Diseño e implementación de un modelo de navegación visual lingüístico para el robot móvil Walter

Abstract

CONDITION FOR PUBLICATION OF PROJECT. The scarcity of employees related to the service area over the last decade, adding limited access to local service robotic development, has been the motivation of this project. A semi-autonomous navigation system on the mobile robot “Walter”, is shown based on Vision-Languaje-Navigation models (VLN). The main objective is to construe instructions from an average person by voice to execute navigation to the asked object. Artificial intelligence models like TinyLlama, were used to keep a human speech between Walter and a person. YOLOv8 model was used to extract the information needed to the system to perform the best decisions in real time. The results showed a success rate of more than 85% in searching and approximation scenarios, with 90% of human acceptance. All of this in a supervised scenario. To summarize, the integration of light agentic AI in a limited hardware allows robotic assistants to carry out functional and economics robotic assistants, for instance Walter, letting the user be free of controlling the system all the time, and performing a better human-robot interaction smoother. Keywords: Artificial Intelligence, object detection, human-robot interaction, accessible hardware, semi-autonomous.