Embodied intelligence is a new paradigm in artificial intelligence.With the development of large models and reinforcement learning,embodied intelligence is moving from theory to reality.As the natural carrier of artificial intelligence,unmanned systems have also made considerable progress,but current unmanned systems still lack the ability to perform autonomous behaviors in dynamic and complex environments.By introducing the concept of embodied intelligence,a new generation of autonomous unmanned systems will be formed.This paper first introduces the basic concepts of embodied intelligence and autonomous unmanned systems,pointing out that autonomous unmanned systems need to autonomously learn through interaction with the environment to achieve autonomous perception,decision-making,and action.Secondly,it summarizes the two approaches to achieving embodied intelligence in autonomous unmanned systems,including large models and reinforcement learning technology,and analyzes their applications in embodied perception cognition,embodied planning and decision-making,and embodied action control.Finally,it looks forward to the key issues that need to be addressed in the next stage of embodied intelligence research,including lightweighting,group coordination,full autonomy,and human-machine integration,with the goal of achieving a fully autonomous unmanned system based on embodied intelligence.
关键词
具身智能/大模型/强化学习/大语言模型/视觉大模型/视觉语言模型/自主无人系统
Key words
Embodied Intelligence/Large Models/Reinforcement Learning/Large Language Model/Large Vision Model/Vision-language Model/Autonomous Unmanned System
维普
万方数据