latest work, “Foundation Models in Robotics: A Comprehensive Review of Methods, Models, Datasets, Challenges and Future Research Directions,” authored by Aggelos Psiris, Vasilis A., Evangelos Markakis, Panagiotis Sarigiannidis, Efstratios Gavves, Kostas Bekris, Arash Ajoudani, and Georgios Th. Papadopoulos.
📖 Read the paper: https://arxiv.org/pdf/2604.15395
In recent years, robotics has undergone a major transformation—from fixed, single-task, domain-specific systems to adaptive, multifunctional, general-purpose agents capable of operating in complex and dynamic real-world environments. This shift has been largely driven by the emergence of Foundation Models (FMs).
This study provides a comprehensive and systematic review of the research landscape of Foundation Models in robotics, with a focus on:
- The evolution of robotic Foundation Models, highlighting key development phases:
- 2018–2021: Integration of Natural Language Processing (NLP) and Computer Vision (CV) models
- 2021–2022: Grounded planning with Vision–Language (VL) representations
- 2022–2023: Embodied Vision–Language–Action (VLA) policies
- 2023–2024: Memory, autonomous task composition, and web-to-robot transfer
- 2024–present: Multi-sensory generalization and real-world deployment
-
A detailed taxonomic analysis of existing approaches, based on multiple criteria, including:
- Type of Foundation Model and input/output modalities
- Neural network architectures
- Learning paradigms and stages
- Task types and application domains
For each category, the paper provides a structured comparative analysis along with key insights.
- An overview of public datasets and benchmarks used for training and evaluation
- A discussion of current challenges and future research directions in this rapidly evolving field
This work aims to serve as a valuable reference for researchers and practitioners interested in the future of robotics and intelligent systems.
