Lunar Lander with Deep Q-Learning in Pytorch

This repository displays a solution to the lunar lander problem from the gymnasium library. Solved using Deep Q-learning (DQL) in pytorch. Notes on DQL shown here.

Results

---

Training was stopped when the average rewards over the latest 100 episodes reached +200, which took 1764 episodes.

After 100 episodes:

lunar_lander_100.mp4

After training completed:

lunar_lander.mp4

---

Resources

• Richard Sutton's book
• Andrew Ng's Reinforcement Learning course in the coursera Machine Learning specialization