This is my Reinforcement Learning Coursework

Reinforcement learning (RL) is one of the hottest research topics in the area of Artificial Intelligence, which has achieved great performance in various problems, including robotic manipulation, Go(AlphaGo), playing games, and autonomous driving.

In this report, we are going to introduce two popular deep reinforcement learning algorithms: Deep Q-Network (DQN) and Policy Gradient (PG). DQN is a value-based approach that first tries to compute optimal state-action value, then extract optimal policy from it, While the latter is a policy-based approach, which can directly optimize the policy. All these two algorithms can learn successful policies from high-dimensional inputs, such as pixels of images, by using end to end reinforcement learning.

We implement a deep Q-network based on Atari game Pong, and an autonomous driving agent on the simulation engine VISTA, which uses techniques in computer vision to synthesize new photorealistic trajectories and driving viewpoints.