Investigating Reinforcement Learning Agents for Continuous State Space Environments

Abstract--Given an environment with continuous state spaces and discrete actions, we investigate using a Double Deep Q-learning Reinforcement Agent to find optimal policies using the LunarLander-v2 OpenAI gym environment. I. INTRODUCTION For this study, we examine performance of reinforcement learning (RL) algorithms for continuous state space MDPs, specifically OpenAI Gym's LunarLander-v2. In this environment, the goal is for the RL agent to learn to land successfully on a landing pad located a coordinate points (0,0) in the frame. The agent receives -0.03 points for firing its main engine for each frame, and landing on the landing pad is 100-140 points, which can be lost if the agent moves away from the pad. Each leg contact with the ground is 10 points.