Multi-disciplinary Enhancements for Autonomous Service Robots: A Case Study on PiBot's Navigation, Perception, and Human Interaction

Abstract

Autonomous Mobile Robots (AMRs) are rising in modern-day applications ranging from material transportation to human-robot interactions, exhibiting the potential to contribute in both indoor and outdoor environments [30]. Tecnologico de Monterrey’s PiBot, a house-built AMR, presents the typical challenges faced by such robots, mainly its dependence on a single 2D LiDAR for Simultaneous Localization and Mapping (SLAM) and obstacle detection, which limits its perception capabilities, especially in dynamically changing environments with human obstacles. Additionally, its interaction capability is restricted to a touchscreen or terminal input, posing a barrier to intuitive human-machine communication. This thesis presents a series of multi-disciplinary enhancements on PiBot, focusing on augmenting its autonomous navigation performance, perception, and interactive capabilities while investigating understanding its power consumption behavior. A known, better-performing local planner algorithm is proposed, with more outstanding results in social environments where dynamic obstacles such as humans are present. Integrating a RealSense D435i depth camera to the obstacle layer of the costmap transcends the single-height obstacle detection limitation from the 2D LiDAR, thereby enhancing PiBot’s perception. Concurrently, an exploration into Natural Language Processing (NLP) and Convolutional Neural Networks (CNN) is displayed, implementing a Spanish speech recognition system using Wav2Vec2 models and evaluating MaskNET’s performance in dynamic video capturing scenarios. Moreover, a meticulous analysis of PiBot’s power consumption under various operational scenarios provides crucial insights into its autonomy duration. The expected results are enhanced PiBot’s indoor navigation, obstacle detection, and human interaction through speech commands, alongside a deeper comprehension of its power autonomy and perception augmentation. The anticipated contributions from this research aim to push the reliability and efficiency of AMR systems, addressing the integrative challenges of multi-sensor navigation, obstacle detection, and interactive capabilities, thereby paving a path towards more adept and intuitive autonomous service robots