The introduction of Large Language Models (LLMs) in the past few years has disrupted the fields of natural language understanding and embodied artificial intelligence. LLMs have previously been used to generate plans from human instructions, enabling a physical agent (robot) to complete abstract long-form tasks without direct programming, when provided with a list of capabilities. On the other hand, Retrieval Augmented Generation (RAG) extends the capabilities of LLMs by querying an easily-updated knowledge base ‘on the fly’, outside of its own parameters, to generate a response. This work is one of the pioneers in exploring the usability of a RAG-based approach to generate robot plans, with RAG used to inform about the capabilities of the robot, and the context in which it is operating, instead of being stored them in the LLM itself. The usability of this approach was evaluated in the context of general purpose service robots (GPSR). Experimental results revealed that the proposed approach outperforms by 15% state-of-the-art LLM-based techniques for robot planning in terms of generating valid plans from ‘implicit’ queries where a GPSR is expected to understand and perform an action based on environmental context, without being explicitly instructed on every detail. Our approach runs locally on with low computational requirements enabling real-time robot operations.