ROS Navigation Stack for Smart Indoor Agents

Abstract

Advances in compute power, sensor technology, and machine learning have facilitated a plethora of assistive and personal agents. These agents are poised to make our life more efficient, safer, feature rich, and more enjoyable. With so much activity in this area, there has been a lot of progress developing algorithms for localization, path planning, path guiding, and obstacle avoidance. Similarly, numerous frameworks for human computer interaction, obstacle recognition, object tracking, and advanced reasoning have been introduced. This research introduces a navigation stack written in Python using the Robot Operating System for modular indoor agent development. The localization system makes use of deep learning and particle filters and is easily trained to localize in new environments. The obstacle avoidance system can be changed to reflect the agents size, required safety margin, sensor properties and behavior. Different path planning algorithms can be substituted and used in the path guiding system. The created navigation stack was tested on an assistive technology wheelchair, exhibiting state of the art localization, collision avoidance, and navigation in complex scenarios.