A Review of Morphology-Based Approaches to Intelligent Control of Embodiment
In recent years,embodied intelligence based on reinforcement learning has been widely applied in the field of robot control.Embodied intelligence focuses on the perception and interaction of an intelligent body on the environment,which is the control and optimization of the physical form of the intelligent body at the physical level.Therefore,the problems studied in embod-ied intelligence are generated by the interaction of morphology,learning and behavior.Among them,the introduction of morpho-logical information can effectively constrain the huge search space brought by the complex morphology of robots.In light of this,this research focuses on the morphological control part of embodied intelligence and centers on the morphological control method based on reinforcement learning.It introduces the current research progress and related problems of morphological control for em-bodied intelligence,focusing on morphological information introduction and model migration,summarizes the current hot meth-ods,graph neural network method and transformer method respectively,and gives an outlook on the future development direction.
万方数据
维普
