Habitat is a relatively new entry by Facebook AI Research for a new form of embodied agents. This platform represents the ability to represent full 3D worlds displayed from real-world complex scenes. The environment is intended for AI research of robots and robotic-like applications that DRL will likely power in the coming years. To be fair though, pun intended, this environment is implemented to train all forms of AI on this type of environment. The current Habitat repository only features some simple examples and implementation of PPO.
The Habitat platform comes in two pieces: the Habitat Sim and Habitat API. The simulation environment is a full 3D powered world that can render at thousands of frames per second, which is powered by photogrammetry RGBD data. RGBD is essentially RGB color data plus depth. Therefore, any image...
