Embodied intelligence for robot manipulation: development and challenges

Abstract Embodied intelligence is regarded as a key pathway to achieving artificial general intelligence (AGI) due to its ability to enable direct interaction between digital information and the physical environment. With the development of artificial intelligence (AI) technology, embodied intelligence has demonstrated strong capabilities in perception, reasoning, and generalization within real-world environments, driving the transformation of AI systems from data-driven to embodiment-based and reshaping the paradigm of robotic manipulation. Currently, research on embodied manipulation is advancing rapidly, and there is an urgent need for a systematic review focusing on embodied manipulation: to construct the core concepts of embodied manipulation, summarize previous literature, break through the limitations of unstructured organization, propose solutions to existing dilemmas, and provide systematic support for the practical implementation of intelligence. This paper focuses on the field of robot manipulation driven by embodied intelligence and conducts a systematic review: it analyzes the conceptual connotation of embodied manipulation and clarifies its definition as taking embodied cognition as the engine and physical robots as the carrier, sorts out its core characteristics and challenges, and reviews the supporting research outcomes in this field; the core part centers on the dual paradigms of data-driven and model-driven approaches, conducts an in-depth analysis of the diverse research status in the field of embodied manipulation, and clarifies the core value of data and models in this context; finally, it specifically analyzes the current research dilemmas and proposes future research directions based on solutions to these dilemmas. This paper aims to clarify the core link of embodied manipulation, provide systematic insights for researchers in the field, and support the development of the embodied intelligence community.