Enhancing the ability of robots to adapt seamlessly to new tasks is essential for advancing intuitive human-robot interaction, particularly for non-expert users. This study introduces a novel approach that integrates scene understanding via scene graphs with large language models (LLMs) to enable flexible task execution guided by natural language commands. By leveraging 6D pose estimation and scene graph generation, our method constructs a detailed, spatially-aware representation of the environment, empowering the system to dynamically interpret and respond to user instructions. The synergy between this contextual understanding and an LLM enables robots to comprehend and execute a wide range of tasks with minimal reprogramming efforts. Preliminary experiments conducted in a simulated setting demonstrate the system’s capability to process diverse tasks by translating general instructions into task-specific actions, substantially reducing the need for user input and setup. This approach establishes a foundation for developing adaptable robotic systems that can intuitively and efficiently understand and perform new tasks.