An Automated Statistical Evaluation Framework of Rapidly-Exploring Random Tree Frontier Detector for Indoor Space Exploration

This paper focuses on the design of an automated statistical evaluation framework for mapping generation of Rapidly-Exploring Random Tree (RRT) frontier detectors. By evaluating the run time and distance traveled of the simulated Kobuki robot agent in a Gazebo environment, the designed framework can automatically evaluate the process on a user-defined Gazebo map for a large number of repeated simulations. We also expanded the experiment platform into customized maps with complex layouts and trial schemes. The key formulas and parameters are provided with different trial settings. During the development of this framework, we have added functions that allow the user to choose among the maps we have designed, and the initial positions of the simulated robots for each map at the beginning of each trial. We have also modified the modules developed by Umari et al. so that the RRT frontier detection process can be started automatically with pre-defined exploration area in place. Modules have also been added so that the run time and distance traveled by the simulated robot for each trial can be measured and saved to the respective CSV files for further statistical analysis. We have created additional procedures that ensure the consistency of each trial. The results show that our designed automated evaluation framework is reliable and suitable for use as a fully automated research platform for robot exploration.