stable-baselines with JAX & Haiku

Berkeley CS 294: Deep Reinforcement Learning

Lunar Lander game from OpenAI Gym using behavioral cloning, DAgger methods, and POMDP(Partially-Observable Markov Decision Processes)

Using DAgger with our MPC treated as the expert, we are able to effectively distill knowledge into relatively simple networks while still being able to retain a large fraction of the performance. (Please see paper for full description).