Embodied AI with Large Language Models: A Survey and New HRI Framework
Ming‐Yi Lin, Ou-Wen Lee, Chi-Jie Lu
Abstract
The study aims to develop an emotional logic engine based on a large language model (LLM), providing emotional connections, personalized interactions, knowledge representation, and logical inference. Using this emotional logic engine, we intend to realize the goals of high-level cognition, autonomous knowledge reasoning, long-horizon planning, and action execution described in embodied artificial intelligence (embodied AI). Ultimately, we will implement an efficient intelligent companion interaction robot (ICIR) based on a novel human-robot interaction (HRI) framework to enhance the interaction between humans and robots. The proposed framework integrates multiple components including a visual language model (VLM), logic reasoning model, pre-trained database integration, and the development of a multi-modal template. Additionally, we introduce a complementary framework termed perception-action loop (PALoop), which is meticulously modeled and constructed to facilitate seamless interactions between human operators and robotic systems. Detailed design aspects of both frameworks are elucidated, providing insights into their architecture and functionality. The research outcomes will be practically applied, offering the robotics industry innovative and practical technology solutions.