Toward Robotic Metacognition: Redefining Self-Awareness in an Era of Vision-Language Models

Kurzfassung

Artificial Intelligence (AI) aims to replicate and even enhance human cognitive capabilities. However, AI researchers have primarily focused on planning and learning, leaving other concepts underexplored. One important concept is metacognition, which enables agents to understand and improve their own processes [1]. Metacognition is key to human self-improvement [2] and even the recovery of physiological limitations originating from stroke [3] or brain injury [4]. Recent advancements have challenged the traditional boundaries of achieving such human-level cognition through conventional methods. In particular, Large Language Models (LLMs) and Vision-Language Models (VLMs) have the potential to boost the development of cognitive architectures significantly. This blue-sky paper explores how VLMs can be leveraged to handle and avoid hardware failure and thus increase the resilience of future robots through metacognitive reasoning. With this we aim to showcases the transformative potential of metacognition as future research direction in robotics.