Machine Learning-Based Agoraphilic Navigation Algorithm

Abstract

This paper presents a novel machine learning-based Agoraphilic (free space attraction) navigation algorithm. The proposed algorithm is capable of undertaking local path planning for mobile robots in unknown dynamic environments with a moving goal. The inability to track and reach a moving goal is one of the common weaknesses of most existing navigation algorithms operating in dynamic environments. High uncertainty involved in dynamic environments is also another major challenge. The novel machine learning-based approach helps the proposed algorithm to successfully overcome these challenges. This paper also introduces the integrated modular-based architecture for free-space attraction-based algorithms. This allows the algorithm to incorporate ten different modules with miscellaneous algorithms to perform sub-tasks such as tracking, prediction, map generation, machine learning-based free space attraction force generation and robot motion command generation. The new modular-based architecture integrates those sub-modules to create the robot’s driving force. This driving force is the single attractive force to pull the robot towards the moving goal via current free space leading to future free space passages. The proposed algorithm was experimentally tested under a dynamic environment. The experiment was focused on testing the behaviour of the algorithm under the challenge of reaching a moving goal. Furthermore, the test results demonstrate that the Agoraphilic algorithm is successful in reaching a moving goal in an unknown dynamically cluttered environment.