Room Navigating Agent in a Virtual Reality Environment

We focus on the application of deep reinforcement learning to 3D navigation in the simulated environment. The agent is placed inside the 3D scenes of houses from the SUNCG dataset and directed to go to a target room. To succeed, the agent learns the scenes from RGBD information with two deep reinforcement learning algorithms, Deep Q Network (DQN) and Deep Deterministic Policy Gradients (DDPG). Using these two models, we make the agent navigate to the target room using the visual inputs in the virtual reality (VR) environment. The result shows that DDPG performs and generalizes better in the VR environment.