Optimization of Industrial Robot Navigation Algorithm Based on Pre-trained Large Model and Knowledge Distillation

In recent years, with the rapid development of intelligent manufacturing, industrial robots have become a crucial component in industrial settings, playing a significant role on production floors. Concurrently, advancements in artificial intelligence technology have provided strong support for the development of industrial robots. Although traditional AI techniques can efficiently perform tasks when deployed to production sites, the complex and variable industrial environment can lead to recognition failures and erroneous outputs due to untrained anomalous data. The emergence of large model technology enables industrial robots to handle complex industrial scenarios more flexibly, thus enhancing navigation accuracy and efficiency. This paper proposes a robot navigation algorithm based on large models, employing pre-trained large language models, vision-language models, and navigation models. To reduce the number of model parameters, we applied knowledge distillation to the vision-language model. To validate the effectiveness of the proposed method, we conducted experiments in a simulated environment using an automated guided vehicle (AGV) robot and compared the results with traditional methods. The experimental results demonstrate that our proposed method achieves higher navigation accuracy and efficiency in complex industrial scenarios, verifying its superiority and practicality.