Optimizing Object Recognition of NAO Robots Using Large Language Models (LLMs) Compared to the YOLO Method in Webots Simulation

This paper explores improving object recognition for NAO robots through the integration of Large Language Models (LLMs), specifically the 豆包·视觉理解模型 (Doubao Vision Understanding Model), and compares this with the widely used YOLO (You Only Look Once) object detection method in a Webots simulation environment. While YOLO provides a real-time, high-speed object detection solution, the LLM-based approach offers superior capabilities in contextual understanding, reasoning, and a more nuanced interpretation of visual data. This research aims to demonstrate the effectiveness of the 豆包·视觉理解模型 for object recognition, investigating how the LLM’s enhanced understanding of visual content and reasoning capabilities can be leveraged to improve object recognition over traditional YOLO models. The research also examines the performance, accuracy, and computational trade-offs in a simulated environment, shedding light on the strengths and weaknesses of each approach.