Title:Integration of Large Language Models within Cognitive Architectures for Autonomous Robots

Abstract:The usage of Large Language Models (LLMs) has increased recently, not only due to the significant improvements in their accuracy but also because of the use of the quantization that allows running these models without intense hardware requirements. As a result, the LLMs have proliferated. It implies the creation of a great variety of LLMs with different capabilities. This way, this paper proposes the integration of LLMs in cognitive architectures for autonomous robots. Specifically, we present the design, development and deployment of the llama\_ros tool that allows the easy use and integration of LLMs in ROS 2-based environments, afterward integrated with the state-of-the-art cognitive architecture MERLIN2 for updating a PDDL-based planner system. This proposal is evaluated quantitatively and qualitatively, measuring the impact of incorporating the LLMs in the cognitive architecture.