Embodied AI: A Survey on the Evolution from Perceptive to Behavioral Intelligence

摘要

ABSTRACT Creating intelligent beings like humans is a long‐standing goal in AI research, such as intelligent robots in science fiction. Classic AI technologies are disembodied, and insufficient to make robots intelligently behave in the real world. In contrast, embodied artificial intelligence (Embodied AI) enables artificial agents with physical embodiment to achieve intelligent behavior through interactions with environments. However, there are few comprehensive surveys on Embodied AI from the perspective of robot behavior within the AI domain. Thus, we provide a comprehensive survey on Embodied AI. According to the process of robot behavior, we categorize Embodied AI into three modules: embodied perception, embodied decision‐making, and embodied execution. For each module, we review its tasks, methods, and challenges. We hope this survey can provide a structural framework for Embodied AI research. Besides, we also pay attention to large foundation models in Embodied AI.