Autonomous Navigation and Room Categorization for an Assistant Robot

Currently, the majority of elderly people live alone and they require daily support for their routines such as nutrition, mobility assistance, and safety. A personal robot assistant could help with these tasks by carrying objects and keeping track of routines. The aim of this paper is to enhance abilities of a mobile robot to assist elderly people, which can explore and map home environments, learn and categorize each room in the map and navigate in accordance with the static and dynamic surroundings. To autonomously explore the new environments; maps are constructed using GMapping and trajectories are planned by Dijkstra global path planner, and Dynamic Window Approach (DWA) local path planner. Rooms in the generated maps are isolated from each other by Template Matching, then classified by Convolutional Neural Network (CNN). Once the environments are explored, the robot is localized in the 2D space utilizing Adaptive Monte Carlo Localization (AMCL), for point-to-point navigation tasks. The performance of the approach was evaluated through various experiments. Maps of the environments were successfully built by GMapping and CNN algorithm was able to classify almost all the rooms, eleven out of twelve, correctly. Likewise, AMCL, Dijkstra, and DWA algorithms were able to navigate the robot for point-to-point tasks.