DQN and PPO implemented from scratch in PyTorch to solve LunarLander-v3; PPO matches Stable-Baselines3 in multi-seed evaluation and ablations on wind robustness and two-command reproduction
Read the original at github.com→★ 2 stars
Original headline: "andresbxa/lunar-lander-rl: DQN and PPO implemented from scratch in PyTorch to solve LunarLander-v3. PPO matches Stable-Baselines3 (260 vs 259). Multi-seed evaluation on held-out seeds, ablations, robustness to wind and two-command reproduction."