Abstract
Building autonomous and intelligent robots has been an elusive dream for researchers for some time. Simultaneous Localization and Mapping (SLAM) systems have contributed towards achieving this goal by making robots better in navigating through complex environments. Until now it has only been possible to train and teach robots to move around in particular environments using a certain set of rules and heuristics. With the sudden surge in interest in AI and Machine Learning, a lot of effort has been put in into making robots intelligent and for them to automatically learn their paths in unknown environments (also referred to as Path Planning). This however has been met with mixed results as either the solution proposed is not too practical (e.g. requires too much training) or has limited success (e.g. works in specific environments). In this research, we develop a novel autonomous path planning framework using Deep Learning which can learn to navigate in unknown environments. The system has been tested on state-of-the-art Active Vision Dataset with promising results.