RL Moon Landing & Reward Shaping
A Deep Reinforcement Learning study analyzing the impact of custom reward shaping and architecture selection in OpenAI Gymnasium's LunarLander-v3.
Investigated the effects of reward function design and algorithm selection on Deep Reinforcement Learning (DRL) agent stability and convergence. Evaluated multiple DRL architectures within OpenAI Gymnasium's LunarLander-v3 environment by benchmarking default reward conditions against custom-engineered reward shaping logic.
Utilized Stable-Baselines3 to train and analyze agents across continuous and discrete control parameters. Evaluated performance trajectories, terminal state success rates, and sample efficiency to highlight the critical role tailored reward engineering plays in complex physical simulation environments.