A Robot Memory Management System for Long-term Interactions with Human

Abstract

This paper describes the implementation of an episodic memory system based on the theory of long-term memory. The system's memory model observes and retains the contents of the user's everyday events. The model relies on a memory support pipeline that includes working memory and long-term memory, and can effectively abstract, encode, and decay past memories through our system. It also produces a generalized semantic memory through classification and retrieval of episodic memory. To verify the system's functionality, we deployed it on a commercial robot platform and tested its ability to manage memory over a long period of time.