The integration of Artificial Intelligence (AI) into robotic arms, particularly through the use of Large Language Models (LLMs), marks a significant advancement in automation, enabling these devices to perform complex tasks with unparalleled precision. Despite the potential of LLMs to revolutionize robotic functionality by understanding and processing human language, their application faces challenges, notably in generating precise, actionable commands for robotic operations. To address these issues, we introduce the “Tree of Thoughts” (ToT) methodology, enhanced by a Depth-First Search (DFS) strategy, designed to optimize LLM performance in robotic control. The ToT approach bridges the gap between LLMs’ abstract language comprehension and the generation of specific instructions, streamlining the command generation process and enabling robotic arms to execute tasks with increased precision and adaptability. This methodology not only demonstrates the feasibility of adapting LLMs as feedback controllers for complex tasks but also significantly improves the performance and adaptability of robotic arms, paving the way for more intelligent, responsive robotic systems.