Answers are grounded in this paper and cite their sources.
Embodied Artificial Intelligence (EAI) is increasingly seen as a promising route toward flexible automation, also in industrial operations. Yet the term is used inconsistently, and existing surveys often address isolated components rather than the combined role of embodiment, learning, perception, action, data, and industrial deployment constraints. This paper presents a PRISMA-guided systematic literature review of recent EAI research in industrial operations. We first consolidate the terminology of EAI and define four guiding characteristics: embodiment, learning, perception-action coupling, and situated intelligence. Based on this framework, we introduce a manipulation-mobility matrix for classifying physical embodiments and analyze the reviewed literature across industrial applications, research focuses, hardware forms, input and output modalities, learning architectures, datasets, simulation environments, and training paradigms. The results show that current research is dominated by assembly and manufacturing scenarios, stationary robotic arms, vision- and language-based interfaces, and modular foundation model-centered architectures. However, tightly coupled sensorimotor learning, tactile and force feedback, situated intelligence, and validated real-world deployment remain limited. This review clarifies the current state of industrial EAI and identifies the key gaps that must be closed to enable scalable deployment in industrial operations.
No takes yet. Share an insight, caveat, or question.
Koch et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: