Robotic Path Planning Based on Episodic Memory Fusion

Abstract

Episodic memory provides a mechanism for recalling past experience, which can be used for path planning in complex environments. This paper describes a path planning method based on memory fusion that combines an episodic memory models with the potential path detection network. In traditional path planning methods based on the episodic memory model, paths were planned based on the trajectory that the mobile robot has experienced in the environment, ignoring the surrounding potential paths. Therefore, the planned path is not necessarily globally optimal. In response to this problem, we proposed a path detection network to find potential safe paths in the environment. Our experimental results demonstrated that a better path can be found by fusing the potential path into the original episodic-cognitive map from the perspective of planned path length and number of turns.