Development of a Mobile Service Robot System: Enhancing Localization, Guidance, and Search Tasks in Indoor Environments
Name: ELIO DAVID TRIANA RODRIGUEZ
Publication date: 03/07/2025
Advisor:
Name |
Role |
|---|---|
| ANSELMO FRIZERA NETO | Advisor |
Examining board:
| Name |
Role |
|---|---|
| ANSELMO FRIZERA NETO | Presidente |
| CARLOS ANDRÉS CIFUENTES GARCÍA | Examinador Externo |
| MARIO FERNANDO JIMENEZ HERNANDEZ | Coorientador |
| THIAGO OLIVEIRA DOS SANTOS | Examinador Interno |
Summary: The current development of robotics is aimed at enhancing collaboration between service and assistive robots and humans in daily, occupational, and healthcare-related activities. These systems seek to provide support that increases users' functional independence. However, such collaboration requires advanced interaction strategies, particularly in natural language processing for interpreting human instructions, as well as the ability to understand and adapt to dynamic and complex environments.
In this context, the present project repurposes an existing mobile robot by upgrading its sensors, low-level control, and software based on the Robot Operating System (ROS), with the objective of evaluating the effectiveness of recent machine learning algorithms and generative artificial intelligence in guidance tasks and user interaction. The proposal is grounded in two prior studies. The first assesses an initial version of the robot and a chatbot-style interface deployed in the cloud. It also evaluates perceived usability through the System Usability Scale (SUS), administered to volunteer participants. The results indicate a favorable perception of usability, along with an accuracy of approximately 91% in interpreting user requests. The second study explores the use of open-source large language models, executed locally, to generate dynamic descriptions of the environment without explicitly storing them in the robot’s map. Real-world tests achieved 93% accuracy. Furthermore, the study investigates two popular concepts in the development of generative AI applications: chain-of-thought workflows and Retrieval-Augmented Generation (RAG) strategies for building customized information bases. Additionally, the methodological development enabled the creation of a model and application that can be easily replicated across different robotic systems. The study also opens avenues for new strategies in designing systems capable of understanding context and leveraging sensory modalities to support more effective communication between users and robotic systems. The development of this project paves the way for future work involving the creation of context-aware relationships using not only text and voice models but also those capable of processing images and video. Moreover, it encourages the development of applications that enable task execution through multi-agent systems.
