AI based Robot Navigation Framework for Industrial and Domestic Applications

A service robot operates autonomously in service areas such as residences, hospitals, restaurants, museums, and nursing homes. It finds the quickest route from a starting point to a final destination. This effort aims to create a universal navigation framework for an interior mobility service robot. Learning from Demonstration (LfD) is task-oriented development technique that teaches a machine or robot a unique behaviour by presenting the activities directly rather than programming it with control codes and instructions. Unlike conventional programming methodologies, these solutions doesn’t necessitate designing or coding knowledge and can quickly translate shown behaviour into executable code. Three LfD approaches—counter-based, encoder-based, and enhanced encoder-based—are proposed for teaching a mobile service robot to navigate from one site to another. The robot is programmed to navigate to all feasible locations sequentially. This paper discusses the Tree-Based Planner (TBP), which creates a path tree with all logical links between all sites in the surrounding. The system employs the Deep Q-Learning (DQL) method to learn the initial paths from a topological map of the environment, i.e. the path tree generated by the TBP system. The outcome is a navigation framework that takes into account the typical characteristics of an autonomous mobility robot deployed in a service context.