摘要
Abstract
With the rapid advancement of manned spaceflight,deep space exploration,and on-orbit servicing missions,the demands for high autonomy,strong robustness,and adaptability to com-plex environments in space intelligent robots are becoming increasingly prominent.This paper systematically reviews the key technologies of Vision-Language-Action(VLA)models and sum-marizes major research progress both domestically and internationally.Representative studies are analyzed from two dimensions:task planning and end-to-end control strategies.In conjunction with space robotic operation scenarios,the application potentials of these models is examined in depth in terms of environment perception,semantic understanding,task planning,and manipula-tion execution for space robots,with particular emphasis placed on the application requirements of multimodal large models in space robotics.On this basis,taking into account the current de-velopment status of space robotic technologies in China,this paper proposes forward-looking de-velopment strategies for multimodal large models for future space intelligent robots from multiple prospects,including software and hardware design,model application capabilities,and intelligent ecosystem development.These strategies provide references for intelligent application of space ro-bots in complex operational tasks in manned spaceflight,deep space exploration,and on-orbit servicing.关键词
空间机器人/具身智能/视觉-语言-动作模型Key words
space robots/embodied intelligence/Vision-Language-Action model分类
航空航天
