Self-evolving Embodied AI

Abstract

Abstract Embodied Artificial Intelligence (AI) is an intelligent system formed by agents and their environment through active perception, embodied cognition, and action interaction. Existing embodied AI remains confined to human-crafted setting, in which agents are trained on given memories and construct models for given tasks, enabling fixed physical embodiments to interact with relatively static environments. Such methods fail in in-the-wild setting characterized by variable physical embodiments and dynamic open environments. This paper introduces self-evolving embodied AI, a new paradigm in which agents operate based on their changing physical embodiment and environment through five software-level components, namely memory self-updating, task self-switching, environment self-prediction, embodiment self-adaptation, and model self-evolution, together with a hardware-level component for physical embodiment evolution, aiming to achieve continually adaptive intelligence through software-hardware co-evolution. Specifically, we present the definition, framework, components, mechanisms, roadmap, and evaluation indicators of self-evolving embodied AI, systematically review state-of-the-art works for realized components, discuss practical applications, and point out future research directions. We believe that self-evolving embodied AI enables agents to autonomously learn and interact with environments in a human-like manner and provides a new perspective toward general artificial intelligence.